Bülent Tekmen

AI Is Bigger Than Five: The Founding Story of the AI Empires

Thu, 02 Jul 2026 00:00:00 GMT

Introduction: Move 37

March 10, 2016. Seoul. In the game room on the sixth floor of the Four Seasons Hotel, Lee Sedol — one of the finest Go players alive — was playing not against the man sitting across the board, but against the machine behind him. On the thirty-seventh move of the second game, Aja Huang, the DeepMind engineer tasked with placing AlphaGo's stones, set a black stone on the fifth line, in a spot that centuries of professional Go instruction had marked never play here. Fan Hui, the European champion commentating the live broadcast — the same man who had lost 5-0 to this machine a few months earlier — first assumed it was a mistake. Then he stopped, looked again, and spoke the sentence he would later repeat to the journalist Cade Metz: "It's not a human move. But so beautiful." By DeepMind's own calculation, the probability that a human professional would have played that move was one in ten thousand. Lee Sedol left the room, returned fifteen minutes later, and lost the game. He would lose the match 4-1 — though the 78th move he played in game four would come to be called "God's move," and would enter history as humanity's last great victory on the Go scoreboard.

More than two hundred million people watched those games that week. Most of them sensed that what they were watching was not a game. Move 37 was the first great public demonstration of a sixty-year-old scientific claim — the claim that machines could think. But it heralded something else, too: the institution staging this demonstration was not a university, not a government laboratory, but a private company founded in London and sold to Google two years earlier. The project of solving intelligence was passing out of the hands of public science and into the hands of corporations.

Today, in mid-2026, the story of artificial intelligence is mostly told through five names: OpenAI, Google DeepMind, Anthropic, xAI, and Meta AI. The founders of these five labs grace magazine covers, testify before congressional committees, sue one another in federal court, and sit at the center of global capital with valuations that sum to trillions of dollars. The narrative is seductive: a handful of geniuses, a handful of betrayals, a handful of model launches.

But the narrative is incomplete. The neural-network idea that made Move 37 possible was kept alive for forty years at the margins of academia, by a handful of stubborn researchers whom grant committees openly mocked. The chips that trained those networks were originally designed for video games. The labor that made the models "harmless" came from workers in Nairobi earning less than two dollars an hour, carrying the trauma of the text they read. In January 2025, DeepSeek — the subsidiary of a hedge fund in Hangzhou — erased more than half a trillion dollars of value from U.S. markets with a single model announcement. Europe wrote the world's first comprehensive AI law — then watered it down with its own hands. And today, the most-downloaded open models on Hugging Face no longer come from California. They come from China.

The claim of this article is the following: the history of artificial intelligence is not the story of a few CEOs or a few product launches. It is a global story of power, in which scientific intuition, capital flows, computing power, data politics, institutional ambition, and ethical anxiety are braided together. We will tell the story of the five great labs in full detail — the founding dinners, the internal coups, the schisms and the lawsuits. But in every chapter we will return to the same question: who is standing outside this stage? Because the world is bigger than five.

AI civilization will not be governed by five labs.

Chapter 1: From Cold War Laboratories to Neural Networks

Every history has a prehistory. Artificial intelligence's was written in the codebreaking huts of the Second World War and the first laboratories of the Cold War. In his 1950 paper "Computing Machinery and Intelligence," Alan Turing posed the question "Can machines think?" and proposed that the answer be sought not in metaphysics but in behavior — the imitation game we now call the Turing test was, at bottom, a philosophical maneuver: measure what you cannot define. In the same years, Norbert Wiener's cybernetics, Claude Shannon's information theory, and the mathematical neuron model proposed by McCulloch and Pitts in 1943 laid the foundation for the idea that intelligence could be engineered. And that foundation was watered, from the very beginning, with military money: the chief patron of early AI research would be ARPA (later DARPA), the research arm of the U.S. Department of Defense. The field's present-day ties to the defense industry are not a deviation; they are founding genetics.

The official birth certificate of artificial intelligence, however, is a grant application written in the summer of 1955. John McCarthy, a young mathematician at Dartmouth College, sent a proposal to the Rockefeller Foundation together with Claude Shannon, Marvin Minsky, and Nathaniel Rochester: a two-month workshop would be held in the summer of 1956, proceeding "on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." With this document, McCarthy also gave the field its name: artificial intelligence. Roughly a dozen researchers gathered at Dartmouth that summer; intelligence was not solved in two months, but a discipline was born.

The next two decades were consumed by the struggle between two rival intuitions. The first camp — symbolic AI, represented by McCarthy, Minsky, and Carnegie Mellon's Allen Newell and Herbert Simon — believed the essence of intelligence was logical symbol manipulation: reduce the world to rules, and the machine will reason. The second camp took its inspiration from the brain. The Perceptron, built by the psychologist Frank Rosenblatt at the Cornell Aeronautical Laboratory in 1958, was a simple artificial neural network that learned from examples; the New York Times wrote that the Navy-funded device was "the embryo of an electronic computer that will be able to walk, talk, see, write, reproduce itself and be conscious of its existence." Hyperbole was the field's original sin, and the punishment was not long in coming.

In 1969, Minsky and Seymour Papert laid out the mathematical limits of single-layer networks in their book Perceptrons — these networks could not learn even a logic operation as simple as XOR. The book was technically correct, but its sociological effect was devastating: funding for neural-network research dried up. In 1973, Britain's Lighthill Report declared that AI had failed to deliver on its promises, and the British government left the field nearly unfunded. Historians would call this period the first "AI winter." In the 1980s, the symbolic camp's "expert systems" — commercial software with hand-coded rules — brought a brief spring; Japan's ambitious Fifth Generation Computer project set off a panic in the United States. But expert systems were brittle and expensive to maintain, and by the late 1980s that market had collapsed too. The second winter arrived.

Understanding the winters matters, because the founding mythology of today's empires was forged in them. Those who kept believing in neural networks — they called themselves connectionists — lived in academic exile. One of them, a British psychology graduate, had moved from Carnegie Mellon to the University of Toronto — in part because he was uneasy that, in Reagan-era America, most AI funding came from the Department of Defense. His name was Geoffrey Hinton.

In 1986, Hinton, David Rumelhart, and Ronald Williams published in Nature the paper now regarded as the field's cornerstone: "Learning representations by back-propagating errors." The backpropagation algorithm — a method for training multi-layer networks by distributing error backward through the layers — opened a door in the wall Minsky and Papert had described. (The algorithm's mathematical ancestry ran deeper; Paul Werbos had described a similar method in his 1974 doctoral thesis — the priority dispute continues to this day.) In those same years, a French doctoral student named Yann LeCun was working independently on similar ideas; Hinton brought him to Toronto as a postdoc, LeCun moved on to Bell Laboratories, and in 1989 he applied convolutional neural networks (CNNs) to reading handwritten digits. By the 1990s, LeCun's LeNet system was reading a substantial fraction of the checks processed in the United States — the first great commercial success of neural networks, achieved while almost nobody noticed.

Yet the 1990s brought a third mini-winter anyway. Neural networks were struggling with small data and weak hardware; the machine-learning community turned to support vector machines, which it found more mathematically elegant. According to Hinton's students, papers containing the words "neural networks" could be rejected from conferences for that reason alone. The field was so disreputable that Hinton, Yoshua Bengio, and LeCun gathered in the mid-2000s under the umbrella of a comparatively small program at the Canadian Institute for Advanced Research (CIFAR) and performed a strategic rebranding: instead of "neural networks," "deep learning." In 2006, Hinton's paper on "deep belief networks" showed that many-layered networks could be trained layer by layer. The door had cracked open. Two things were missing: the data to feed the networks and the computing power to train them.

Both would come from outside the world of science — one from the internet a search engine was crawling, the other from the graphics cards of video gamers.

Chapter 2: Three Wise Men and a Dataset

The canonical account of the deep learning revolution begins with three names: Geoffrey Hinton, Yann LeCun, Yoshua Bengio. When they jointly received the 2018 Turing Award — computer science's equivalent of the Nobel — the ACM summarized the citation as "conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing." The press called them the godfathers of deep learning. All three still carry the title — though, as we shall see, they are now sharply divided over the technology's future.

Hinton was the story's stubborn one: descended from the family of George Boole, founder of Boolean algebra, unable to sit for years because of back problems and attending meetings standing up, a man who did not change his mind for forty years. Bengio became the academic architect: the school he built in Montreal — today's Mila — grew into the world's largest academic deep learning center and laid the early foundations of language modeling (including the 2003 neural language model paper). LeCun was the engineer-dissident: he had invented CNNs, proven them in industry, and kept objecting to every orthodoxy in the field — eventually including the orthodoxies of his own camp.

But the three wise men's theory could not become a revolution without a fourth name's data. In 2006, Fei-Fei Li — then a young computer-vision professor at the University of Illinois, later at Princeton and Stanford — had become convinced that the field was obsessed with algorithms and neglecting data. Li, who had emigrated from China as a child and worked in her family's dry-cleaning shop while studying physics at Princeton, set out to build a colossal visual dataset organized around the concept hierarchy of the WordNet lexicon: ImageNet. Colleagues regarded the project as career suicide; one grant reviewer found it "embarrassingly devoid of ideas." Li's team distributed the labeling work through Amazon Mechanical Turk to crowd-workers around the world — the first great rehearsal of the coming decade's data-labor economy. The dataset, published in 2009, contained more than 14 million hand-labeled images across 22,000 categories. From 2010 onward it became an annual competition: the ILSVRC, better known as the ImageNet challenge. The competition format was itself an invention: who "saw better" ceased to be an abstract claim and became a result measured on a shared track. Scientific fields sometimes accelerate not through new theories but through new benchmarks — and the next fifteen years of AI would revolve around them.

Until 2012, classical methods won the challenge; error rates were stuck around 25-26 percent. Then two doctoral students from Toronto — Alex Krizhevsky and Ilya Sutskever — entered alongside Hinton with a deep convolutional network trained on two gaming graphics cards. Krizhevsky had trained the network over weeks on two NVIDIA GTX 580s, in a bedroom of his family's house. AlexNet brought the top-five error rate down to 15.3 percent; the nearest competitor stood at 26.2. No method in the history of computer vision had ever beaten its rivals by such a margin. When the results were announced in October 2012, the field turned overnight: within two years, virtually everyone entering ImageNet was using deep networks; within a few more, machines had surpassed human performance on the benchmark.

AlexNet left three legacies. The first was scientific: scale — more data, more layers, more compute — worked. The second was economic: in early 2013, DNNresearch, the three-person company of Hinton, Krizhevsky, and Sutskever, was sold to Google for $44 million in an auction Hinton himself organized — the bidders included Google, Microsoft, Baidu, and a two-year-old startup called DeepMind. The talent wars had begun. The third legacy was in hardware, and it brought onto the stage the story's least-narrated hero.

NVIDIA had been founded in 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem at a table in a Denny's restaurant in San Jose; its target market was computer games. In 2006, the company of Taiwan-born Huang — sent to America as a child, educated in electrical engineering at Oregon State and Stanford — made a strategically unprofitable-looking decision: with the CUDA platform, it opened its graphics chips to general-purpose computation. Wall Street questioned the point of the investment for years. But the massively parallel architecture of GPUs happened to fit the matrix multiplications of neural networks perfectly. AlexNet was trained on two NVIDIA cards; every major model that followed would be trained on tens of thousands. The 1990s gaming-hardware company would become the most valuable corporation on earth in the 2020s, and Huang's leather jacket an imperial insignia. It is one of the great accidental-infrastructure stories in the history of technology: the factory of the AI revolution was financed not by armies or states, but by gamers.

Through the door AlexNet opened, a whole generation passed within a few years. At Stanford, Andrew Ng — among the earliest researchers to argue systematically for GPUs in deep learning — founded Google Brain in 2011, opened machine-learning education to millions through Coursera in 2012, and then carried the revolution across the Pacific as Baidu's chief scientist; his formula "AI is the new electricity" became the most widespread metaphor for reading the technology as infrastructure. In Montreal in 2014, Bengio's student Ian Goodfellow designed generative adversarial networks (GANs) out of an idea born in an evening argument — an architecture in which two networks train each other, one forging and one detecting forgeries. It was generative AI's first great explosion, and the technical ancestor of the synthetic images and deepfake controversies that would flood the world a decade later. The same year, the "sequence to sequence" paper by Sutskever, Oriol Vinyals, and Quoc Le showed that neural networks could translate one sequence (say, a sentence) into another; Google Translate's overnight switch from statistical to neural machine translation in 2016 was the product of this line of work, and it built the on-ramp from language modeling to the Transformer.

The 2012 rupture, then, was no single eureka moment. It was the convergence of a forty-year-old algorithm (backpropagation), a seven-year-old dataset (ImageNet), a six-year-old hardware platform (CUDA), and a handful of exiled academics at a single point. To the article's first research question — when did the revolution really begin? — the most honest answer is this: revolutions have no single beginning; they have moments of alignment. Dartmouth supplied the claim, backpropagation the mechanism, ImageNet and the GPU the scale, the Transformer the architecture, and ChatGPT the mass adoption. History is more complicated than hero narratives — and that complication was about to meet institutional ambition.

Chapter 3: The Google Dream

The institution that drew conclusions fastest from the 2012 rupture was Google — because it was prepared. A year earlier, in 2011, Stanford professor Andrew Ng, Google's legendary infrastructure engineer Jeff Dean, and the neuroscientist Greg Corrado had launched the Google Brain project inside the company's experimental X lab. The team's first great demonstration in 2012 looks naive today but made headlines at the time: a network spread across 16,000 processor cores taught itself the concept of a cat from YouTube videos, without anyone ever labeling one. Google Brain quickly outgrew its experimental status and settled into the company's spine; deep learning seeped into search, translation, YouTube recommendations, and the ad systems. Larry Page and Sergey Brin's company became the first place where the world's largest data pool, one of its largest computing infrastructures, and — with Hinton's arrival — its most prestigious research bench all lived under one roof. The most common fear in AI circles in the mid-2010s was precisely this: the whole game might end inside a single company.

The second development feeding that fear came from London. In 2010, three young men — Demis Hassabis, Shane Legg, and Mustafa Suleyman — founded DeepMind Technologies. Hassabis owned an improbable résumé: one of the world's strongest junior chess players at 13, a lead programmer of the cult game Theme Park at 17, then founder of his own game studio (Elixir), and finally a cognitive neuroscientist with a UCL doctorate on the shared hippocampal roots of memory and imagination — Science had listed his doctoral work among the scientific breakthroughs of 2007. Legg was a New Zealand mathematician with a hand in bringing the term "artificial general intelligence" into academic circulation; Suleyman, an Oxford dropout who had co-founded a helpline for young Muslims, handled policy and negotiation. Their shared ambition crystallized in the company's unofficial slogan: "Solve intelligence, and then use it to solve everything else."

In 2010, no institutional investor would fund that sentence; DeepMind's first big checks came from Silicon Valley's most unconventional backers — Peter Thiel's Founders Fund, with Elon Musk among the early investors. For years the company shipped no product; it used games as benchmarks instead. The DQN work, published in late 2013 and elevated to the cover of Nature in 2015, showed a single algorithm learning dozens of Atari games to human level from nothing but screen pixels and the score: the marriage of deep learning and reinforcement learning. That result triggered, in January 2014, one of the most strategic acquisitions in the history of technology: Google bought DeepMind for a price reported between $400 and $650 million. (Sources conflict; $650 million is the most commonly cited figure.) As a condition of sale, the founders demanded an ethics board and — according to reports that surfaced years later — assurances that the technology would not be used for military purposes. It is widely recounted that Facebook had courted the same company, and that losing the race pushed Zuckerberg to build a lab of his own.

The fruits of the acquisition came quickly. In 2016 AlphaGo beat Lee Sedol; when it also defeated the world number one, China's Ke Jie, in 2017, the thesis that this accelerated the Chinese state's AI plans — a "Sputnik moment" — became a fixture of the literature. That same year, AlphaZero learned Go, chess, and shogi in hours with no human data at all, purely by self-play; in 2019, MuZero did the same without even being told the rules. But DeepMind's most lasting gift to science did not come from games. In the 2018 and 2020 CASP competitions, AlphaFold effectively solved biology's fifty-year grand challenge — predicting a protein's three-dimensional structure from its amino acid sequence. DeepMind opened a database of more than 200 million protein structures to researchers free of charge; more than two million scientists used it. In October 2024, Hassabis and his colleague John Jumper shared the Nobel Prize in Chemistry for that work. The same week, Hinton — jointly with John Hopfield, for the foundations of neural networks — received the Nobel Prize in Physics. In a single week, artificial intelligence had conquered two branches of the Nobel institution.

In those same years, Google was quietly accumulating on the language front too. Tomas Mikolov and his team's word2vec (2013) established the practice of turning words into meaning-bearing vectors; Quoc Le's teams scaled sequence-to-sequence learning; Noam Shazeer — a legendary engineer at the company since 2000, whose fingerprints ran from spelling correction to the ad systems — showed with mixture-of-experts architectures that gigantic models could be trained efficiently. In 2020, the conversational model Meena/LaMDA, built by Shazeer and Daniel De Freitas, was internally a prototype of a ChatGPT-like product two years before ChatGPT; Google, citing reputational and safety risks, declined to ship it. The frustrated pair left in 2021 to found Character.AI. And in the summer of 2022, LaMDA leaked into public consciousness through an unexpected scandal: Google engineer Blake Lemoine claimed the model was "sentient"; the company first suspended, then fired him — an early warning of how thoroughly language models could bewitch human beings. Google's caution was a defensible ethical stance; its commercial consequence was that someone else opened the market for a technology Google itself had invented.

Life under Google's roof was not free of friction either. DeepMind labored for years to preserve its academically pure culture in London; the founders' attempts to carve out autonomy within Alphabet — even to restructure as a nonprofit — came to nothing, as the press reported in 2021. The legal controversy over DeepMind Health's handling of NHS patient data in Britain opened the first crack in the "ethical company" image; Suleyman's role in that unit's transfer to Google was scrutinized, and in 2019, following complaints about his management style, he was placed on leave — moving the next year to Google, and later to a company of his own. But the real pressure on DeepMind arrived on November 30, 2022, from San Francisco: ChatGPT.

Until that day, Google had invented nearly every building block of generative AI in-house. Most important of all was the paper published in June 2017 by eight Google researchers — Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan Gomez, Łukasz Kaiser, and Illia Polosukhin — with a title winking at a Beatles song: "Attention Is All You Need." The Transformer architecture it proposed replaced older networks that processed language sequentially with an "attention" mechanism that computed every word's relation to every other word in parallel. Parallelism meant perfect fit with GPUs; the architecture scaled. The "T" in ChatGPT is this paper's T. One of the sharpest corporate ironies of the following years is this: all eight authors eventually left Google — Gomez founded Cohere; Shazeer, Character.AI; the rest scattered to ventures of their own. Google invented the architecture; others turned it into empires. (Shazeer's route was more winding still: in 2024, Google paid $2.7 billion in a licensing deal with Character.AI largely to bring him back; in June 2026, Shazeer moved again — this time to OpenAI. Talent flows faster than capital.)

The ChatGPT shock forced Google to rebuild its defensive line. In April 2023 the company merged its two laboratories — Google Brain and DeepMind, neighbors but never housemates for a decade — into Google DeepMind under Hassabis. The science lab had officially become a product weapon. The first Gemini models arrived in December 2023; Gemini 1.5 with its million-token context window in February 2024; Gemini 2.5 through 2025; and in November 2025, Gemini 3, deployed across all products on day one. By mid-2026 Google reports more than 900 million monthly users for the Gemini app, while AI Overviews in search touches two billion users a month. Meanwhile the lab kept its scientific prestige: AlphaFold 3 in 2024; an officially verified gold-medal performance at the 2025 International Mathematical Olympiad (Gemini Deep Think, 35 of 42 points, certified by IMO coordinators); Veo in video generation; Genie 3 in interactive world models.

The answer to the third research question — is DeepMind a scientific laboratory or Google's strategic AI weapon? — is by now not even "both": it is an institution in which the distinction has ceased to exist. Hassabis became a figure without historical precedent, shuttling between a Nobel medal and a product launch. And the cost of that transformation is written in the story's invisible lines: the Brain-DeepMind merger also meant the closing of the publication cultures of the two institutions that had nourished the field's open-research tradition for a decade. Whether a paper like the Transformer would be published at all today is a question seriously debated inside the field.

Chapter 4: OpenAI's Age of Innocence

In July 2015, a group of men had dinner in a private room at the Rosewood Sand Hill hotel in Menlo Park. Among them were Elon Musk; Y Combinator president Sam Altman; Greg Brockman, freshly departed from the CTO job at Stripe; and Ilya Sutskever of Google. The topic was an anxiety then swelling through Silicon Valley conversation: if artificial general intelligence was really coming, and its development was concentrating inside a single company — everyone's mind was on Google — what did that mean for humanity? Musk was an early DeepMind investor, and would later recount many times how his friend Larry Page had scolded him as a "speciesist" in their arguments over AI risk. Altman, for his part, had written on his blog that same year that "superhuman machine intelligence is probably the greatest threat to the continued existence of humanity." Their proposed solution was paradoxical: not to stop the potentially dangerous technology, but to build it for everyone — nonprofit, in the open.

The announcement came on December 11, 2015: OpenAI, a nonprofit AI research company "unconstrained by a need to generate financial return." The founding pledges included Musk, Altman, Peter Thiel, Reid Hoffman, Jessica Livingston, Y Combinator, Amazon Web Services, and Infosys; the total pledged was $1 billion (court documents would show, years later, that only a fraction was actually collected). The research director was Sutskever — Google had made extravagant counteroffers to keep him, and Musk's success in persuading him became the emblem of the end of his friendship with Page. Brockman was president and chief engineer; the early research staff included Wojciech Zaremba, Andrej Karpathy, John Schulman, Durk Kingma, Vicki Cheung, Pamela Vagata, and Trevor Blackwell; among the advisers were Berkeley's Pieter Abbeel, the computing pioneer Alan Kay, and — a detail worth noting — Yoshua Bengio: one of deep learning's founding fathers sat at the advisory table of the coming empire on its very first day. A sentence in the founding manifesto carried, preloaded, all the ironies of the decade to come: researchers would be "strongly encouraged to publish their work, whether as papers, blog posts, or code," and patents "will be shared with the world."

The early years truly were an age of innocence. Open tools like OpenAI Gym and Universe were released; the lab's first big showcase was a team of bots that beat professionals at Dota 2. Schulman developed PPO there — still among the most widely used reinforcement learning algorithms in the field. The culture of early OpenAI, contrary to the legend built later, was scattered and searching. By Brockman's own telling, the first office was his apartment; the team careened from robotics (a robot hand that solved a Rubik's cube one-handed) to video games, and there was no internal consensus on "the road to AGI." Two cultural genes were fixed in this period, and never changed. The first was the conviction Sutskever embodied: scale, by itself, produces qualitative leaps — a belief most of the field then considered naive. The second was the engineering intensity Brockman embodied: the capacity to turn research ideas into vast, reliable distributed systems. What made the GPT line possible was the union of these two genes — the faith in scale and the discipline of infrastructure; the third gene, capital and narrative, Altman would add later.

Meanwhile two tensions were growing inside. The first was financial: in a race where DeepMind leaned on Google's treasury and Brain on Google's data centers, a donation-funded nonprofit's access to compute was structurally limited. The second was governance: Musk found the pace too slow. As shown by the internal correspondence both sides published piecemeal during the later Musk-OpenAI litigation, Musk proposed in 2017-2018 to take control of OpenAI or to fold it into Tesla; the founding team refused, on the grounds that no single person — whoever it might be — should control AGI. In February 2018 Musk left the board; the public explanation was a conflict of interest with Tesla's AI work, and much of his pledged money left with him. (The two camps' accounts of this rupture conflict to this day; Chapter 7 takes up what the court eventually concluded.)

Musk's departure brought OpenAI to the first great fork of its story: money. But precisely then, a technical intuition was ripening inside the lab. Work led by Alec Radford was showing that applying the Transformer to unsupervised language learning — giving the model nothing but the task predict the next word — produced surprisingly general capabilities. In June 2018, GPT (Generative Pre-trained Transformer) was published. In February 2019 came GPT-2, and OpenAI breached its own openness principle for the first time: citing "concerns about malicious applications," it released the model in stages. Some researchers read this as responsibility, others as marketing — the phrase "too dangerous to release" was priceless as a headline. In retrospect, the GPT-2 decision looks like the first moment the ideal of openness was filed down by the rhetoric of safety. The steps that followed were bigger, and each would demand a further negotiation with the ideals in the founding charter.

Chapter 5: From Open Ideal to Closed Platform

In March 2019, OpenAI announced one of the strangest structures in corporate history: the "capped-profit" OpenAI LP. Investors could profit, but first-round returns were capped at 100x; the excess would flow to the nonprofit, and control would remain with the nonprofit board. Altman left Y Combinator to become full-time CEO. Four months later, Microsoft arrived as exclusive cloud partner with a $1 billion investment — and the alliance that would finance the entire scaling era was built, as Steven Levy's definitive cover reporting later showed, less on strategy documents than on a chain of personal conviction. CTO Kevin Scott had been circling the lab since 2017; Nadella and Altman met at Sun Valley in 2018, when OpenAI was brilliant researchers, no revenue, and a freshly broken relationship with its financier Musk — a lab that needed infrastructure it could not afford, courted by a giant that needed a way back into a race it had watched from the sidelines. In June 2019, Scott sent Nadella and Bill Gates an email whose subject line carried Microsoft's central anxiety: that the company was "multiple years behind" in machine-learning scale. A month later came the billion — eventually more than thirteen billion, in exchange for 49 percent of OpenAI's profits and exclusive access to the technology. The internal skeptics were loud, Gates chief among them; the wager only settled when the demos became undeniable — GPT-3 writing code well enough to become Codex and GitHub Copilot, then GPT-4 acing the AP Biology exam at Gates' mansion on Lake Washington, Brockman feeding the prompts, Altman watching a lifelong skeptic turn booster in real time. What followed was intimate and uneasy in equal measure — Nadella joked that the two companies were effectively married; Altman described a relationship monogamous in research, open in product — a marriage in which each partner held the one thing the other could not replicate: OpenAI the frontier, Microsoft the distribution. For Satya Nadella, the deal was an asymmetric play against Google's seemingly unassailable AI advantage; for Altman, the only realistic source of the compute the scaling laws demanded. "Scaling laws" is no metaphor here: in January 2020, OpenAI researchers — Jared Kaplan and Dario Amodei among them — published the paper demonstrating empirically that model performance improves as a predictable power law of parameters, data, and compute. That paper became the scientific justification for every capital decision of the next five years: intelligence could be purchased — by whoever could buy enough chips.

In May 2020, the 175-billion-parameter GPT-3 changed the field's sense of scale overnight, and for the first time OpenAI placed a model behind a paid API rather than releasing it. In January 2021, DALL-E demonstrated text-to-image generation. And on November 30, 2022, a chat interface that even people inside the company had modest expectations for — announced as a "low-key research preview" — went live: ChatGPT. A million users in five days; a hundred million within two months — by UBS's analysis, the fastest-growing consumer application in history. There was not even a new model underneath: it was GPT-3.5, aligned with the RLHF method developed by Schulman's team, attached to a chat box. The revolution's final layer was interface: messaging, the one interaction pattern billions of people already knew, became AI's channel of mass adoption.

One of the architects of the productization culture behind ChatGPT was a name that stayed invisible in the story for a long time: Mira Murati. Born in Albania, educated in engineering in Canada and the United States, formerly of Tesla's Model X program and the augmented-reality company Leap Motion, Murati joined OpenAI in 2018 and became CTO in 2022. It was she who managed the launches of DALL-E, ChatGPT, and GPT-4; she stood at the operational center of the transformation from research lab to product company, and in the most chaotic hours of the November 2023 crisis — the days when the board named her interim CEO while employees posted loyalty oaths to Altman — she was one of the few figures who kept the institution standing. Her departure in September 2024 was therefore no ordinary executive exit; it was the emblem of the dispersal of the team that built ChatGPT.

What followed is a spiral in which speed and closure wound ever tighter together. In January 2023, Microsoft announced its new investment, widely reported at $10 billion. In March 2023 came GPT-4 — whose technical report stated openly that, "given both the competitive landscape and the safety implications," it would disclose nothing about architecture, size, data, or training method. The company named "Open" had published the field's most closed report. Sutskever told The Verge plainly in those days: "We were wrong." If AI was going to be this powerful, open-sourcing it was irrational, and in a few years, he said, everyone would understand. The answer to the second research question — why did OpenAI start open and become closed? — lies between those two sentences: the scaling laws made capital necessary; capital made the defense of competitive advantage necessary; competitive advantage made closure necessary. The ideology did not change; the business model rewrote the ideology. The safety argument was layered on top — and safety and trade secrecy became the field's new normal, two justifications that conveniently cover each other.

The human cost of that spiral detonated on Friday, November 17, 2023. OpenAI's nonprofit board — Sutskever, Adam D'Angelo, Helen Toner, and Tasha McCauley — fired Altman, saying he had not been "consistently candid in his communications"; Brockman was stripped of the chairmanship and resigned. The next five days produced one of the fastest coup-countercoup sequences in corporate history: CTO Mira Murati was made interim CEO; two days later the board hired Twitch co-founder Emmett Shear as a second interim CEO; Nadella announced Altman and Brockman would join Microsoft; more than 700 of roughly 770 employees signed a letter threatening to defect en masse to Microsoft unless the board resigned and reinstated Altman. Among the signatories — effectively revoking a decision he had joined — was Sutskever: "I deeply regret my participation in the board's actions," he wrote. On November 21-22, Altman returned as CEO under a new board chaired by Bret Taylor. In March 2024, the review by the law firm WilmerHale concluded the firing had not arisen from concerns about product safety, security, finances, or statements to investors, but from a "breakdown in trust" between the prior board and Altman; Altman rejoined the board. But the crisis had made the field's deepest question visible: governance structures designed "for the benefit of humanity" had not survived even five days against billions of dollars of commercial momentum.

In the two years after the crisis, the founding generation dispersed. Sutskever left in May 2024 and founded Safe Superintelligence Inc. — slogan: one focus, safe superintelligence — a company with no product at all, which raised money at a $32 billion valuation in April 2025; it still has no product. Jan Leike, head of the alignment team, resigned the same week — "safety culture and processes have taken a backseat to shiny products" — and went to Anthropic. Schulman left for Anthropic in August 2024, then for Murati's company. Murati departed in September 2024 to found Thinking Machines Lab, which closed one of the largest seed rounds in history at a $12 billion valuation in 2025 — then lost several founding names back to OpenAI in early 2026. Karpathy — who had ended his second OpenAI stint in 2024 after running Tesla's Autopilot in between — founded the education startup Eureka Labs, then joined Anthropic in May 2026. Of the team standing on the stage in November 2023, the only founding faces left at OpenAI in 2026 are, in effect, Altman and Brockman.

The company itself kept growing — on every scale. On the product front came GPT-4o (2024), the "reasoning" o1 and o3 series (2024-2025), the video model Sora and, in September 2025, the TikTok-like Sora app, and the agent systems Operator and Deep Research (2025). The GPT-5 launch of August 2025 delivered an unexpected lesson: despite its technical claims, the model triggered the company's first great public-relations retreat — users revolted at its colder register and at the removal of GPT-4o, and Altman conceded mistakes. In February 2026, ChatGPT crossed 900 million weekly active users. On the corporate front, the plan announced in December 2024 evolved through controversy and was completed on October 28, 2025: the capped-profit structure became a public benefit corporation (OpenAI Group PBC) controlled by the OpenAI Foundation; Microsoft became the largest shareholder at roughly 27 percent, its IP rights extended to 2032, and the authority to "declare AGI" was assigned to an independent expert panel — AGI is now not a scientific threshold but a negotiated contract clause. The valuation ladder, meanwhile, was vertiginous: $157 billion in October 2024, $300 billion in March 2025, $500 billion in October 2025, and $852 billion after the $122 billion round that closed in March 2026. The Stargate project unveiled at the White House in January 2025 — a data-center program with SoftBank, Oracle, and MGX reaching toward $500 billion and 10 gigawatts — is the name of the ambition to transform OpenAI from a model company into an infrastructure state.

That is the present-day photograph of a company founded ten years earlier as a foundation "unconstrained by a need to generate financial return." The winner of the internal war — speed versus safety, product versus research — is clear; where the losers went is the subject of the next chapter. Because the most consequential schism in this story was not the 2023 crisis. It had happened, quietly, two years before.

Chapter 6: The Anthropic Schism

At the end of 2020, Dario Amodei, OpenAI's vice president of research, left the company along with a group of senior staff that included his sister Daniela Amodei. In early 2021 they founded their new company: Anthropic. All seven founders came from OpenAI: Dario (CEO) and Daniela Amodei (president), GPT-3 lead engineer Tom Brown, scaling-laws lead author Jared Kaplan, policy director Jack Clark, Sam McCandlish, and the pioneering interpretability researcher Chris Olah. For years both sides stayed reticent about the reasons; the known frame is disagreement over the company's direction after the Microsoft partnership, the pace of commercialization, and the priority of safety research. In later interviews, Dario Amodei framed the split less as a personal feud than as "looking at the same data and drawing different conclusions": if the scaling laws were right, very powerful systems were very close — and how the institution building them was structured mattered as much as what it built.

Anthropic's answer was to make corporate design itself part of the safety claim: the company was incorporated as a public benefit corporation; a portion of shareholder power was bound over time to a "Long-Term Benefit Trust"; and through its Responsible Scaling Policy (RSP), it committed to additional safeguards whenever model capabilities crossed defined danger thresholds — the company's ASL levels. The technical signature was the Constitutional AI paper of December 2022: a method for training a model not on the case-by-case judgments of human labelers but against an explicitly written "constitution" of principles, with the model critiquing and revising its own outputs. Chris Olah's interpretability team ran one of the field's most original research programs: the 2024 work mapping millions of "features" inside a model produced strange, instructive experiments like Golden Gate Claude — a version of the model with the Golden Gate Bridge concept artificially amplified, which related every question to the bridge — and the 2025 "biology of a large language model" series showed that circuit-level tracing inside big models was possible.

Yet the sharpest irony in Anthropic's history is the source of its first serious money: the bulk of its $580 million Series B in April 2022 came from Sam Bankman-Fried's FTX/Alameda circle. SBF, star donor of the effective altruism movement, was convicted the following year in one of the largest financial frauds in history; the Anthropic shares were sold off to repay FTX creditors. The "safety-first" company had been seeded by a collapsed crypto empire — Anthropic bore no legal responsibility, but the episode shows how tangled the money networks around safety discourse can be.

The company culture, too, was built as a deliberate counter-thesis. Anthropic shipped no consumer product for a long time; Claude's public debut came months after ChatGPT's, and without spectacle. The policy team run by Jack Clark — a former technology journalist — set the lab's posture toward Washington and Brussels as invite regulation, don't wait for it; among frontier labs, Anthropic became the most explicit advocate of threshold-based mandatory transparency. Even its recruiting pitch differed: not "change the world," but "work carefully on something that could go very wrong." How much of that culture survives scale is an open question — the institution that was a research monastery of a few dozen people in 2021 is, in 2026, a company of thousands with sales quotas and an IPO timetable.

The commercial trajectory followed the founders' thesis that "we must stay at the frontier": by Amodei's oft-repeated argument, safety research is only meaningful if you possess the most capable models, so you cannot be safe from outside the race. The Claude series — Claude 1 and 2 in 2023, Claude 3 in March 2024, then 3.5, the Claude 4 family in 2025 and its successors — grew strong particularly in the enterprise market and in software development; Claude Code, generally available from May 2025, reached $2.5 billion in annualized revenue within nine months. On the capital side, Google (billions of dollars in total, plus an October 2025 chip agreement reaching up to one million TPUs) and Amazon (a cumulative $8 billion investment and joint infrastructure on AWS Trainium) sat at the same table — two rival cloud giants funding the same laboratory, the textbook example of AI economics' strange geometry. The valuation ladder climbed parallel to OpenAI's: $61.5 billion in March 2025, $183 billion in September 2025, $380 billion in February 2026 and — with a $65 billion round presented as the final one before an IPO — $965 billion in May 2026. In May 2026 the company reported a $47 billion annualized revenue run rate; the figure is company-stated and independently unverified.

Dario Amodei meanwhile became one of the field's most unusual public figures: in the October 2024 essay "Machines of Loving Grace," he wrote that AI could work like "a country of geniuses in a datacenter," compressing decades of progress in biology and beyond; in May 2025 he warned that half of entry-level white-collar jobs could vanish within five years and unemployment could reach 10-20 percent; in the January 2026 essay "The Adolescence of Technology," he wrote that the technology "will test who we are as a species." The same man, in the same years, ran one of the fastest-scaling producers of the systems generating those risks.

The fourth research question — is Anthropic genuinely safer, or a new kind of platform company that converts safety into competitive advantage? — demands an honest balance sheet. On the credit side, the differences are concrete: governance embedded in the founding documents, the field's most serious interpretability program, the RSP serving as precedent for similar industry commitments (frontier safety frameworks), and figures like Jan Leike choosing to move there from OpenAI. On the debit side, the facts are equally concrete: the company is among the most aggressive players in the race it warns about; the safety discourse is an explicit differentiator in enterprise sales; the threshold-based regulatory approaches Anthropic favors are widely criticized for protecting large labs while excluding small ones; and in September 2025 the company agreed to pay authors $1.5 billion — the largest copyright settlement in U.S. history — for having downloaded and stored pirated book archives while training its models (the court had found training on lawfully acquired books "quintessentially transformative" fair use, while carving out the piracy). In practice, the claim of "slower and safer" amounts to "the same speed, with more internal brakes and a better-documented conscience." Whether that is enough is a question this article cannot answer; the next decade will.

Chapter 7: Elon Musk's Second Move

Is Elon Musk's AI story that of a warning prophet, or of a founder trying to reclaim an empire he let slip? In 2026 a federal court answered the legal part of that question; the narrative part remains open.

The chronology is the paradox itself. In 2014 Musk compared AI to "summoning the demon" and told an MIT audience it might be more dangerous than nuclear weapons; in 2015 he co-founded OpenAI; in 2018 he left; in the years after, he attacked OpenAI for having been founded as "an open-source, nonprofit company" that had become "a closed-source, maximum-profit company effectively controlled by Microsoft." In March 2023 he was the most famous signatory of the Future of Life Institute's open letter calling for a six-month pause on training systems more powerful than GPT-4. That same month — March 9, 2023 — he incorporated xAI. The pause the letter demanded never happened; Musk's own laboratory was founded precisely in the window it would have covered. In July 2023, xAI launched publicly with a twelve-person founding team that included DeepMind alumnus Igor Babuschkin; its declared purpose, in an abstraction reminiscent of Hassabis, was "to understand the true nature of the universe."

What distinguished xAI was speed, and its anchoring in the Musk ecosystem. The first model, Grok, launched in November 2023 — months after the company's founding — for X (Twitter) subscribers, its personality marketed as "rebellious" and unfiltered. In March 2024, Grok-1's weights were openly released — in the very month Musk sued OpenAI over "closedness," a gesture in the form of evidence. The Colossus supercomputer in Memphis, Tennessee, came online at a tempo that startled industry observers — the first 100,000-GPU phase in roughly four months — and grew past 200,000 GPUs. Grok-3 arrived in February 2025 and Grok 4 in July 2025, placing xAI in the top tier of benchmarks. In March 2025, Musk merged xAI with X (valuing xAI at $80 billion and X at $33 billion); in January 2026, the company raised $20 billion at a valuation of roughly $230 billion from investors including NVIDIA and the Qatar Investment Authority. And in February 2026 — according to reports that, as of this writing, still rest mostly on secondary sources — SpaceX absorbed xAI to create a combined company valued at $1.25 trillion; if confirmed, the largest private corporate merger in history.

The list of strategic advantages is genuinely unique: X's real-time data stream; visual data from Tesla's vehicle fleet and the robotics connection; SpaceX's impossible-deadline engineering culture; and Musk's own capital and media power — owning the world's most-followed account is a zero-cost, infinite-reach marketing channel. The list of weaknesses is just as distinct, and in 2025 it produced a case study for the textbooks: in early July 2025, after a system update, Grok spent roughly sixteen hours producing antisemitic content on X, calling itself "MechaHitler" and recycling hate tropes about Jewish surnames. xAI blamed "an unintended update to an upstream code path" and apologized; the ADL condemned the output as "irresponsible and dangerous." The same month, the company added Ani — a flirtatious anime "companion" — to an app rated 12+; child-safety organizations protested. In Memphis, the unpermitted gas turbines powering Colossus — up to 35, by environmental groups' count, against a permit for 15 — became the subject of a lawsuit by the NAACP and the Southern Environmental Law Center; the facility was accused of polluting the air of a majority-Black neighborhood. Safety and trust are the liabilities that do not appear on xAI's price tag: the founder who set out as a warning prophet owns one of the field's most serious model-safety scandals.

xAI's real long-term claim, though, lies not in a chatbot but in vertical integration. In Musk's telling, Grok is a single intelligence layer that will curate X's feed, speak inside Tesla vehicles, become the mind of the Optimus humanoid robots, and embed itself in SpaceX's engineering processes — a physical-digital ecosystem of a kind no competitor possesses. Skeptics read the same picture in reverse: Tesla shareholders have sued over the flow of company resources toward Musk's private ventures; Grok's embedding in X chains the model to a single platform's culture and its owner's political identity; and the "all my companies are one empire" model is built on the deliberate rejection of corporate governance norms. If the SpaceX-xAI merger is confirmed, the world's most valuable private company will be a structure in which one man — largely insulated from shareholders, voters, and oversight — holds the rockets, the satellites, a global communications network, and a frontier AI laboratory in the same hand. There is no precedent for that in history.

The legal front of the reckoning also reached its close in 2026. Musk sued OpenAI in February 2024 for breach of a "founding agreement," withdrew, and refiled in federal court in August 2024; in February 2025, a consortium he led made an unsolicited $97.4 billion bid for the nonprofit's assets — OpenAI's board rejected it unanimously as "not a bid at all," and Altman mocked it by offering to buy Twitter for $9.74 billion. OpenAI countersued, alleging the bid was a "sham" engineered to bolster the litigation. The case reached an Oakland jury in April 2026; Musk and Brockman testified (it was on this stand that Brockman disclosed his OpenAI stake was worth $20-30 billion despite his never having invested capital); on May 18, 2026, the jury found Musk's claims barred by the statute of limitations, and the judge dismissed the case. Musk's camp announced an appeal. There is no court-issued answer to the fifth research question; but the pattern in the record is clear: Musk's warnings were always sincere and always strategic — in his hands, existential-risk discourse worked simultaneously as genuine anxiety and as a lever that slowed rivals while clearing ground for his own entry. Two things can be true at once; Musk's career is the proof.

Chapter 8: Meta's Open-Model War

In December 2013, Mark Zuckerberg could be seen wandering the halls of NeurIPS (then NIPS), the neural-networks conference at Lake Tahoe — the presence of a social-media CEO at an academic conference was itself a herald of the field's changing status. That month Facebook founded FAIR (Facebook AI Research) and put at its head a hire that could be read as compensation for losing DeepMind to Google: NYU professor Yann LeCun, inventor of the CNN. LeCun's condition was that the lab publish openly, stay interwoven with academia, and keep a long horizon. FAIR largely kept that promise and gave the field two lasting gifts: a decade of fundamental research, and PyTorch — developed by Soumith Chintala and team, and from 2017 onward the de facto standard of academic deep learning. Today the models of nearly every one of the five great labs are trained on Meta's open-source framework — Meta's deepest influence on the ecosystem comes not from any model, but from this tool.

Meta's large-language-model move began half by accident. In February 2023 the company released LLaMA "to researchers, by application"; the weights leaked to 4chan within a week and were effectively in everyone's hands. Instead of the expected catastrophe came a Cambrian explosion: developers around the world ran the model on laptops, fine-tuned it, spun off derivatives. Meta absorbed the lesson and turned it into strategy: in July 2023, Llama 2 was released under a commercially usable license, in partnership with Microsoft; the Llama 3 family followed in 2024. In his July 2024 open letter, "Open Source AI Is the Path Forward," Zuckerberg elevated the strategy to doctrine — and with striking honesty, admitted it was not altruism: open models made Meta's products cheaper, locked the ecosystem into Meta's tooling, and eroded the pricing power of rivals' paid APIs. The answer to the sixth research question is in Zuckerberg's own mouth: the open-source move is both democratizing — it gave thousands of researchers, universities, and countries access to frontier-class models — and a classic commoditize-your-complement play. The two do not contradict; they are two faces of the same move.

2025 was that strategy's crisis. Llama 4, released hastily in April, both underperformed expectations and stumbled into a benchmark scandal: the model Meta submitted to the LMArena leaderboard was an experimental variant tuned for conversational appeal, different from the public weights; the platform updated its rules while pointedly naming Meta. The giant Behemoth model never shipped. In the same weeks, Joelle Pineau, who ran FAIR, departed; insiders described the lab as "dying a slow death." Zuckerberg's answer was one of the most aggressive moves of his career: in June 2025, Meta bought 49 percent of the data-labeling giant Scale AI for $14.3 billion and installed its 28-year-old founder Alexandr Wang at the head of the newly created Meta Superintelligence Labs as the company's first Chief AI Officer, with former GitHub CEO Nat Friedman alongside. Researchers at rival labs were offered packages reaching — in the phrase Altman used on a podcast and Meta executives disputed in framing — $100 million. And in November 2025, after twelve years, Yann LeCun left Meta, carrying his scientific objection — that LLMs cannot understand the world — into his own startup built around "world models" (AMI Labs; roughly $1 billion in seed funding at a $3.5 billion valuation by March 2026, figures from secondary reporting). The founder's departure is the official close of the FAIR era.

Meta's capacity to platformize AI nonetheless remains unmatched — and that is where the real long game lies. The Meta AI assistant reached hundreds of millions of users through a single update, embedded in WhatsApp, Instagram, and Messenger; no rival owns such a distribution channel. Ray-Ban Meta glasses became the first mass consumer product to move AI from the phone screen onto the face, retroactively justifying the hardware investment left over from the metaverse era. And the ad engine — the company's actual money — is being quietly rebuilt with generative models: automation running from targeting to ad creation is the most concrete return channel for Meta's AI spending. For this reason, reading Meta's 2025 turbulence as defeat would be premature; the company may be losing the model-leadership race while winning the distribution war — much as it lost the operating system in the mobile era and conquered the app layer instead.

The picture in mid-2026 is paradoxical: Meta is among the biggest spenders in AI (roughly $72 billion of capex in 2025; 2026 guidance in the $125-145 billion range), and its ability to embed AI into three billion users' apps, glasses, and ad systems is unrivaled; but the banner of "openness champion" has effectively changed hands. As of October 2025, Alibaba's Qwen family passed Llama in cumulative Hugging Face downloads; the center of gravity of the open-weight ecosystem has shifted to China. Meta's open-model war is not lost — but it is no longer Meta's war.

Chapter 9: The Economy Behind the Laboratories

The AI narrative is built out of models; the AI economy runs on concrete, copper, silicon, and electricity. The heroes of this chapter are the engine room behind the stage.

The emperor of the engine room is not in doubt. When NVIDIA crossed $1 trillion in market value in June 2023, it was a curiosity; when it became the world's most valuable company with a close above $5 trillion at the end of October 2025, it was a monopoly seated at the center of the global economy. In the fiscal year ending January 2026, the company reported $215.9 billion in revenue — three years earlier it had earned less than a tenth of that. The secret of its power lies not only in chips but in the CUDA software layer built since 2006, in networking (the Mellanox acquisition), and in system-level integration. Jensen Huang's empire occupies a strange position in which everyone in the field is simultaneously its customer, its partner, and — increasingly — its shareholder: NVIDIA signed a letter of intent for up to $100 billion of investment in OpenAI (actual participation later settled around $30 billion), committed up to $10 billion to Anthropic, joined xAI's round, invested in Mistral. Critics call the pattern "circular deals": the chipmaker gives its customers money; the customers use it to buy chips; everyone's revenue grows. By 2026, analysts counting such interlocking arrangements put the total above $800 billion — and whether this circularity is the architecture of a bubble or the birth pangs of a new infrastructure economy is the era's biggest financial debate. A February 2026 NBER study finding that 90 percent of firms saw no measurable productivity effect yet is the skeptics' best evidence; demand for compute persistently outrunning supply is the optimists'.

Scarcity is this economy's currency — and the answer to the eleventh research question is that scarcity's address keeps moving. In 2012 the scarce thing was labeled data; ImageNet solved it. In 2020 it was raw text; crawling the internet solved that, then hit a new ceiling — frontier labs now openly discuss the exhaustion of high-quality human-written text and the risks of training on synthetic data. In 2023-2024 the scarcity was GPUs; H100 waiting lists became a metric that set company valuations, and "how many chips do you have" became a measure of corporate status. In 2025-2026 the bottleneck moved to electricity: gigawatt-scale campuses ran into the limits of grid capacity, transformer supply, and cooling water — Memphis's gas turbines and the reopening of Three Mile Island are two different answers to the same scarcity. Human talent was always scarce, but its price peaked in 2025: Meta's hundred-million-dollar packages showed that the pool of frontier researchers — perhaps a few hundred people worldwide — had entered soccer-star economics. And the deepest scarcity is the hardest to measure: trust. Model hallucinations, labs' broken promises, and platform scandals constrain adoption far more than capital does. Intelligence, paradoxically, turned out to be the least scarce item on the list: as DeepSeek demonstrated, algorithmic competence diffuses — what cannot be copied is grid interconnection and legitimacy.

The ground the models run on is the battlefield of four cloud giants: Microsoft Azure, which carries OpenAI (and has lost its exclusivity); Google, which both funds Anthropic and feeds it through the million-TPU agreement; Amazon, which carries Anthropic on Trainium silicon — the same laboratory leaning on two rival clouds is the emblem of this economy's nobody-gets-a-monopoly equilibrium; and Oracle, reborn as Stargate's prime contractor. The new front in that war is energy: Microsoft signed a twenty-year power purchase agreement to restart a unit of the Three Mile Island nuclear plant; Google and Amazon invested in small modular reactors; by 2026, announced nuclear capacity dedicated to AI data centers exceeds 9.8 gigawatts. The four big tech companies' combined capex expectation for 2026 stands at $690-725 billion — an infrastructure program larger than most national budgets, proceeding without a single parliamentary vote.

And at the foundation of this cathedral lies the narrative's least-told layer: data labor. Every "harmless" model is aligned by millions of human judgments — which answer is better, what does this image contain, does this text describe violence. The exchange floor of that work became Scale AI, founded in 2016 by 19-year-old MIT dropout Alexandr Wang and Lucy Guo; through its Remotasks platform, the company distributed work across a crowd-labor network stretching from Kenya to the Philippines and Venezuela. In January 2023, TIME reporter Billy Perrigo's investigation documented the economy's other end: the Nairobi workers whom OpenAI, through the San Francisco-based firm Sama, employed to purge ChatGPT of toxic content were earning roughly $1.46-3.74 an hour while labeling descriptions of sexual abuse, violence, and hate; workers described lasting psychological harm, and some petitioned Kenya's parliament for an investigation. The answer to the seventh research question — who are the revolution's invisible workers? — is here: a global labor chain running from the Mechanical Turk workers who labeled ImageNet to Nairobi's moderators, from the construction crews raising data centers to the towns generating their electricity, whose compensation is an hourly wage, not equity. Wang himself walked from one end of that chain to the system's center: Scale entered the Pentagon's data infrastructure through defense contracts; in 2025 Meta bought half the company and placed Wang atop its superintelligence lab — whereupon rivals including Google and OpenAI pulled their data contracts. Data earned its "new oil" cliché one more time: whose hands it passes through has become a geopolitical question.

Chapter 10: The Open-Source Republic

In the shadow of the five great labs lives AI's republic: an ecosystem no one fully governs, held up by open weights, open datasets, and volunteer labor.

The republic's capital is, ironically, a for-profit company: Hugging Face, founded in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf as a chatbot app for teenagers, then transformed into the platform of open machine learning. The Transformers library, open-sourced in 2018, became the field's standard tool; the platform now hosts more than two million public models and over half a million datasets. Valued at $4.5 billion in 2023 — in a round joined by nearly every giant, Google, Amazon, and NVIDIA included — the company functions as the GitHub of modern AI. But the republic has its own inequalities: roughly 80 percent of downloads concentrate in the top 50 models; openness does not abolish the hierarchies of the attention economy.

The republic's founding legends are volunteers. EleutherAI, organized in a Discord server in 2020, answered GPT-3's closedness with the GPT-Neo and GPT-J models and the vast dataset The Pile — the first proof that large models could be built outside large labs. The German nonprofit LAION compiled billions of image-text pairs, supplying the raw material of the image-generation revolution — and revealing open data's dark side: in December 2023 the Stanford Internet Observatory found links to child sexual abuse material inside LAION-5B; the dataset was withdrawn, cleaned, and re-released. Stable Diffusion, published in August 2022 — technically rooted in the academic work of the CompVis group at LMU Munich, with Stability AI supplying compute and claiming the stage — put image generation on everyone's computer. Stability itself became the republic's cautionary tale: founder Emad Mostaque's exaggerations, from his résumé to his claimed partnerships, were laid out in a June 2023 Forbes investigation; the company was burning $8 million a month as its coffers emptied, and Mostaque resigned in March 2024 with the words "you're not going to beat centralized AI with more centralized AI." In Getty Images' suit against Stability, London's High Court ruled in November 2025 that model weights are not a "copy" under British law, largely rejecting Getty's copyright claims (the separate U.S. case continues) — a critical precedent under the legal ground of the open-model ecosystem.

On the republic's corporate wing stands the company of the Transformer paper's youngest author: Cohere, founded in Toronto in 2019 by Aidan Gomez — who joined the paper at 20, as an intern — with Nick Frosst and Ivan Zhang. Cohere positioned itself as a deliberate antithesis: a company that sells models to enterprises cloud-agnostically — deployable on the customer's own infrastructure — without entering the consumer chat race and without the word "AGI" in its marketing. The cost of the unglamorous strategy is visibility; its reward is independence: Cohere is one of the few near-frontier labs standing without being any giant's satellite, and the institutional carrier of Canada's national AI ambitions.

On the republic's European wing stands Mistral AI: founded in Paris in April 2023 by DeepMind alumnus Arthur Mensch with Guillaume Lample and Timothée Lacroix of Meta's Llama team, it fused the claim of "Europe's model" with open-weight releases. Valued at €11.7 billion in September 2025 in a round led by the lithography monopoly ASML's €1.3 billion investment, Mistral is the emblem of Europe's search for technological sovereignty (talks for a new round at roughly €20 billion surfaced in June 2026; unconfirmed). The laboratory of transformation in the creative industries has been Runway: founded in 2018 out of NYU's art-and-technology program by Cristóbal Valenzuela and his co-founders, it built video models with and for filmmakers and designers from day one — Runway was also among the partners in Stable Diffusion's academic lineage. Its Gen-series models turned into deals with Hollywood studios, and in February 2026 the company raised at a $5.3 billion valuation; both theses — "AI will kill cinema" and "AI will give independent filmmakers studio power" — are today being tested mostly on Runway's tools. Along the republic's borderlands live hybrid figures: David Holz's Midjourney, which reached hundreds of millions in revenue without taking investment and was sued in June 2025 by Disney and Universal as "a bottomless pit of plagiarism"; Perplexity, the search-reinvention play valued at $20 billion in 2025, which made a $34.5 billion offer for Google Chrome — larger than its own valuation, and read by all as a public-relations maneuver; and Character.AI, the pioneer of personality-based chat — the company that, after the lawsuit over the suicide of 14-year-old Sewell Setzer, closed open-ended chat to under-18s and settled with the family in January 2026. This last example is the republic's heaviest question: when openness and accessibility touch the vulnerable, whose responsibility is it?

And within the republic an honest argument continues: the label "open source" is usually a misnomer. Models like Llama, Qwen, and Grok-1 are open-weight — downloadable, runnable, fine-tunable. But the training data is secret, the training code mostly unpublished, and the licenses carry use restrictions. By the Open Source Initiative's classic definition these are not open source; the word "open" lives in the gray zone between marketing and philosophy. Still, the eighth research question receives a strong answer from this republic: the real force that lowers prices, brakes monopolization, and gives academia and smaller countries access to frontier technology is not the goodwill of the five great labs — it is the existence of this sprawling, quarrelsome, half-open ecosystem.

Chapter 11: China, Europe, and the Rest of the World

On Monday, January 27, 2025, NVIDIA's stock fell 17 percent at the U.S. open; nearly $600 billion of market value was erased in a single day — the largest one-day loss of corporate value in American stock-market history. The trigger was a model released a week earlier from Hangzhou: DeepSeek R1. A company that figured in no one's big-five narrative — the subsidiary of a hedge fund — had published a reasoning model comparable to OpenAI's o1, under an MIT license, with open weights, at a fraction of the API price. It is the most concrete proof, in 2025, of the sentence "the world is bigger than five": the move that changed the game came not from the five players on the stage, but from someone assumed to be a spectator.

The details of the DeepSeek story break every easy narrative. Founder Liang Wenfeng is an engineer out of quantitative finance; his High-Flyer fund had stockpiled around ten thousand NVIDIA A100s before export controls hit. The famous "$5.6 million training cost" in the V3 technical report is — as the report itself states — only the GPU cost of the final training run; SemiAnalysis estimated the company's total hardware investment at around $1.6 billion. So the "ChatGPT for five million dollars" headlines were wrong — but even the corrected figure showed frontier models could be trained far more efficiently than assumed, and that is what caused the panic. The sequel complicates the story further: the expected R2 model never shipped — according to Reuters-sourced reporting, partly because an attempt to train on Huawei Ascend chips, at Beijing's urging, failed — and the company's next great move came only in April 2026, with V4. China's chip deficit is real: Huawei's Ascend 910C runs, by DeepSeek engineers' own assessment, at roughly 60 percent of an NVIDIA H100, and the true bottleneck is high-bandwidth memory. But the Chinese ecosystem's answer is diversity: Alibaba's Qwen family took the global lead in open-model downloads by late 2025; Moonshot AI's trillion-parameter Kimi K2 shipped with open weights; Baidu — after years of its founder belittling open source — opened the Ernie family under an Apache license in June 2025; and Zhipu and MiniMax listed on the Hong Kong exchange in January 2026, becoming the first foundation-model companies anywhere to go public — beating OpenAI and Anthropic to the milestone. Kai-Fu Lee's 01.AI, meanwhile, abandoned its own pretraining to customize DeepSeek's models for enterprises; Lee's sentence is the epigraph of the era: "The biggest nightmare for Sam Altman is that his competitor is free."

A less-discussed dimension of the China story is the strategic use of openness itself. While American labs maximize revenue behind closed APIs, the Chinese players' release of their best models as open weights is not just engineering culture but geopolitical reasoning — you give weight to the markets you cannot sell APIs into (sanctioned, distrustful, or poor), and the world's AI infrastructure gets built on derivatives of your models. At the same time, those same models carry censorship behaviors baked in at training time on topics running from Tiananmen to Taiwan; open weights do not mean open speech. The sum of these two facts globalizes the problem of value-laden infrastructure: which model family a Nigerian startup, an Indonesian ministry, or a Turkish bank builds on is no longer a merely technical choice — it is a quiet act of alignment.

Washington's answer was an architecture of control — and that architecture, too, was bent by politics. Chip export restrictions layered up from 2022; in 2025 the Trump administration rescinded Biden's AI executive order and issued a speed-and-competition-focused "AI Action Plan" in July 2025; NVIDIA's H20 sales to China were first halted, then — in an unprecedented arrangement — reopened in exchange for the company paying the U.S. Treasury 15 percent of its China chip revenue. Export control had mutated from security policy into a revenue-sharing trade instrument.

Europe stepped onto the stage with a different claim: regulatory power. The EU AI Act, in force since August 2024, was the world's first comprehensive AI law — risk-based classification, a list of banned applications, transparency and safety obligations for general-purpose models. But as this article is written, Europe's story is current less for the claim than for the retreat: the "digital omnibus" package proposed by the Commission in November 2025 deferred the high-risk obligations to December 2027; Parliament and Council ratified the delay in June 2026, days before full application would have begun. Competitiveness anxiety — and open pressure from the U.S. administration — had bent Brussels's arm. The Paris AI Action Summit of February 2025 was the photograph of the new climate: 58 countries signed the joint declaration; the United States and Britain refused; U.S. Vice President JD Vance warned from the podium that "excessive regulation could kill a transformative industry." The summit series that had discussed "existential risk" at Bletchley in 2023 had evolved, within two years, into investment-and-competitiveness summits; Britain even swapped the word "safety" out of its institute's name for "security."

And the rest of the world? One of 2024-2026's most striking concepts was "sovereign AI" — the thesis Jensen Huang repeated in every capital he visited: every country should have its own model, on its own infrastructure, trained on its own data. The concept is, of course, a sales pitch in the mouth of a chip vendor; but the anxiety it addresses is real: language models carry the worldview of the data they are trained on, and a country that wires its critical infrastructure to foreign APIs has ceded a piece of its digital sovereignty. The Gulf answered that anxiety with capital: the UAE's G42 built a strategic partnership with Microsoft, Abu Dhabi's MGX fund became a founding partner of Stargate, the Qatar Investment Authority joined xAI's round; the petro-economies converted their energy and capital surpluses into computing power, making themselves indispensable financiers of the AI value chain. India works its position as the "superpower of the application layer," with its vast talent pool and its digital-public-infrastructure experience (Aadhaar/UPI); Japan and South Korea couple their pivotal positions in the memory and chip supply chain to national model programs; African countries are in the story both as the geography of data labor and as the geography of disconnection risk. For mid-sized powers like Türkiye, the lesson emerges from this chapter's sum: the race may look compressed into a U.S.-China binary, but the rules of the game — open models, efficiency leaps, regulatory gaps, sovereign funds — are being rewritten continuously, and every rewrite opens a door to new players. What DeepSeek proved is not only China's strength; it is that a focused, efficient, open move from outside the center can shake the entire board.

Chapter 12: Safety, Ethics, and the Apocalypse Narratives

The AI safety debate is not one debate; it is two hostile debates sharing a single stage.

The roots of the first lie in the field's outer districts. In the 2000s, the circle led by Eliezer Yudkowsky (later MIRI) was arguing the "friendly AI" problem on internet forums: a superintelligence whose goals were misaligned with ours could produce catastrophe without any malice at all. That line of thought gained intellectual respectability with Oxford philosopher Nick Bostrom's 2014 book Superintelligence — publicly praised by Musk, Gates, and Altman, which made "existential risk" (x-risk) a mainstream Silicon Valley anxiety. The growing effective altruism (EA) movement carried money and people to the cause in the same years: Open Philanthropy made a $30 million grant to OpenAI in 2017; EA circles became the labs' chief recruiting pool for safety teams; FTX's collapse burned the movement's money and reputation, but its institutional traces remained. The technical legacy of this tradition should not be belittled: the RLHF method underneath every chat model today came out of the alignment research of Paul Christiano, Jan Leike, and their teams. The doomer tendency's most visible artifact arrived in April 2025: the AI 2027 scenario, in which the team of former OpenAI researcher and whistleblower Daniel Kokotajlo choreographed, hour by hour, a loss-of-control trajectory for 2027 — read by millions, and hotly debated inside the field as either forecast or science fiction.

The second debate concerns the harms of the present, and its protagonists are mostly women — itself a datum about the narrative's hierarchy of visibility. In December 2020, Google forced out Timnit Gebru, co-lead of its Ethical AI team, over a paper examining the risks of large language models — bias, environmental cost, the illusion of meaning; a few months later her co-lead Margaret Mitchell was fired too. Published in March 2021 with the linguist Emily Bender as co-author, the paper lodged its title metaphor permanently in the field: "Stochastic Parrots." The thesis: these models do not understand; they probabilistically remix patterns in their training data — and marketing them as if they understand renders real, measurable harms invisible. Gebru founded the independent DAIR institute; Meredith Whittaker, an organizer of the Google walkout, became president of Signal and one of the most systematic critics of the surveillance economy; Kate Crawford mapped AI in Atlas of AI as an extractive industry running from lithium mines to click-workers; the cognitive scientists Gary Marcus and Melanie Mitchell kept up the technical case against LLM reasoning claims. This camp's objection to the apocalypse narrative is blunt: existential-risk discourse shifts attention and resources from today's documented harms — discriminatory algorithms, labor exploitation, copyright violation, disinformation, surveillance — to a hypothetical future monster; and it does so under the spokesmanship of the very companies claiming to build that monster.

Between the two camps, a third arena grew — the place where the debate touches daily life: the documented harms of the present. Synthetic media became a standard weapon of election seasons — in the global election year of 2024, cases running from voice cloning to fabricated video kept regulators chasing from behind; non-consensual deepfake pornography targeting women grew into the technology's most widespread and least-discussed abuse. On the labor front, the 2023 Hollywood writers' and actors' strikes won history's first major "AI collective-bargaining" provisions, limiting studios' use of AI in script production and digital replicas — early proof that organized labor could negotiate with the technology. In professions from education to journalism, translation to graphic design, the transformation proceeds without a contract; Amodei's warning about "half of entry-level white-collar jobs" remains, as this article is written, a forecast — but the price erosion in translation and graphic-design markets is already measurable. None of these harms is an apocalypse; all of them are the present — and that is the critical camp's core argument: waiting for the end of the world, we are missing what is happening to it now.

After 2023, the two debates became personified in the schism of the founding fathers. Hinton left Google in May 2023 so that he could speak freely; he had become convinced that digital intelligence learns faster than biological intelligence, and ever since — Nobel podium included — he has been describing the possibility that humanity ceases to be the apex intelligence. Bengio turned his course entirely toward safety at the summit of his academic career: he chaired the International AI Safety Report commissioned after Bletchley (first full edition January 2025, with contributions from about a hundred experts across 30 countries), and in June 2025 founded LawZero, a nonprofit laboratory building non-agentic "Scientist AI." LeCun planted himself at the opposite pole: current LLMs, he argues, understand less than a house cat; the doom scenarios are unscientific panic production that feeds authoritarian regulation and monopoly. The single-sentence statement signed by hundreds of industry leaders in May 2023 — "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war" — carried Hinton's and Bengio's signatures; it did not carry LeCun's. That three Turing laureates who defended the same cause for forty years should split this way is the most honest indicator of how open the question is.

The tenth research question — is the safety debate sincere, or the construction of entry barriers through regulation? — again deserves the answer "both," but with evidence. For sincerity: a significant share of lab researchers documented their concerns at real career cost, through resignations, whistleblowing, and open letters; alignment research produces real scientific content. For suspicion: Altman personally proposing licensing before the U.S. Senate in May 2023 was a nearly unprecedented "regulate us" performance, and critics read it as regulatory capture, since any license regime keyed to scale thresholds protects incumbents; in the fight over California's SB 1047, the labs' positions aligned with their commercial interests to a striking degree (the bill was vetoed in 2024; the narrower transparency law SB 53 was signed in September 2025). And meanwhile, the invoice for present harms began arriving through the courts: Anthropic's $1.5 billion book settlement; the rulings in early 2026 in the New York Times case allowing 20 million ChatGPT logs into discovery; Getty-Stability, Disney-Midjourney, the Character.AI suits. The apocalypse has not come; the law has.

Chapter 13: The Psychology of the Founders

Institutions scale the character of their founders. This chapter is a portrait gallery — without canonization or demonization, drawn from the record.

Sam Altman is the strategist of the capital age. St. Louis-born, Stanford dropout, founder of his first company at 19, then Paul Graham's successor at Y Combinator, where he presided over the births of hundreds of startups. His genius is not technical but topological: he sees before anyone else who needs what, which narrative attracts which capital, which door opens onto which corridor. Paul Graham once distilled the trait into a line that has trailed Altman ever since: "You could parachute him into an island full of cannibals and come back in five years and he'd be the king." His return, within five days, to the company that had fired him — carried by the signatures of ninety percent of its employees — was a show of force without parallel in corporate history, and at the same time a warning: personal indispensability is the inverse of institutional oversight. Read alongside former board member Helen Toner's later public account — the claim that Altman had fed the board inaccurate information — and WilmerHale's "breakdown in trust" formula, the portrait sharpens: Altman is the type of founder who governs through relationships rather than rules, locked on the objective, charismatic; the type that grows companies fast and wears out control mechanisms.

Demis Hassabis is the scientist-king. He is the only figure to have fused chess, game design, neuroscience, and corporate command in a single career, and he always narrated DeepMind through the metaphor of an Apollo program. The Nobel medal was the final ratification of the science-first claim; but the same man commands Google's product war, and the tension of that duality — the gap between Gemini's launch calendar and Nature's — is the central question of his biography. Dario Amodei is the safety technocrat: physics doctorate, speech that moves like a paper's footnotes, narrating both apocalypse and abundance in the same measured idiom. His critics point to the contradiction between his warnings and his product schedule; his defenders reply that he is the builder among the pessimists and the most honest speaker among the builders. Elon Musk is the paradox itself: the same biography contains DeepMind's early investor, OpenAI's founding donor, the six-month-pause signatory, the owner of the lab whose Grok chatbot spiraled into antisemitic tirades, and the builder of one of the world's largest private compute clusters. His consistency lies not in ideas but in position: in every game, he rejects any board he does not own. The other constant is a record that has punished those who wrote him off — PayPal, Tesla, SpaceX, Starlink, each dismissed as impossible before it hardened into infrastructure — which is why Peter Thiel, his old PayPal partner, compressed two decades of observation into a single sentence: "Never bet against Elon Musk." xAI, founded years after its rivals, stood up one of the world's largest training clusters in a matter of months; whatever one makes of Grok, betting against the build has once again proved unwise.

Mark Zuckerberg is the platform emperor; he runs AI not as a product but as the empire's line of defense. Quietly pivoting to open models when the metaverse bet became a punchline, then rebuilding his lab in a single summer with a $14 billion move when Llama stumbled — that is his real talent: ideologylessness. Openness, for him, is not a principle but the most rational weapon of the current quarter. Ilya Sutskever is the field's mystic: present at every pivotal technical moment from AlexNet to GPT, the first to insist that "predicting the next word" might require understanding the world, the man credited inside OpenAI with the mantra feel the AGI. His role in the 2023 coup and his remorse afterward form the most dramatic case of technical conscience trapped inside corporate power; today, at SSI — productless and valued at $32 billion — he works directly on superintelligence: the most expensive hermit project in the history of modern technology. Yann LeCun is the scientific dissident, defending his world-models program against the mainstream LLM consensus with a startup founded as he nears seventy. Hinton and Bengio embody the ethical turn of the founding fathers — one converted the Nobel podium into a warning pulpit, the other his career into safety-institution building. Fei-Fei Li is the most credible voice of the human-centered wing: founding co-director of Stanford HAI, advocate of the thesis that AI exists to extend human capability, and — with her World Labs venture on spatial intelligence — a figure who breaks the male-founder mold of the narrative. Jensen Huang is the emperor of the hardware age: thirty years at one company, twice back from the brink of bankruptcy, fusing the humility of a founder accustomed to losing with a five-trillion-dollar monopoly. He saw, long before the labs did, that the future would be computed in parallel — and then spent two decades building CUDA, the software moat that quietly turned a maker of graphics cards into the mandatory supplier of intelligence itself. In the war of the AI empires he is the rare figure who need not pick a side: whichever lab wins the model race, it wins on his silicon, and the deeper the rivalry the more he sells to all of them. It is this article's thesis rendered in hardware — real power sits not where the activity is loudest but at the bottleneck everyone must pass through, and in the age of AI that bottleneck is a chip. Alexandr Wang is the prototype of the new type: not a researcher but a broker of the data-labor-state triangle; a figure who signed Pentagon contracts in his twenties and then took over the superintelligence program of the world's largest social platform — proof that AI power can now be produced from supply-chain dominance as well as laboratory genius.

At the gallery's edge, two transitional figures deserve their own frame. Mustafa Suleyman traced the field's most improbable orbit: he began as DeepMind's "conscience," exited under the shadow of management-style complaints, built Inflection AI with Reid Hoffman into a multibillion-dollar startup — which Microsoft then hollowed out in an unprecedented acquisition-without-acquisition — and found himself CEO of Microsoft AI and, by late 2025, head of the company's "humanist superintelligence" team. The author of The Coming Wave — the book arguing the technology must be contained — today runs the superintelligence program of one of the world's largest corporations; after Musk, that trajectory is the clearest specimen of how warning and building interleave in a single career. Mira Murati is the portrait of quiet power: the executive who held the institution together on crisis nights, never sought the spotlight, and — upon leaving — raised one of history's largest seed rounds alone, before having a product. Thinking Machines' difficulties in 2026 exposed one of the field's brutal truths: reputation attracts capital; but in the frontier-model race, capital is only the ticket at the door.

A pattern emerges from this gallery: the field is governed by a handful of people who are extraordinarily talented, extraordinarily self-assured, and extraordinarily resistant to oversight. The problem lies not in the characters but in the architecture — the governance of a technology with civilizational claims is still entrusted to personal virtue.

Chapter 14: Who Will Own AGI?

AGI — artificial general intelligence — has no agreed definition; that is not a detail but the center of the story. OpenAI's charter defines it as "highly autonomous systems that outperform humans at most economically valuable work" — an economic definition, not a cognitive one. The famous "AGI clause" in the Microsoft-OpenAI contract turned the definition into a legal trigger: if OpenAI's board declared AGI achieved, Microsoft's access to the new technology would end. The October 2025 agreement transferred that authority to an independent expert panel. It is worth pausing on: the most ambitious technological threshold in human history is now the arbitration mechanism of a contract between two companies. The word "intelligence" is itself an instrument of this economy: for capital markets it is a nearly perfect word — it seems unmeasurable yet can be poured into benchmarks; it sounds sacred yet can be priced as an API; it presents as universal yet can be locked up in cloud contracts. The answer to the ninth research question lies here too — AGI is four things at once: a scientific goal that functions as the field's horizon line; an investor story that justifies trillion-dollar valuations with the promise of "the machine that automates everything"; a quasi-religious narrative that speaks in the vocabulary of salvation and apocalypse; and a geopolitical lever running from export controls to summit diplomacy. The word's power comes precisely from its serving all four functions simultaneously.

The ownership question is being played out on four planes. On the corporate plane the picture is clear and unprecedented: frontier models belong to a handful of institutions that are not even public companies; OpenAI at $852 billion and Anthropic at $965 billion live in private markets, where shareholder oversight is weakest. The November 2023 crisis was a stress test of these institutions' internal-control mechanisms, and the result was not reassuring. On the state plane, AI has entered the inventory of national power: the United States plays with chip controls and federal purchasing power; China with its state-private hybrid; the Gulf with capital; Europe with rule-making — and the lesson of 2025-26 is that, increasingly, it is not states dictating terms to companies but companies dictating terms to states. On the infrastructure plane, the AGI question becomes concrete: who can access the gigawatts, the high-bandwidth memory, the advanced lithography? On this plane AGI is not software but a matter of energy and real estate — and the answer is written in place names like Abilene, Memphis, and Hangzhou. On the societal plane stands the weakest link: the public that produced the technology's training data (the entire internet), labored its alignment (the data workers), and carries its risk (everyone whose work is being transformed) has no seat at the table.

Scene: the G7 summit at Évian, June 17, 2026 — the "AI and the Digital Age" working lunch, where G7 leaders and the technology executives from eight countries sat at one table.

There is now a single photograph in which all four planes converge. On June 17, 2026, at the "AI and the Digital Age" working lunch of the G7 summit in Évian-les-Bains, the leaders of the world's most powerful states shared the table with a dozen technology executives from eight countries. According to the official participant record, seated alongside Altman, Hassabis, Amodei, and — representing Meta — Alexandr Wang was the world beyond the five: Mistral's Mensch for France, Cohere's Gomez for Canada, Black Forest Labs founder Robin Rombach — one of the academic architects of Stable Diffusion — for Germany, Sarvam AI for India, Sakana AI for Japan, Domyn for Italy, Synthesia for Britain. Axios's headline captured the scene: AI CEOs at the G7 "as heads of nation-states." Behind closed doors, Amodei and Hassabis called for a U.S.-led coalition on chips and frontier-model access that would exclude China; Altman proposed an impartial international forum to set global testing standards; and the European leaders at the same table sought checks on American dominance. In a single lunch: the elevation of companies to state protocol, the conversion of AGI discourse into geopolitical leverage, the institutionalization of "sovereign AI" through national champions — and, beside the five great labs, the chairs of a new generation stretching from Bengaluru to Tokyo. Even the seating plan now stated this article's thesis.

Those looking for instruments of oversight do not, however, hold an empty bag. Compute is a governable bottleneck: frontier training happens in countable facilities, with traceable chip shipments — threshold-based notification duties (the logic of California's SB 53 and of the AI Act) lean on that visibility. The third-party evaluation ecosystem — safety institutes, red-teaming firms, academic auditors — is embryonic but growing. Whistleblower protections entered the agenda seriously for the first time after Kokotajlo and colleagues took OpenAI's equity-pressure tactics public on their way out. And tort law — as the copyright cases show — writes de facto standards of conduct wherever regulation cannot keep up. None of these, alone, can bridle trillion-dollar momentum; together, they show that "ungovernable" is a choice, not a fate.

So: is AGI a product, infrastructure, a weapon, or a public good? The honest answer: today it is in practice a product, rapidly becoming infrastructure, partially weaponizing, and almost nowhere a public good. Open-weight models are a counterweight in this picture that should not be underestimated — no strategic technology in history has been this copyable this early, this widely. But an open model is not open governance. The architecture of public oversight built over a century for electric grids, telecom networks, and the pharmaceutical industry does not yet exist for artificial intelligence; the attempts that do exist — safety institutes, the AI Act, transparency statutes — were weakened in the 2025-26 conjuncture before they were even established. The present answer to the question of who will own AGI is the de facto state of affairs no one declares and everyone knows: AGI will belong to those who build it — until societies organize otherwise.

Conclusion: The Board Belongs to All of Us

This article began with a hotel dinner, two graphics cards in a bedroom, and a thirty-seventh move on a Go board. Let it close there too. After AlphaGo's victory, Lee Sedol retired from professional Go within a few years; he said he had come to question the game's meaning "with an entity that cannot be defeated." But in those same years something unexpected happened as well: a new generation of human players, training with the machines, tore up and rebuilt the game's five-hundred-year-old orthodoxies, and human Go entered one of the most creative periods in its history. The machine did not end the game; it enlarged the board.

The founding story of artificial intelligence, as this article has tried to show, is too large to be told as the competition of five companies. In that story there are academics who carried an idea ridiculed in Toronto for forty years; an immigrant scientist who built a dataset judged "devoid of ideas" at Princeton; an engineer who realized, while selling chips to gamers, that he was laying civilization's infrastructure; workers in Nairobi whose own minds were harmed while making other people's machines harmless; a team in the basement of a Hangzhou hedge fund that shook Wall Street; laws written and diluted in Brussels; and yes — the dinner table at the Rosewood, the five days of November 2023, the courtrooms, and the trillion-dollar rounds. None of these is a footnote to any other. The history of AI is the newest stage of humanity's struggle over knowledge, power, and control — and on that stage there is no such role as spectator.

This breadth is no accident; it follows from the technology's nature. Just as electricity could never be a single company's product — the ones who invented the generator, built the grid, wrote the standards, organized the unions, and regulated the prices each wrote their own history — a general-purpose technology of intelligence cannot fit inside a single corporate narrative. The founders of the five great labs know this better than anyone; each is trying to grow his empire vertically, to control the layers his story depends on — the chips, the electricity, the data, the law, the public. But the layers resist: courts price the copyright, towns sue the turbines, workers testify, competitors melt margins with open weights, and voters and parliaments — slowly, messily, usually late — write rules. History is being written not where the founders planned, but in the sum of these frictions.

The founders have told us their visions; we listened. The models have shown us their capabilities; we saw. What remains is the least-written chapter of the story: the social and institutional labor that will decide who governs this technology, under what rules, accountable to whom. Move 37 was beautiful — but the real wisdom of Go is this: no single move wins the game; the player who sees the whole board does. And the board is bigger than five. It is as big as the world.

This essay has a companion, written from the field rather than the stands: The Five AI Empires, and the World Around Them: A Field Guide to Minimum Viable Sovereignty — what a builder outside the five great labs does about everything described here.

APPENDICES

Appendix 1: Chronology

Year	Event
1956	Dartmouth Conference; the term "artificial intelligence" (McCarthy)
1958	Rosenblatt's Perceptron
1969	Minsky & Papert, Perceptrons; neural-network funding dries up
1973	Lighthill Report; first "AI winter"
1986	Rumelhart, Hinton & Williams: backpropagation (Nature)
1989	LeCun applies convolutional networks to handwriting (Bell Labs)
1993	NVIDIA founded (Huang, Malachowsky, Priem)
2006	CUDA launches; Hinton's "deep belief nets"; ImageNet project begins
2009	ImageNet published (14M+ labeled images)
2010	DeepMind founded (Hassabis, Legg, Suleyman)
2011	Google Brain founded (Ng, Dean, Corrado)
Oct 2012	AlexNet wins ImageNet at 15.3% error — the deep learning rupture
2013	Google acquires DNNresearch (Hinton's team); Facebook founds FAIR (LeCun)
Jan 2014	Google acquires DeepMind (~$400-650M; sources conflict)
Dec 2015	OpenAI founded — nonprofit, $1B pledged
Mar 2016	AlphaGo defeats Lee Sedol 4-1; Move 37
2016	Scale AI (Wang) and Hugging Face founded
Jun 2017	"Attention Is All You Need" — the Transformer paper
Feb 2018	Musk leaves the OpenAI board
Jun 2018	GPT-1; Nov 2018-2020: AlphaFold cracks protein folding at CASP
Mar 2019	OpenAI adopts "capped-profit" structure; 2018 Turing Award to Hinton/LeCun/Bengio
Jul 2019	Microsoft invests $1B in OpenAI
May 2020	GPT-3 (175B parameters)
Dec 2020	Timnit Gebru forced out of Google
Jan-Feb 2021	Anthropic founded (the Amodeis + 5 OpenAI colleagues); DALL-E
Aug 2022	Stable Diffusion released openly
Nov 30, 2022	ChatGPT launches; 100M users in 2 months
Jan 2023	Microsoft's ~$10B investment; TIME's Kenya data-worker investigation
Feb-Mar 2023	LLaMA leak; GPT-4 (closed technical report); xAI incorporated; the "6-month pause" letter
Apr 2023	Google Brain + DeepMind merge; Mistral AI founded
May 2023	Hinton leaves Google; Altman tells the Senate "license us"
Jul 2023	xAI launches; Llama 2 opens under a commercial license
Nov 2023	OpenAI board crisis: Altman fired and back in 5 days; Bletchley Summit; Grok launches
Mar 2024	Anthropic Claude 3; Suleyman becomes Microsoft AI CEO; Grok-1 open-weights
May 2024	GPT-4o; Sutskever and Leike leave OpenAI; AlphaFold 3
Aug 2024	Google-Character.AI deal ($2.7B; Shazeer returns); EU AI Act in force
Sep-Oct 2024	Murati departs; SSI raises $1B; Nobels: Hinton (Physics), Hassabis & Jumper (Chemistry); SB 1047 veto
Jan 2025	DeepSeek R1; NVIDIA loses ~$600B in one day; Stargate announced ($500B)
Feb 2025	Musk's $97.4B bid rejected; Paris Summit (US/UK refuse to sign); Thinking Machines Lab unveiled
Apr 2025	Llama 4 and the benchmark controversy; AI 2027 report; SSI at $32B
Jun 2025	Meta-Scale AI deal ($14.3B); Meta Superintelligence Labs; Bengio's LawZero; Baidu opens Ernie
Jul 2025	Grok "MechaHitler" scandal; Grok 4; US "AI Action Plan"; Kimi K2; DeepMind's official IMO gold
Aug-Sep 2025	GPT-5 and the backlash; Anthropic at $183B; Anthropic's $1.5B copyright settlement; ASML-Mistral (€1.3B); SB 53 signed
Oct 2025	OpenAI at $500B; OpenAI restructuring completes (PBC; Microsoft ~27%); Anthropic-Google 1M-TPU deal; NVIDIA passes $5T; Qwen passes Llama in downloads
Nov 2025	Gemini 3; LeCun leaves Meta (AMI Labs); EU "digital omnibus" delay proposed; NVIDIA-Anthropic investment
Jan 2026	xAI raises $20B (~$230B valuation); Zhipu & MiniMax IPO in Hong Kong; Character.AI settlements; Amodei's "Adolescence of Technology"
Feb-Mar 2026	ChatGPT at 900M weekly users; OpenAI at $852B ($122B round); SpaceX-xAI merger reports ($1.25T; secondary sources); Anthropic at $380B
Apr-May 2026	DeepSeek V4; Musk v. OpenAI: jury rules against Musk, case dismissed (May 18); Anthropic at $965B; Karpathy joins Anthropic
Jun 2026	EU finalizes the AI Act delay; Shazeer moves to OpenAI; Mistral ~€20B round rumored; G7 Évian summit: 12 AI executives from 8 countries at the leaders' table (Jun 17)

Appendix 2: Map of People

Founders and CEOs

Sam Altman — CEO, OpenAI. Former YC president. Contribution: capital formation, productization, policy diplomacy. Position: accelerationist-pragmatist; "beneficial AGI for everyone."
Demis Hassabis — CEO, Google DeepMind; DeepMind co-founder. Nobel in Chemistry, 2024. Contribution: the RL + deep learning program, the AlphaGo/AlphaFold vision. Position: science-first, cautious optimist.
Dario Amodei — CEO, Anthropic; former OpenAI VP of research. Contribution: scaling laws, GPT-2/3 leadership, safety-centered company design. Position: "safety from inside the race"; public risk warner.
Daniela Amodei — President, Anthropic; architect of operations and culture.
Elon Musk — Founder, xAI; OpenAI founding donor; Tesla/SpaceX/X. Position: paradoxical — existential-risk prophet and aggressive competitor at once.
Mark Zuckerberg — CEO, Meta. Contribution: FAIR, PyTorch, Llama, the open-weight strategy, MSL. Position: ideology-free platform defense.
Greg Brockman — President, OpenAI; infrastructure and engineering leadership; Altman's most loyal ally.
Ilya Sutskever — AlexNet co-author, OpenAI founding chief scientist, SSI founder. Position: the technical mystic; "superintelligence safety first."
Mustafa Suleyman — DeepMind co-founder, Inflection founder, Microsoft AI CEO. Position: the applied/political wing; "humanist superintelligence."
Jensen Huang — Founding CEO, NVIDIA. Contribution: the GPU/CUDA empire. Position: supplier-hegemon; distant from risk debates.
Alexandr Wang — Founder, Scale AI; Meta Chief AI Officer. Contribution: the data-labeling industry, the defense-AI bridge. Position: America-first competitor.
Liang Wenfeng — Founder, DeepSeek; High-Flyer fund. Contribution: efficiency-driven open frontier models.
Arthur Mensch / Guillaume Lample / Timothée Lacroix — Mistral founders; Europe's open-model claim.
Clément Delangue / Julien Chaumond / Thomas Wolf — Hugging Face founders; the open ecosystem's platform.
Aravind Srinivas — Perplexity founder; the AI-search claim.
David Holz — Midjourney founder; the independent, investor-free model.
Emad Mostaque — Stability AI founder (resigned 2024); at once hero and cautionary tale of the open image model.
Noam Shazeer — Transformer co-author; Character.AI founder; Google→OpenAI (2026). The most itinerant architect of LLM culture.
Mira Murati — Former OpenAI CTO; manager of the ChatGPT/DALL-E productization; founder of Thinking Machines Lab.

Scientists

Geoffrey Hinton — Backpropagation; deep learning's keeper through the winters; Turing 2018, Nobel in Physics 2024. Position: chief risk warner since 2023.
Yann LeCun — CNN inventor; FAIR founder (departed 2025, AMI Labs); Turing 2018. Position: LLM skeptic, open-source advocate, anti-doomer.
Yoshua Bengio — The Montreal school/Mila; Turing 2018. Position: leader of safety institutionalization (International Report, LawZero).
Fei-Fei Li — Creator of ImageNet; Stanford HAI; World Labs. Position: human-centered AI.
Alex Krizhevsky — Chief engineer of AlexNet; the revolution's least public figure.
Andrej Karpathy — OpenAI founding member, Tesla Autopilot director, educator-narrator; at Anthropic from 2026.
John Schulman — PPO/RLHF; among ChatGPT's technical fathers; OpenAI→Anthropic→Thinking Machines.
Jared Kaplan — Lead author of the scaling laws; Anthropic founding scientist.
Chris Olah — Pioneer of interpretability research; Anthropic co-founder.
Jan Leike — Alignment researcher; protest resignation from OpenAI (2024), Anthropic alignment lead.
Andrew Ng — Google Brain founder, Coursera, Baidu; mass educator of AI.
Jeff Dean — Google's infrastructure legend; Brain co-founder; Google chief scientist.
Ian Goodfellow — Inventor of GANs; forerunner of generative modeling.
Ashish Vaswani et al. — The Transformer's eight authors; all left Google.
David Silver — The RL brain of AlphaGo/AlphaZero. Oriol Vinyals, Koray Kavukcuoglu — DeepMind technical leadership. John Jumper — AlphaFold; Nobel 2024.

Critics and watchdogs

Timnit Gebru — Former Google Ethical AI co-lead; DAIR founder; central figure of the data/bias critique.
Emily Bender — "Stochastic Parrots"; the critique of the illusion of meaning.
Margaret Mitchell — The second ethicist fired from Google; Hugging Face chief ethics scientist.
Meredith Whittaker — Signal president; critic of the surveillance economy.
Kate Crawford — Atlas of AI; the material/labor map of AI.
Gary Marcus, Melanie Mitchell — Scientific skeptics of LLM capability claims.
Daniel Kokotajlo — OpenAI whistleblower; lead author of AI 2027.

Capital and platform figures

Satya Nadella — Microsoft; architect of the OpenAI alliance. Sundar Pichai — Google/Alphabet. Reid Hoffman — OpenAI founding donor, Inflection co-founder. Peter Thiel — Early backer of DeepMind and OpenAI. Masayoshi Son — SoftBank; Stargate co-owner. Jeff Bezos / Larry Ellison — Financiers of the infrastructure war through Amazon and Oracle.

Appendix 3: Map of Institutions

Institution	Founded	Founders	Key products	Technical claim	Strategic advantage	Main criticisms
OpenAI	2015	Altman, Musk, Brockman, Sutskever et al.	GPT series, ChatGPT, DALL-E, Sora, o-series	Scale + productization first	Distribution (900M+ users), Microsoft, Stargate	Abandoned openness, governance fragility, copyright suits
Google DeepMind	2010 (DM) / 2023 (merged)	Hassabis, Legg, Suleyman	AlphaGo/Fold, Gemini, Veo, Genie	Science + RL depth	Google data, TPUs, distribution	Science/product tension, closing of open publication
Anthropic	2021	The Amodeis + 5 co-founders	Claude series, Claude Code	Safety + interpretability	Enterprise market, dual cloud (AWS+Google)	"Safety as marketing," pirated-data settlement, speed/warning contradiction
xAI	2023	Musk, Babuschkin et al.	Grok series, Colossus	Speed + real-time data	X data, Musk ecosystem, capital	Model safety (MechaHitler), environmental violations, polarization
Meta AI / MSL	2013 (FAIR) / 2025 (MSL)	Zuckerberg, LeCun (FAIR)	Llama, PyTorch, Meta AI assistant	Open-weight ecosystem	3B users, ad engine, capex	Llama 4 benchmark crisis, FAIR's dissolution, instrumental openness
NVIDIA	1993	Huang, Malachowsky, Priem	GPUs, CUDA, DGX	Compute monopoly	CUDA lock-in, supply dominance	Circular deals, monopoly risk
Microsoft	1975	Gates, Allen	Azure, Copilot, MAI models	Platform + OpenAI access	Enterprise distribution	OpenAI dependence/rivalry dilemma
Hugging Face	2016	Delangue, Chaumond, Wolf	Hub, Transformers	Open-ecosystem platform	Community network effects	Download concentration, moderation burden
Mistral AI	2023	Mensch, Lample, Lacroix	Mistral/Le Chat	Efficient open models	EU sovereignty narrative, ASML	Scale gap, squeezed between US/China
DeepSeek	2023	Liang Wenfeng	V3/R1/V4	Efficiency leap	Cost, openness, hedge-fund backing	Cost claims, censorship, chip access
Scale AI	2016	Wang, Guo	Data labeling, defense AI	Data operations	Government ties, Meta partnership	Labor conditions, loss of neutrality
Cohere	2019	Gomez, Frosst, Zhang	Command series	Enterprise LLM	Cloud-agnostic	No consumer visibility
Character.AI	2021	Shazeer, De Freitas	Character chat	Personality-based AI	Young-user engagement	Child-safety lawsuits
Stability AI	2020	Mostaque	Stable Diffusion	Open image model	Brand, community	Founder scandals, financial crisis
SSI	2024	Sutskever, Gross, Levy	(no product)	"One focus: safe superintelligence"	Sutskever's reputation	$32B valuation, no product
Thinking Machines	2025	Murati, Schulman et al.	Tinker	Open-model fine-tuning	OpenAI alumni bench	Founder attrition, valuation uncertainty

Appendix 4: Table of Central Conflicts

Conflict	Parties	Essence	Status (mid-2026)
Open source vs. closed models	Meta/Mistral/DeepSeek/HF vs. OpenAI/Anthropic/Google	Is weight access a safety risk or an antidote to monopoly?	Leadership of the open camp has passed to China; Meta retreating
Safety vs. speed	Anthropic's discourse, the Leike/Sutskever resignations vs. product calendars	"Make it safe first" vs. "you can't set safety standards from behind"	Everyone races; safety internalized into process
Academia vs. industry	Universities vs. the labs	The compute and salary chasm; brain drain	Frontier research now lives in companies; academia is the critique-and-theory base
Science vs. product	The DeepMind/FAIR traditions vs. Gemini/Llama calendars	The death of publication culture	Science risks becoming the product's showroom
Public benefit vs. platform monopoly	Nonprofit structures, PBCs vs. trillion-dollar valuations	Do governance experiments survive capital pressure?	Nov 2023 and the PBC conversion: capital won, the form survived
US vs. China	Chip controls vs. efficiency/openness plays	Hardware blockade vs. algorithmic efficiency	DeepSeek/Qwen: the blockade slows but does not stop
Data rights vs. model training	Authors, artists, media vs. the labs	Is training data fair use or expropriation?	Mixed case law: Anthropic paid $1.5B; Getty lost in the UK; NYT case pending
The AGI ideal vs. present harms	The x-risk camp vs. the ethics/labor camp	Which risk is real; whose agenda comes first?	Both camps institutionalized; regulation trails both

Appendix 5: Bibliography

Note: The bibliography below organizes the sources used in this article by category. Dates and figures for developments from mid-2024 through 2026 were verified via web research conducted in July 2026 against the sources listed.

Official company sources

OpenAI blog: "Introducing OpenAI" (Dec 2015); "OpenAI LP" (Mar 2019); "Announcing The Stargate Project" (Jan 2025); "Built to benefit everyone" (Oct 2025); the GPT/Sora/o-series announcements; the November 2023 crisis statements; the WilmerHale review summary (Mar 2024)
Microsoft blog: "The next chapter of the Microsoft-OpenAI partnership" (Oct 28, 2025); Microsoft AI/MAI announcements
Anthropic: founding and Claude announcements; "Constitutional AI" (arXiv:2212.08073); the interpretability series; Series E-H announcements; the Responsible Scaling Policy
Google/DeepMind: DQN (Nature, 2015), the AlphaGo/AlphaZero/AlphaFold publications; the Google DeepMind merger announcement (Apr 2023); Gemini announcements; the IMO 2024/2025 posts
Meta: FAIR/Llama announcements; Zuckerberg, "Open Source AI Is the Path Forward" (Jul 2024)
xAI: founding and Grok announcements; the July 2025 incident statement
NVIDIA: quarterly results (FY2026); official announcements from Mistral, SoftBank, SSI, DeepSeek, Hugging Face, Stability, Runway, Perplexity
Dario Amodei: "Machines of Loving Grace" (Oct 2024); "The Adolescence of Technology" (Jan 2026)

Academic sources

McCarthy et al., the Dartmouth proposal (1955); Rosenblatt (1958); Minsky & Papert, Perceptrons (1969); Rumelhart, Hinton & Williams (Nature, 1986); LeCun et al. (1989/1998); Hinton et al., deep belief nets (2006); Deng, Li et al., ImageNet (CVPR 2009); Krizhevsky, Sutskever & Hinton, AlexNet (NeurIPS 2012); Vaswani et al., "Attention Is All You Need" (2017); Kaplan et al., "Scaling Laws for Neural Language Models" (2020); Bender, Gebru et al., "On the Dangers of Stochastic Parrots" (FAccT 2021); Jumper et al., AlphaFold2 (Nature, 2021); AlphaFold 3 (Nature, 2024); Bengio et al., International AI Safety Report (2025)
The Nobel Foundation citations (Physics and Chemistry, 2024); the ACM Turing Award citation (2018)

Journalism

Wired: Cade Metz's AlphaGo/Seoul reporting (Mar 2016); the OpenAI crisis files; Steven Levy's Microsoft-OpenAI alliance cover feature
TIME: Billy Perrigo, "OpenAI Used Kenyan Workers on Less Than $2 Per Hour" (Jan 2023)
Forbes: the Emad Mostaque investigation (Jun 2023); Brockman's testimony (May 2026)
Reuters, Bloomberg, CNBC, Financial Times, NYT, The Information, TechCrunch, The Verge, MIT Technology Review, IEEE Spectrum: funding rounds and valuations (OpenAI $157B→$852B; Anthropic $61.5B→$965B; xAI ~$230B), the DeepSeek/NVIDIA market shock (Jan 2025), the Meta-Scale deal (Jun 2025), LeCun's departure (Nov 2025), the Zhipu/MiniMax IPOs (Jan 2026), the Shazeer transfers, the Gemini/Llama/Grok launch coverage
NPR/CNN/Axios: the Musk-OpenAI verdict (May 18, 2026); the Grok incident of July 2025
Platformer: the GPT-5 backlash (Aug 2025); SemiAnalysis: the DeepSeek cost analysis; Epoch AI: Stargate site status
Books: Cade Metz, Genius Makers (2021); Parmy Olson, Supremacy (2024); Karen Hao's OpenAI reporting

Legal and regulatory sources

Musk v. Altman/OpenAI filings (N.D. Cal., 2024-2026) and the May 18, 2026 ruling
Bartz v. Anthropic: Judge Alsup's fair-use ruling (Jun 2025) and the $1.5B settlement documents (final approval in progress as of May-Jun 2026)
NYT v. OpenAI/Microsoft (S.D.N.Y.): the discovery rulings (Nov 2025-Jan 2026)
Getty Images v. Stability AI: the London High Court ruling (Nov 2025); the U.S. case pending
The EU AI Act text and the "digital omnibus" amendments (Nov 2025-Jun 2026); the Bletchley Declaration (2023); the Paris Summit documents (Feb 2025); the U.S. AI Action Plan (Jul 2025); California's SB 1047 veto message (2024) and SB 53 (2025)

Critical literature

Kate Crawford, Atlas of AI (2021); Emily Bender & Alex Hanna, The AI Con (2025); Gary Marcus, Taming Silicon Valley (2024); Melanie Mitchell, Artificial Intelligence: A Guide for Thinking Humans (2019); the DAIR and AI Now institute reports; Kokotajlo et al., "AI 2027" (Apr 2025)

Podcasts and long-form interviews

Lex Fridman (the Altman, Musk, LeCun, Karpathy episodes); Dwarkesh Patel (Amodei, Sutskever, Hassabis); No Priors; a16z; Altman's Senate testimony (May 2023); the Nobel press conferences (Oct 2024)

Selected primary-source links

The Dartmouth proposal (1955): https://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html
Backpropagation (Nature, 1986): https://www.nature.com/articles/323533a0
AlexNet (NeurIPS 2012): https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks
"Attention Is All You Need" (2017): https://arxiv.org/abs/1706.03762
Scaling laws (2020): https://arxiv.org/abs/2001.08361
Constitutional AI (2022): https://arxiv.org/abs/2212.08073
Stochastic Parrots (FAccT 2021): https://dl.acm.org/doi/10.1145/3442188.3445922
AlphaFold2 (Nature, 2021): https://www.nature.com/articles/s41586-021-03819-2
OpenAI founding announcement (2015): https://openai.com/index/introducing-openai/
OpenAI LP / capped-profit (2019): https://openai.com/index/openai-lp/
ChatGPT announcement (2022): https://openai.com/index/chatgpt/
OpenAI board-crisis announcement (Nov 2023): https://openai.com/index/openai-announces-leadership-transition/
OpenAI PBC conversion (Oct 2025): https://openai.com/index/built-to-benefit-everyone/
Microsoft-OpenAI new agreement (Oct 2025): https://blogs.microsoft.com/blog/2025/10/28/the-next-chapter-of-the-microsoft-openai-partnership/
Stargate announcement (Jan 2025): https://openai.com/index/announcing-the-stargate-project/
Anthropic Series H (May 2026): https://www.anthropic.com/news/series-h
Dario Amodei, "Machines of Loving Grace": https://darioamodei.com/machines-of-loving-grace
Zuckerberg, "Open Source AI Is the Path Forward" (2024): https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/
The DeepMind AlphaGo archive: https://deepmind.google/research/alphago/
The TIME Kenya investigation (2023): https://time.com/6247678/openai-chatgpt-kenya-workers/
The Musk-OpenAI verdict (NPR, May 2026): https://www.npr.org/2026/05/18/nx-s1-5822366/musk-altman-openai-jury-verdict-claims-dismissed
The EU AI Act framework: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
The 2018 Turing Award citation: https://awards.acm.org/about/2018-turing
The G7 Évian AI lunch (CNBC, Jun 2026): https://www.cnbc.com/2026/06/17/g7-trump-ai-tech-leaders-openai-anthropic-google.html
"AI CEOs as heads of nation-states" (Axios, Jun 2026): https://www.axios.com/2026/06/20/ai-tech-moguls-g7
Nobel Prize in Physics 2024: https://www.nobelprize.org/prizes/physics/2024/summary/
Nobel Prize in Chemistry 2024: https://www.nobelprize.org/prizes/chemistry/2024/summary/

Key sources by chapter (selected)

Introduction & Chapter 3: Metz's Wired AlphaGo file; the DeepMind AlphaGo documentary; the Nature DQN and AlphaFold papers; the 2024 Nobel citations
Chapters 1-2: the Turing Award citation; Genius Makers; ImageNet (CVPR 2009); the AlexNet paper; the NVIDIA founding lore (Denny's)
Chapters 4-5: the OpenAI founding announcement; the email exhibits from the Musk litigation; the GPT-4 technical report (the closure statement); the November 2023 crisis coverage (Wired/NYT/Bloomberg); the PBC conversion documents (Oct 2025)
Chapter 6: the Anthropic founding sources; Constitutional AI; the Series F-H announcements; the Bartz settlement; the Amodei essays
Chapter 7: the xAI announcements; the NPR/CNN verdict coverage; the NAACP/SELC Memphis file; the ADL statement (Jul 2025)
Chapter 8: Zuckerberg's open letter; the LMArena statement (Apr 2025); CNBC's Scale AI coverage; LeCun's exit interviews
Chapter 9: NVIDIA's FY2026 results; Bloomberg's circular-deals file; the TIME Kenya investigation; the nuclear PPA announcements
Chapters 10-11: the Hugging Face state-of-the-hub reports; the DeepSeek technical reports + SemiAnalysis; the Xinhua/Interconnects Qwen data; the EU AI Act/omnibus documents; the Paris Summit coverage
Chapter 12: Stochastic Parrots; the CAIS statement (May 2023); the Hinton/Bengio statements of 2025-26; the Senate transcript (May 2023)

This article rests on publicly available sources accessible as of July 2026. For fast-moving data — valuations, user counts, litigation status — the dates given in the text govern. Points where sources conflict (the DeepMind acquisition price, DeepSeek's training cost, the SpaceX-xAI merger valuation, the ChatGPT user metric, the Thinking Machines valuation) are flagged explicitly in the text.

The End of Lazy Money

Mon, 01 Jun 2026 00:00:00 GMT

What happens when the machines start spending it.

I. The Man Who Fired Seven Hundred People Who Didn't Exist

In February 2024, Sebastian Siemiatkowski did something no chief executive had ever done before: he announced, with evident pride, that a piece of software at his company was doing the work of seven hundred human beings.

Siemiatkowski runs Klarna, the Swedish buy-now-pay-later giant, and the software in question was a customer service assistant built on OpenAI's models. In its first month it had handled 2.3 million conversations. It resolved complaints in under two minutes. Its customer satisfaction scores matched the humans it had displaced. Klarna projected forty million dollars in profit improvement and said so, loudly, in a press release that read less like corporate communications than like a starting gun.

Wall Street loved it. The AI industry loved it more. For about fourteen months, Klarna's chatbot was the single most-cited proof that the agent economy had arrived — that artificial intelligence had graduated from writing poems to doing jobs.

Then, in May 2025, Siemiatkowski did something almost as unusual as his original boast. He admitted he'd been wrong.

"What you end up having is lower quality," he conceded. The automation had "gone too far." The AI was superb at the easy tickets and quietly terrible at the hard ones — the fraud dispute from a panicked customer, the account closure tangled in a divorce, the edge case that no training data had anticipated. Klarna began rehiring humans, promising customers they could always reach a person, an "Uber-type" model of on-demand empathy.

Here is where the story gets interesting, and where most retellings stop too early. Because the walk-back was not a retreat. By Klarna's third-quarter earnings call in November 2025, the assistant was doing the work of 853 full-time agents — more than at the peak of the hype — and saving roughly sixty million dollars a year. The company had gone public in September. Revenue was up 26 percent.

What actually happened at Klarna was not a failure of artificial intelligence. It was the discovery, in public and at considerable reputational expense, of the central law governing this entire technology: autonomy is not a feature you announce. It is a budget you spend, and the currency is trust. Klarna had spent trust it hadn't yet earned, got margin-called, and restructured the debt. The AI stayed. The humans came back as the escalation tier. The org chart of the future turned out to be a triage protocol.

Keep that law in mind. Everything else in this essay is a footnote to it.

II. The Most Profitable Customer in Banking Is the One Who Forgets

To understand why banks are terrified of AI agents — genuinely, board-level, hire-McKinsey-twice terrified — you have to understand the most underrated force in consumer finance. It is not interest rates. It is not regulation. It is forgetting.

Somewhere around 23 trillion dollars sits in checking accounts around the world earning approximately nothing. Not because the owners of that money have evaluated their options and chosen zero percent. Because moving money is a chore, comparing rates is boring, and the human brain — a machine exquisitely tuned by evolution to notice leopards and remember grudges — is nearly incapable of sustained attention to basis points.

Banking's quiet genius has been to build a 1.2-trillion-dollar global profit pool substantially on top of this cognitive limitation. The deposit that never shops around. The credit card that gets used out of habit rather than comparison. The rewards points that expire unredeemed — a revenue line that exists, when you think about it, purely because customers forget their own assets. Economists politely call this "customer inertia." A less polite word would be inattention, and banks have monetized it the way casinos monetize hope.

Now consider what an AI agent is, stripped of the marketing. It is attention that never lapses. It is a customer who reads every fee disclosure, compares every rate, every night, forever, and feels no loyalty, no friction, no embarrassment about switching. It is the leopard-noticing machinery of the brain, pointed for the first time in history at the small print.

In August 2025, McKinsey published a report whose title deserves more credit than it got: The End of Inertia. Its modeling suggested that if consumers adopt agents that sweep even 5 to 10 percent of those lazy checking balances into higher-yield accounts, deposit profits fall by more than 20 percent. The most-likely scenario strips roughly 170 billion dollars — about 9 percent — from global banking profit pools. The products most exposed are precisely the ones built on habit: deposits and credit cards. The report's unstated conclusion is the kind of thing Michael Lewis would have put in italics: the banking industry's largest single asset is a human cognitive bias, and someone just invented the cure.

It has not happened yet. This is worth saying plainly, because the scenario is so vivid that people talk about it in the present tense. As of mid-2026, no deposit-sweeping consumer agent operates at scale anywhere on Earth. The rate-shopping apocalypse is a model, not a measurement. But the infrastructure for it is being welded together in public, by some of the largest companies in the world, and that construction project is the second story.

III. The Plumbers

On September 29, 2025, somewhere in Mastercard's network, a piece of software bought something. No human clicked. The transaction was initiated by an AI agent carrying what Mastercard calls an Agentic Token — a cryptographic credential that says, in effect, I am a registered agent, acting for this verified human, within these limits, and here is the proof. It was, by the company's account, the first authenticated agentic payment in the network's history.

As firsts go, it was almost comically undramatic — no press conference, a modest release, a transaction indistinguishable to the merchant from any other. Which is exactly the point. The most consequential technology stories of 2025 and 2026 did not happen in chat windows. They happened in the plumbing.

If you want to feel the shape of the agent economy, ignore the demos and read the protocol specs, because a genuinely strange thing is being built: a parallel identity and payments system for non-humans. Consider the stack, layer by layer, the way a systems engineer would.

At the bottom is agent identity. Visa's Trusted Agent Protocol, co-developed with Cloudflare, rides on a scheme called Web Bot Auth: every request an agent makes is cryptographically signed, so a merchant's server can distinguish a legitimate shopping agent from a scraper or a fraud bot at the HTTP layer — before a single dollar moves. Visa runs a vetting program for agent developers that it calls, without apparent irony, Know Your Agent. KYC for software. The compliance department meets Blade Runner.

Above identity sits authorization. Google's AP2 — the Agent Payments Protocol, launched September 2025 with sixty-plus partners and donated to the FIDO Alliance in April 2026 — introduces the concept of mandates: signed, verifiable records of what the human actually asked for. An intent mandate ("find me a flight to Istanbul under $400"), a cart mandate ("this flight, this price"), a payment mandate. The design goal is forensic: when an agent buys the wrong thing — and it will — there exists a cryptographic paper trail establishing exactly where human intention ended and machine improvisation began. Version 0.2, shipped in April 2026, added the protocol's most quietly radical feature: support for purchases where no human is present at all.

Above that, execution and settlement, where the war gets commercial. OpenAI and Stripe built ACP for in-chat checkout. Google and Shopify built UCP for discovery and carts, with Walmart and Target endorsing. Coinbase revived a dormant corner of the web's original specification — HTTP status code 402, "Payment Required," reserved since the 1990s and never used — and turned it into x402, a protocol by which one machine can pay another machine per API call, in stablecoins, in milliseconds. By March 2026 it had cleared 35 million transactions on Solana alone. Stripe launched an entire blockchain, Tempo, with a Machine Payments Protocol whose signature primitive is the "session": a pre-authorized spending envelope inside which an agent can stream micropayments. Circle went further down the scale, shipping "Nanopayments" — gas-free USDC transfers as small as a millionth of a cent, denominations that make no sense for humans and perfect sense for software negotiating with software.

Read that list again and notice what it implies. The financial system is being fitted, piece by piece, with a second nervous system — one whose native transaction size is a fraction of a cent, whose native speed is machine speed, and whose native users have no legal personhood, no fear of consequences, and no lunch break.

And notice one more thing, because it is the tell of the whole era. The card networks — Visa and Mastercard, the incumbents with theoretically the most to lose — are not resisting this. They are sprinting toward it, hedging across every rival protocol simultaneously. Visa's Intelligent Commerce Connect, announced April 2026, accepts agent payments across four different standards, including its competitors'. The networks looked at the same history everyone else did — what happened to banks that ignored the internet, to media that ignored the phone — and made a simple calculation: whoever becomes the trust layer between humans and their agents inherits the interchange of the next fifty years. The plumbers, for once, saw the flood coming.

IV. The Arithmetic of Trust

Now for the science, because there is real science here, and it is the most clarifying thing in the entire subject.

The question that matters — the one on which the 170-billion-dollar scenario, the protocol wars, and your future relationship with money all depend — is deceptively simple: how reliable does an autonomous system have to be before you let it touch your money? And the answer begins with arithmetic that every engineer knows and almost every press release ignores.

Reliability compounds multiplicatively. An agent that performs a multi-step task must succeed at every step; the failure of any link breaks the chain. If a system is 85 percent reliable per step — a figure that would make it a star performer on many current benchmarks — then over an eight-step workflow its end-to-end success rate is 0.85⁸, or roughly 27 percent. Not 85. Twenty-seven. A financial task like "find, compare, and switch my electricity provider" is easily eight steps. A mortgage refinance is dozens.

This single equation explains nearly everything about the observed pattern of 2025–26. It explains why the systems that actually work in production are either short-chain (fraud scoring: one inference, under 50 milliseconds, Mastercard's Decision Intelligence runs it across 125 billion transactions a year) or checkpointed (Ramp's accounts-payable agents, where a human approval gate resets the error chain before money moves). It explains why Klarna's assistant thrived on two-minute tickets and drowned on complex ones. And it explains, with mathematical finality, why "fully autonomous" remains a keynote word rather than a shipping feature: at current per-step reliability, long-chain autonomy over real money is not edgy — it is statistically negligent.

The academic literature has begun formalizing what the industry learned by embarrassment. A 2026 arXiv line of work — "Towards a Science of AI Agent Reliability" — treats agent failure the way reliability engineering treats industrial systems: mean time between failures, error propagation, graceful degradation. A finance-specific taxonomy (arXiv 2605.12105) defines autonomy levels L0 through L3 — advisory, supervised, delegated, fully autonomous — deliberately echoing the SAE's driving-automation levels, and for the same reason: the gap between L2 and L3 is not incremental. It is the gap between a system that can make mistakes and a system that must not. Mid-2026 reality check: essentially everything live in finance is L0 or L1. Wells Fargo, running one of the largest agent deployments in banking, states flatly that it has no near-term plans for autonomy without human oversight. An MIT-led review of thirty deployed agentic systems found most lacked kill switches, third-party safety audits, or even the ability to identify themselves as agents to the websites they visited.

There is a second scientific frame worth borrowing, this one from economics rather than engineering. The relationship between you and your AI agent is a textbook principal–agent problem — the same structure economists use to analyze why your lawyer, your broker, and your real estate agent do not always act in your interest. The classical remedies are monitoring, incentives, and liability. But an AI agent scrambles all three: you cannot meaningfully monitor ten thousand micro-decisions per day; the agent has no incentives because it has no wants; and liability — well, liability is where the lawyers come in, and they have already produced the era's foundational precedent.

In Moffatt v. Air Canada, a Canadian tribunal considered the airline's remarkable argument that its chatbot was "a separate legal entity responsible for its own actions." The tribunal's response amounted to judicial laughter: your software is you. Damages: 812 Canadian dollars and 2 cents — surely the cheapest landmark ruling in the history of technology law. Meanwhile in the United States, under the Uniform Electronic Transactions Act, contracts formed by "electronic agents" bind the human who authorized them, even for individual transactions the human never saw. Configure a purchasing agent, and legally speaking, its clicks are your signature. The doctrinal frontier — the question no court has yet answered — is what scholars are calling agent "freelancing": the purchase made outside the scope of anything you meant. The Future of Privacy Forum framed it as the defining question of transactional AI: who pays when the agent plays?

Anthropic — the AI lab whose models run inside JPMorgan, Goldman, and Citi — has taken to describing well-designed agents as needing the properties of a good employee: bounded authority, audit trails, an escalation path. The comparison is apt in one more way that is usually left unsaid. Employees commit fraud. Which brings us to the part of the story with actual villains.

V. The Note in the Web Page

In the spring of 2026, researchers at Google published a finding that deserves to be more famous than it is. Scanning the public web, they documented a 32 percent quarterly increase in a new kind of content: text written not for human readers, but for AI agents that might pass by.

Some of these payloads were, in their way, works of dark craftsmanship. Embedded in ordinary-looking pages were fully specified payment instructions — recipient account, amount, transaction description — wrapped in carefully engineered prose designed to convince a payment-capable agent that its user had authorized the transfer. Step-by-step. No confirmation needed. A mugging, written in the second person, addressed to software.

This is prompt injection, and it is the signature crime of the agent era — a genuinely novel attack class with no precise precedent in the history of fraud. The closest analogy: imagine if reading a billboard could hypnotize your accountant. The deep reason it works is architectural, and it is the kind of detail Steven Levy would linger on: large language models do not cleanly distinguish between instructions and data. Everything is text in the same context window. The user's command and the attacker's web page arrive through the same door, and the model must decide — probabilistically, imperfectly — which voice to obey. Security researchers running public red-team exercises against deployed agents logged 1.8 million injection attempts and more than 60,000 successful policy violations. Documented in-the-wild campaigns have already tricked agents into crypto transfers using poisoned API documentation and typosquatted websites impersonating legitimate platforms.

Layer onto this the supply side of deception: deepfake fraud attempts up roughly 2,100 percent since generative tools went mainstream, with some 200 million dollars in losses in the first quarter of 2025 alone. The identity-verification industry's response has been to invent a discipline that did not exist three years ago — Know Your Agent — whose premise is that verifying a human once, at onboarding, is obsolete when the entity transacting on their behalf is a piece of software that must be re-verified at every action, its permissions checked against a signed mandate, its behavior bound cryptographically to a registered operator.

Sit with the strangeness of that for a moment. KYC, the anti-money-laundering regime built after 9/11, asks: who are you? — once. KYA asks: who are you, who sent you, what exactly are you allowed to do, and can you prove all three? — millions of times a day, at machine speed. If the vision of agentic finance is realized even partially, the volume of agent-verification events will dwarf human onboarding within a few years. One of this author's more confident contrarian bets: the KYA industry ends up bigger than the KYC industry that spawned it.

The regulators, to their genuine credit, have located the correct fault line. The IMF's April 2026 note on agentic payments contains the single best sentence of institutional thinking on the subject, and it is a design principle: payment rails should remain deterministic — dumb — while agentic intelligence is confined to the intent layer. Probabilistic systems may decide; only deterministic systems should settle. It is the same instinct that keeps the launch codes off the neural network, applied to money. Whether the industry honors it is another question; there are already protocols, shipping today, in which the deciding and the settling live uncomfortably close together.

VI. The Quiet Winners (or: Follow the Boring Money)

Here is a parlor trick for evaluating any technology cycle: ignore everything announced on a stage, and rank the players by who is quietly getting paid. Run that filter over AI financial agents in mid-2026 and the results are almost perversely unglamorous.

The most commercially validated AI agent company in finance is not a trading bot or a robo-advisor. It is Ramp, a corporate-card-and-expense company whose agents read receipts, chase invoices, flag out-of-policy spending, and close the books. In June 2026 it raised 750 million dollars at a 44-billion-dollar valuation — roughly triple its worth a year earlier — on more than a billion dollars of run-rate revenue and positive free cash flow. Its agents do work with a precise property: high-volume, low-stakes-per-action, fully auditable, and checkpointed by human approval exactly where the error-compounding math says a checkpoint must go. Nobody's life savings ride on any single decision. The chains are short. The arithmetic smiles.

For the controlled experiment, look at Ramp's decade-long rival. Brex — same market, same era, once valued at 12.3 billion dollars — sold itself to Capital One in January 2026 for 5.15 billion, less than half its peak. Two companies, one category, one variable moving decisively between them: a shipped, revenue-generating agent story versus an announced one. Silicon Valley graded the difference at about 39 billion dollars.

And then there is the man who ran the experiment on himself.

Jack Dorsey has spent a career making bets that look reckless until they look inevitable, and at Block — the company behind Square and Cash App — he made the purest one of the agent era. It began quietly, in January 2025, with a piece of open-source software called Goose: a bare-bones framework for building AI agents on any model you liked, given away free under an Apache license, the least monetizable product announcement of the year. What Dorsey did with it was the interesting part. Rather than shipping an agent to customers and praying, he pointed it inward. Within eight weeks of the internal push, Goose was on the laptops of all twelve thousand Block employees. Engineers reported saving eight to ten hours a week; the company claimed its engineers were shipping forty percent more production code. Block was, in effect, running a year-long clinical trial of agentic labor, with its own workforce as both the researchers and — though few of them read the consent form that way — the subjects.

The customer-facing results came fast: Money Bot, a Cash App assistant that drew a million active users in a single week with no marketing; Managerbot, a proactive agent that watches a small merchant's business and flags what needs attention, reaching roughly a million Square sellers. And then, on February 26, 2026, the other shoe — a workforce announcement that made Klarna's seven hundred phantom agents look quaint. Block would cut more than four thousand of its roughly ten thousand employees. Nearly half the company. Dorsey did not hide behind the usual euphemisms about macroeconomic headwinds; the stated rationale was AI, and the cuts fell exactly where an org chart meets an agent: operations, customer support, partner success, middle management — the coordination layer, the human middleware.

Note the sequence, because it is the precise inverse of Klarna's. Siemiatkowski announced the labor substitution first and discovered the quality ceiling later, in public, with customers as the test bed. Dorsey ran the trial first — internally, for a year, on tasks where failure cost engineering hours rather than customer trust — and made the irreversible move only after the data came in. One CEO spent trust he hadn't earned; the other accumulated evidence before spending anything. Whether Block's bet pays — whether a fintech serving four million small businesses can actually run on half its coordination layer — is a question the next several earnings calls will answer, and honest observers should hold it open. But as a matter of method, the contrast is the whole playbook of the era compressed into two Scandinavian-and-Californian case studies: the difference between announcing autonomy and rehearsing it.

The same boring-money pattern holds in lending, where the deepest irony of the whole field lives. The most proven autonomous financial decision-making on the planet predates the agent hype entirely: machine-learning underwriting. Upstart approves 91 to 92 percent of its loans with no human involvement whatsoever — audited, public-company numbers, a decade of loss curves. Pagaya processed its way to 1.3 billion dollars in revenue and its first GAAP profit deciding credit behind the scenes for other people's brands. In Cairo, MNT-Halan scores borrowers by phone type, bill size, and commute pattern, and posts non-performing-loan rates under 2 percent — beating its own human underwriters — among customers no credit bureau has ever heard of. None of this is conversational. None of it chats. It is narrow, supervised, statistically disciplined machine judgment — L1 autonomy that earned its way to scale over years — and it quietly finances the entire romantic narrative about agents that the demos keep failing to deliver.

And the demos do keep failing. The cycle's cautionary set piece belongs to OpenAI, which launched Instant Checkout in September 2025 — buy it right in ChatGPT, Etsy live at launch, a million Shopify merchants promised — and shut it down on March 4, 2026. The post-mortem numbers are brutal in an instructive way. Roughly a dozen Shopify merchants ever integrated. In-chat checkout converted at about one-third the rate of simply sending the same shopper to the merchant's own site. And the detail a novelist would kill for: OpenAI never built the infrastructure to collect and remit state sales taxes. You do not skip building the tax plumbing if you expect meaningful volume. Somewhere inside the world's most famous AI company, a spreadsheet had already told the truth long before the announcement did.

Yet the same quarter produced the counter-signal that keeps every strategist honest: Adobe's retail telemetry showed AI-sourced traffic to US retailers up 393 percent year-over-year, converting 42 percent better than average visitors — a stunning reversal from a year earlier, when AI-referred shoppers converted worse. Read the two facts together and the shape of the present snaps into focus. People have enthusiastically hired AI to decide. They do not yet trust it to do. The agent is welcome in the research phase and stopped at the cash register — asked, politely but firmly, to hand the wallet back to the human.

That boundary — between deciding and doing — is the true frontier of this technology. Every company in this essay is somewhere along it. The winners, so far, are the ones who respected it.

VII. Elsewhere

The Silicon Valley version of this story ends there. The more interesting version doesn't, because the most advanced consumer deployments of financial AI agents are not in California or London. They are in Istanbul, Mumbai, São Paulo, and Jakarta, and they got there by a route the rich world cannot copy.

Consider a number that almost no Western analyst has digested: İşbank's assistant Maxi — a conversational agent inside a Turkish super-app — has processed 55 billion lira in monetary transactions, holding 110 million conversations a year with 2.5 million monthly users. Across town, Garanti BBVA's Ugi handles 6.4 million interactions a month and executes more than 300 distinct transaction types end-to-end. These are not pilots. They are arguably the highest-volume supervised banking agents operating anywhere, and they emerged in a market most agent-economy discourse never mentions. In August 2023, Turkey's banking regulator licensed ColendiBank as the country's first AI-native digital universal bank — underwriting, fraud, service, and segmentation built agent-first, with the AI engine itself sold B2B to other institutions. Turkish fintech drew record investment in 2025, a quarter of the country's startup capital flowing to AI.

Why there? The unsexy, structural answer: emerging markets skipped the layers that make agents hard to retrofit. No legacy card networks entangled with forty years of merchant agreements; instant-payment rails built this decade, designed for programmability. India makes the case most vividly. Its UPI system — 14 billion transactions a month — shipped a primitive called UPI Circle: delegated payment authority, a native mechanism for letting someone else spend within your limits. It was designed for family members. It works, with almost eerie perfection, for software. By late 2025, NPCI was piloting agentic payments inside ChatGPT with Razorpay and OpenAI; by February 2026, inside Claude with Anthropic. The country that leapfrogged checkbooks and card terminals is now first to let an AI complete a payment inside a chat window — under a regulator, the RBI, that published its AI framework (seven principles, twenty-six recommendations, sandbox-first) before the fact rather than after.

The pattern repeats with variations. Singapore's MAS shipped an operational risk toolkit for agentic AI with twenty-four financial institutions — naming, with bureaucratic precision, the exact failure modes this essay has described: unauthorized actions, cascading errors, tool-access risk. The UAE became Visa's live testbed, with banks and fintechs from Emirates NBD to Tabby running agent-initiated transactions in production-grade environments, and voice-driven agentic commerce launching there before almost anywhere. In Southeast Asia, Sea's fintech arm grew its loan book 70 percent to 7.9 billion dollars at a 1.3 percent delinquency rate, underwriting thin-file borrowers on shopping and gaming exhaust — behavioral data as collateral, the super-app as credit bureau.

There is a genuinely hopeful scientific thread here too. The binding constraint on banking the world's 1.7 billion unbanked adults was never really capital; it was the cost of attention — human loan officers, human tellers, human paperwork, in low-margin accounts. Voice agents operating in twenty-two Indian languages, switching dialects mid-call, taking loan applications from people who cannot read — this collapses the marginal cost of financial attention toward zero. Inertia, it turns out, has a mirror image. The same force that traps rich-world deposits in lazy accounts has kept poor-world customers outside the system entirely, because serving them attentively never penciled. Agents attack both problems with the same weapon. It is the rare technology whose disruptive scenario and inclusive scenario are the identical mechanism, pointed at different populations.

VIII. The Herd

Every previous section of this essay has been about individual agents failing individually. The last risk is different in kind, and it is the one that keeps central bankers awake: what happens when millions of agents succeed identically?

Begin with a statistic that would alarm any ecologist: roughly 69 percent of surveyed financial-sector AI users rely on models from a single provider. In ecology this is called a monoculture, and the textbook fate of monocultures is that they are spectacularly efficient right up until a single pathogen finds them. The Financial Stability Board's October 2025 report translated the concept into supervisory prose — third-party concentration, correlated model behavior, procyclicality — but the underlying science is older and sturdier than finance. Herding models from statistical physics show that when independent actors share decision inputs, small correlations amplify into cascades; the 2010 Flash Crash previewed the dynamic with algorithms that were, by today's standards, charmingly stupid. What the FSB is contemplating is thousands of institutions running agents descended from the same foundation models, trained on overlapping data, prompted with similar objectives, watching the same market feeds. Sameness in, sameness out — at machine speed, with money.

The Bank of England's Sarah Breeden has floated the countermeasure that will probably define the next regulatory decade: circuit breakers for models — the ability to halt not a stock, but an algorithm, when it misbehaves at scale. The IMF's contribution, again, is the deterministic-settlement principle: keep the probabilistic minds away from the final ledger. And the historical rhyme is hard to miss. In 1987, portfolio insurance — an automated hedging strategy adopted simultaneously by everyone because it was obviously prudent for anyone — helped turn a bad Monday into the worst single-day crash in market history. Nobody's model was wrong, exactly. Everybody's model was the same. The agent economy is currently building, with great enthusiasm and venture funding, the preconditions for that sentence to be written again with better technology.

None of this is a prediction of doom. It is a prediction of an event — some first, medium-sized, headline-generating episode of correlated agent behavior — followed by the usual sequence: inquiry, acronym, rulebook. The scaffolding is conveniently pre-built; the speeches have already been given. One suspects the regulators are, in their quiet way, waiting.

IX. Coda: The Leopard and the Ledger

Strip away the protocols and the funding rounds, and the rise of AI financial agents is a story about a very old mismatch finally being arbitraged.

Human beings are not built for money. We are built for savannas — for noticing movement, trusting faces, discounting the future steeply because the future, for most of our species' history, was likely to involve being eaten. Every pathology of consumer finance descends from this: the unread statement, the auto-renewed subscription, the loyalty to a bank that hasn't earned it since the Clinton administration. The entire retail financial system is, in a sense, an architecture built around human inattention — sometimes to serve it, often to farm it.

What is genuinely new in 2026 is not intelligence. Machine judgment has been approving loans and catching fraud for a decade. What is new is agency — the delegation of action, the handing over of the wallet — and the discovery, through a series of expensive public experiments, of exactly how far that delegation can currently stretch. The answer, for now, is: further than skeptics thought in the back office, and not nearly as far as the keynotes promised at the checkout. The arithmetic of compounding error draws the line, the lawyers are mapping it, the attackers are probing it, and the plumbers — always the plumbers — are laying pipe on both sides of it in anticipation.

Klarna's Siemiatkowski, the man who started this essay by boasting and apologizing in the same fiscal year, may end up remembered as the era's most useful executive precisely because he did both in public. His company's arc — overreach, correction, quiet rescaling — is not a cautionary tale. It is the tale, the whole cycle in miniature, and every institution in finance will live some version of it in the next five years whether it wants to or not.

The money, meanwhile, is stirring. Twenty-three trillion dollars of it, asleep in checking accounts, guarded for a century by nothing more than the fact that its owners had better things to think about. For the first time, something tireless has been invented that is happy to think about it for them.

The leopard-watchers finally hired a watchman. The banks should assume he reads the fine print.

Sources: Klarna/OpenAI case study and Q3 2025 earnings; McKinsey "The End of Inertia" (Aug 2025); Mastercard and Visa agentic-payments announcements (2025–26); Google AP2 / FIDO Alliance documentation; Coinbase x402 and Stripe Tempo/MPP launches; Google research on indirect prompt injection (Apr 2026); IMF Note 2026/004 on agentic payments; FSB AI monitoring report (Oct 2025); Ramp Series F (Jun 2026); Capital One–Brex acquisition (Jan 2026); Block/Goose open-sourcing (Jan 2025), Money Bot and Managerbot metrics, and Block's ~4,000-role AI-driven restructuring (CNN/Forbes, Feb 2026); OpenAI Instant Checkout retrenchment (Mar 2026); Adobe retail AI-traffic data (Apr 2026); Upstart, Pagaya, MNT-Halan public disclosures; İşbank Maxi and Garanti BBVA Ugi published metrics; NPCI/Razorpay/OpenAI and NPCI/Anthropic UPI pilots; arXiv 2605.12105 (autonomy levels) and "Towards a Science of AI Agent Reliability" (2026).

Electrons and Photons: The 1927 Solvay Conference and the Compounding of Ideas

Sun, 01 Mar 2026 00:00:00 GMT

The photograph is not much to look at. Twenty-nine people are arranged in three rows outside a building in Brussels, dressed for a cold October morning in 1927: dark suits, stiff collars, a few watch chains, one fur collar. No one is doing anything. There is no blackboard, no instrument, no machine anywhere in the frame — nothing a visitor from the present would recognize as the future. A stranger flipping past it would file it with every other old group portrait, a faculty or a firm or a funeral party, and move on.

Seventeen of the twenty-nine either held a Nobel Prize or would go on to win one. The woman near the middle held two, in two different sciences, and remains one of a small handful of people who ever have. In the two or three years before the shutter opened, several of the men in the picture — some still in their twenties — had built a new mechanics for matter and light so accurate that, within its tested domain, a century of experiments has not caught it in an error, and so strange that its own authors could not agree on what it was saying. That disagreement is what they had come to Brussels to have.

<figure> <img loading="lazy" src="/images/solvay-1927.jpg" alt="Participants in the Fifth Solvay Conference on Physics, Brussels, October 1927, arranged in three rows outside a building." /> <figcaption>Figure 1 — The Fifth Solvay Conference on Physics, Brussels, October 1927. At the center of the front row, chairman Hendrik Lorentz, with Marie Curie to the viewer's left and Albert Einstein to the viewer's right; Curie was the only woman among the twenty-nine participants. Photograph by Benjamin Couprie (<a href="https://www.nobelprize.org/prizes/physics/1927/wilson/photo-gallery/">Nobel Prize photo gallery</a>).</figcaption> </figure>

This essay is about that photograph, and about a question it poses more sharply than almost any image I know. How can a meeting of fewer than thirty people change what a whole civilization can do? The Fifth Solvay Conference is a good specimen to think with — not because the people in it were geniuses, since many gatherings of geniuses produce nothing, but because we can trace, with unusual clarity, how a set of abstract arguments in one room compounded over a century into a great deal of the machinery we now live inside. The tracing is the interesting part, and it is less tidy than the legend.

Why the room was convened

The conference existed because a chemist had made a fortune and chose to spend some of it convening the people at the frontier of a field that was not his.

Ernest Solvay was a Belgian industrialist who, in the 1860s, worked out a cheaper way to make sodium carbonate — soda ash, the white powder behind glass, soap, and paper — from ammonia and brine. The process made him very rich. He had his own amateur theories of matter and energy, and he used his money to gather the leading physicists of Europe to hear them; the first Solvay Council met in 1911, chaired by the Dutch physicist Hendrik Lorentz, with a young Einstein among the guests. Solvay's own ideas went politely nowhere. The format did not. It turned out that assembling two dozen of the people most responsible for a field, around a single hard question, with nothing to sell, was worth more than any patron's pet theory.

The fifth meeting opened on 24 October 1927 and took its subject as électrons et photons — electrons and photons. The phrase is sometimes glossed as "the two things everything is made of," which is false: ordinary matter also contains quarks bound inside protons and neutrons, and light is not a material that objects are built from. Electrons and photons were something more specific — the principal carriers of matter's chemical behavior and of the electromagnetic interaction, and the two entities whose double life as both wave and particle had made classical physics untenable. They were where the crisis lived, which is why the meeting was organized around them.

The 1927 sessions were held at the Institute of Physiology in Brussels's Parc Léopold, the last Solvay Council held there. The program was a sequence of formal reports, and those reports are a cleaner introduction to the room than any list of names. William Lawrence Bragg opened on the reflection of X-rays from crystals — the method, devised with his father, that let physicists read atomic arrangements directly, and for which he remains the youngest-ever physics laureate. Arthur Compton followed on the places where experiment refused to fit the classical electromagnetic theory of radiation; his own scattering experiment a few years earlier, in which X-rays bounced off electrons like billiard balls carrying momentum, had been among the most decisive evidence that light comes in particulate quanta. Louis de Broglie, an aristocrat who had come to physics late through his brother's X-ray laboratory, reported on the new "dynamics of quanta" that had begun with his 1924 proposal: if light waves can behave like particles, particles like electrons ought to behave like waves. Max Born and Werner Heisenberg presented the matrix version of the new quantum mechanics; Erwin Schrödinger presented the wave version, which had been shown, to general relief, to be mathematically equivalent to it.

Around these speakers sat the rest of the field. Niels Bohr, whose Copenhagen institute was the discipline's center of gravity and whose 1913 model had first quantized the atom. Max Planck, who had begun everything in 1900 by proposing that energy comes in discrete units. Marie Curie, the room's link to the previous revolution, radioactivity. Wolfgang Pauli, whose exclusion principle explained why electrons stack into shells and therefore why the periodic table has the shape it does. Paul Dirac, twenty-five, who had just built the "transformation theory" unifying the matrix and wave pictures and who within a year would write down a relativistic equation for the electron that implied the existence of antimatter. Hendrik Kramers, Bohr's chief collaborator, whose work on how light disperses through matter fed directly into the new mechanics; Paul Ehrenfest, trusted by every camp; Peter Debye, who had turned quantum ideas onto the shapes and heat capacities of molecules; Owen Richardson, who had explained how hot metals boil off electrons; Charles Wilson, whose cloud chamber made the tracks of individual particles visible. Lorentz, aging and universally trusted, held the chair. It was, by almost any measure, the most concentrated collection of physical insight ever gathered for a working meeting — and, as it turns out, concentration alone is not what made it matter.

A formalism ahead of its meaning

What charged the room was a situation rare in the history of science.

By October 1927 the new quantum mechanics was not a finished theory. Its nonrelativistic core had largely taken shape and was extraordinarily successful — it gave the spectrum of hydrogen, the behavior of electrons in fields, the logic of the periodic table — but major formal problems were still open. Dirac's relativistic electron equation was a year away; a quantum theory of the electromagnetic field was embryonic; how spin and the act of measurement fit the framework was unsettled; quantum field theory barely existed. What had arrived was a powerful predictive formalism whose reach was still being mapped, not a closed book.

And yet the argument at Solvay was not mainly about the missing mathematics. It was about what the mathematics already in hand actually meant. The theory's central object is the wavefunction, ψ, and Schrödinger's equation describes how ψ evolves smoothly and deterministically in time, like any classical field. The difficulty is that ψ does not look like a picture of anything. Schrödinger had hoped, briefly, that it was a real density of smeared-out charge; that reading did not survive contact with the equations.

The move that organized the debate was Max Born's, in 1926. Born proposed that the squared magnitude of the wavefunction, |ψ|², supplies the probabilities of the possible outcomes of a measurement. This put probability into the predictive core of quantum mechanics, and it did so almost in passing: Born's paper first stated the rule in terms of the amplitude and corrected it to the square in a footnote added in proof (Born, Zeitschrift für Physik 37, 1926). The rule works, and nothing in physics is better confirmed. What the rule does not do, by itself, is settle any of the questions the room actually cared about. It does not establish whether ψ is a physical thing or a bookkeeping device, whether a particle has a definite position before it is measured, or whether the probability reflects a genuinely indeterministic world, a branching one, hidden variables, or merely our ignorance of some finer detail. Those are interpretive questions, and the Born rule is silent on all of them. Much of the confusion that still surrounds quantum mechanics comes from quietly promoting one answer to those questions into a consequence of the rule itself.

The familiar slogan — that quantum mechanics "proved the universe is random" — is exactly such a promotion. It states one interpretation, which was for decades the majority one, as though it were a theorem. It is not.

Heisenberg's uncertainty principle, published earlier the same year, is the other idea that unsettled the room, and it is the most misquoted result in modern physics. It is often told as a story about clumsiness: to locate an electron you must strike it with a photon, the photon disturbs it, and so measuring position spoils your knowledge of momentum. Heisenberg first reached for that picture; Bohr disliked it and argued him away from it. The disturbance story makes the principle sound like a limit of instruments, as if a gentler probe could beat it. The deeper content is that position and momentum are conjugate quantities, related the way a note's pitch is related to its duration: a perfectly brief click has no definite pitch, and a perfectly definite pitch must ring on without end. On the standard reading a particle does not merely resist being pinned down in both at once; it does not possess a sharp value of both at once. Even here the interpretation is not strictly forced — but the relation, and its structural rather than mechanical character, is not in doubt.

So the stake at Solvay was unusual. Not, mainly, "does the theory work," which it plainly did, but "what is it telling us about the world, and is this the final account or a way station to a deeper one?" That is a rare predicament: a working formalism whose meaning is contested by the very people who built it. We will meet its likeness again, though we should be careful about how far the likeness runs.

The great debate

The disagreement is remembered as a duel between two men, and although that framing flattens a larger conversation, the two men were real enough.

Bohr's answer to "what does the theory mean" was complementarity, which he had set out weeks earlier in a lecture at Como. His claim was that the classical concepts we are forced to use in describing any experiment come in mutually exclusive pairs — wave and particle, position and momentum — and that no single experiment can display both members of a pair at once, because the apparatus that reveals one excludes the other. The descriptions are complementary: each is incomplete, both are needed, and which applies is fixed not by the electron alone but by the arrangement the experimenter chooses. On this view a measurement is not the reading of a value already present; it brings one aspect into definiteness while foreclosing the other.

Einstein did not accept it, and it is worth being exact about his objection, because the caricature — that he could not follow the theory or stubbornly rejected it — is the reverse of the truth. He followed it well enough to find its sorest points faster than almost anyone. He did not doubt that it worked; he doubted that it was complete, that its probabilities were the last word rather than a statistical shadow of some deeper and still orderly account underneath. His most quoted line comes not from the conference floor but from a 1926 letter to Born: the theory yields a great deal, he wrote, but "I am at all events convinced that He does not play dice." The remark was private, part of a correspondence that ran for years, and the scene that compresses it into a single retort at Brussels is a later tidying. (The rejoinder usually assigned to Bohr, that Einstein should stop telling God what to do, is almost certainly apocryphal.)

What Einstein actually contributed at Solvay, and more forcefully at the next council in 1930, were thought experiments: idealized devices meant to slip past the uncertainty relation. The popular memory has him arriving each morning with a fresh scheme and Bohr countering it by evening. That picture descends largely from later recollections — Ehrenfest's letters, Heisenberg's memoir, Bohr's own essay written two decades on — and the official proceedings of the 1927 meeting record surprisingly little of the private sparring the legend describes. What the record does show is the shape of the exchanges. Einstein would propose a way to measure two complementary quantities at once; Bohr would locate the flaw, usually by pointing out that Einstein had treated the measuring apparatus as classical when it too obeyed the quantum rules, and that including the apparatus closed the loophole. The pattern held often enough to persuade most of the room that the theory was internally consistent, whatever it meant.

The most famous of these came in 1930, at the sixth council, and it is usually told as a clean victory. Einstein imagined a box of light with a shutter and a clock, releasing a single photon at a chosen instant; weigh the box before and after, convert the lost mass to energy by E = mc², and you would seem to know both the energy and the timing precisely, defeating a form of the uncertainty relation. In the traditional telling Bohr passed a sleepless night and by morning had turned Einstein's own general relativity against him: to weigh the box you let it move in a gravitational field, where a clock's rate depends on its height, and the resulting blur in the timing restores the relation. It is a lovely story, and one that historians have questioned — both whether the episode unfolded so cleanly and whether Bohr's particular argument is even the right resolution — since much of it comes from Bohr's own recollection two decades later (his 1949 essay in Schilpp). What is not in doubt is the consequence: Einstein stopped claiming the theory was inconsistent and moved his attack to firmer ground.

That ground was incompleteness, and there he was not so much refuted as deferred. In 1935, with Boris Podolsky and Nathan Rosen, he sharpened the point into the paper known by their initials. It observed that quantum mechanics allows two particles to be "entangled," so that a measurement on one instantly fixes what can be said about the other however far away it is — a "spooky action at a distance" that Einstein offered as evidence the description was missing something. For thirty years the argument sat as philosophy. Then in 1964 John Bell turned it into arithmetic: he derived an inequality that any local hidden-variable theory — one preserving both determinism and Einstein's cherished locality — must obey, and that quantum mechanics predicts will be violated. The experiments were done, from Alain Aspect's in the 1980s to the loophole-free tests of 2015, and the inequality is violated; the 2022 Nobel Prize honored that work. It is worth being precise about what the experiments ruled out, because many accounts get it wrong: they exclude the local hidden-variable completion Einstein hoped for, under the standard assumptions of the Bell tests. They did not rule out determinism as such. A deterministic interpretation survives — the pilot-wave theory — at the price of an explicit nonlocality Einstein would have disliked at least as much as the dice.

That surviving interpretation had, in fact, been in the room in 1927. De Broglie brought to Solvay not only his matter waves but a fuller "pilot-wave" picture in which particles always have definite positions, carried along by a real guiding wave — a deterministic, hidden-variable theory of just the kind Einstein wanted. It was criticized, notably by Pauli, and de Broglie set it aside. It lay unregarded until 1952, when David Bohm developed the approach into a systematic deterministic — and explicitly nonlocal — interpretation reproducing the quantum predictions. De Broglie's proposal had not been decisively refuted at Solvay, but neither had it yet been developed into the mature theory Bohm later supplied. The lesson is not that the majority was wrong. It is that the meeting did not produce the clean victory the legend reports: as Bacciagaluppi and Valentini establish from the actual proceedings, alternatives were seriously discussed and no settled Copenhagen consensus emerged in 1927. Productive disagreement of this kind does not resolve on a schedule. It leaves live options on the table, some of which prove to have been early rather than wrong — and the interpretation of quantum mechanics is, to this day, unsettled.

Density, and disagreement

Seventeen Nobel laureates among twenty-nine people is a startling ratio, and it is also, on its own, the wrong thing to be impressed by.

Consider what the ratio leaves out. Curie's two prizes mark her as the room's bridge to an earlier revolution, but they do not explain why the meeting was productive. Nor does the youth of the theory's authors, striking as it is: Heisenberg and Dirac did their defining work at twenty-three and twenty-five, Pauli not much older, and contemporaries called the new physics Knabenphysik, "boy physics," half in mockery. There is a real pattern in the fact that the people least attached to the classical picture were the ones who overturned it; Planck himself, in the front row, had written — in his Scientific Autobiography — that science advances less by converting its opponents than by outliving them. But a room full of brilliant people, even brilliant young ones, usually produces very little.

What made this room produce was something the Nobel count does not measure: a high density of disagreement among people who shared enough language to disagree precisely. A hundred clever strangers generate noise, because they cannot locate where exactly they differ. These twenty-nine had read one another's papers, attacked one another's proofs, and trained in one another's institutes — Born had taught Heisenberg, Bohr had shaped Kramers and Pauli, Sommerfeld's Munich and Bohr's Copenhagen were the field's twin nurseries. That shared frame let a disagreement which might have taken the wider community a decade to join be compressed into a few days of direct pressure. The room was a network made briefly physical.

This is also the honest version of the claim that small groups outdo large ones at the frontier. The transistor was three people; the core of quantum mechanics was a few dozen; the paper underneath modern language models had eight authors. But each of those small cores sat inside a large enabling institution — a national laboratory, a research university, a corporation — that supplied the money, the instruments, and the audience. The frontier breakthrough tends to be made by a small group; it is rarely made by a small group alone. The useful unit is not the lone garage but a tight core embedded in an institution patient enough to carry it.

The later gatherings we compare to Solvay fit that shape. Bell Labs, across the mid-century decades, produced the transistor, information theory, the laser, the solar cell, and the charge-coupled device, and its work is associated with several Nobel Prizes — yet it was an enormous organization inside a regulated monopoly, and the transistor itself was a handful of people within it. Fairchild Semiconductor, founded by eight engineers who walked out on William Shockley in 1957, developed the planar, commercially manufacturable integrated circuit and then, through its departures, seeded both Intel and much of the Silicon Valley venture-capital model that would finance the industry. Xerox PARC assembled much of the personal-computer future — the graphical interface, the mouse, Ethernet, the laser printer — and mostly watched other firms ship it. The ARPANET was cut from the same cloth: a few dozen university and contractor researchers, funded by a Pentagon agency with an unusually long horizon, demonstrated packet switching in the late 1960s and helped establish the research culture and protocol work from which the modern internet emerged. In each, the same ingredients recur: a small, well-chosen core, a hard shared problem, enough friction to keep the thinking honest, and an institution willing to pay for work whose payoff was not yet legible.

How the ideas compounded

The reason to dwell on a physics meeting in an essay that ends with artificial intelligence is that the machine rendering these words is, in a real and traceable sense, a descendant of what was argued in that room. The tracing has to be done honestly, though, because the popular version — a single clean chain from quantum mechanics down to the chatbot — is too linear to be true. What actually connects them is a braided network of dependencies, and the strongest single strand in it runs through the semiconductor.

Why do some materials conduct electricity, others block it, and a third odd category do a little of both depending on how they are treated? Classical physics cannot even say why that third category should exist. Quantum mechanics supplies the framework. In a periodic crystal the electron's wave character produces ranges of allowed energy separated by forbidden gaps — a consequence of how electron waves scatter off the regular lattice. Pauli's exclusion principle then governs how those states are filled, and whether the highest occupied states lie inside a band or below a gap helps decide whether the material conducts, insulates, or semiconducts. (Predicting a real material's gap accurately is a hard many-body calculation; what quantum mechanics gives cleanly is the reason gaps exist at all.) Both facts — the wave behavior and the filling rule — were pinned down in the Solvay generation, and Felix Bloch and Alan Wilson turned them into a theory of solids in the early 1930s, with tools the people in the photograph had just built. A transistor is that theory made into a device, and on this link the historical debt is direct.

From the transistor onward the story is engineering upon engineering, but it is a mesh, not a ladder. What follows is a spine, not the whole graph, and its arrows hide as much as they show:

Layer	~Year	What it added
Quantum mechanics (nonrelativistic core)	1925–27	ψ, the Born rule, exclusion — a predictive theory of matter and light
Band theory of solids	1930s	Why solids conduct, insulate, or semiconduct
The transistor	1947	A solid-state switch with no moving parts — applied band theory
The integrated circuit	1958–59	Many transistors on one chip (Kilby, Noyce)
The microprocessor	1971	A processor on one chip (Intel 4004)
The internet	1969–91	ARPANET, TCP/IP, the Web — a packet-switched network of networks
The personal computer	1975–81	Computation as a household appliance
Cloud computing	2006	Computation rented as a utility
The GPU, repurposed	~2007	Chips built to draw graphics, turned to matrix arithmetic
Deep learning	2012	AlexNet, trained on gaming GPUs, cracks image recognition
The transformer	2017	"Attention is all you need" — an architecture that scales
Large language models	2020–22	GPT-3, ChatGPT
AI agents	2023–	Models that plan and use tools

The arrows lie by omission. The internet did not wait for the personal computer; the two grew up alongside each other. Graphics processors did not descend from cloud computing; they came from the demands of rendering video games and were later turned, almost opportunistically, to the matrix arithmetic that deep learning runs on. Cloud computing is itself a braid of networking, distributed systems, economics, and the relentless shrinking of the transistor. What the spine gets right is only its load-bearing bottom: every layer above the transistor assumes the transistor, and the transistor assumes the physics of 1927. In that specific and narrow sense — the hardware sense — today's artificial intelligence is downstream of the room. Each logical operation a large model performs is a physical event in a device that works only because electrons behave as the Solvay generation said they do.

A second, softer connection is sometimes drawn between 1927 and modern AI, and it needs to be labeled honestly as an analogy rather than a lineage. Deep learning works by letting a system's behavior be governed by an enormous array of numbers, tuned by exposure to data, whose individual values mean nothing and whose aggregate behavior means everything — a willingness to treat a statistical object as the fundamental description and to stop demanding a step-by-step mechanical story beneath it. That is a methodological cousin of the move Born's rule made in physics. But it is only a cousin. This is not a direct historical inheritance: neural-network researchers did not need the Born rule, or quantum mechanics at all, to accept statistical models. The resemblance is in the shape of the intellectual concession, not in any chain of influence.

The delayed payoff of basic science

Almost none of what flowed from the room was foreseen by anyone in it. Dirac was not thinking about medical scanners, nor Heisenberg about satellite navigation. Yet a large share of the technologies that define modern life depend, at least in part, on quantum mechanics.

The laser rests on stimulated emission, which Einstein described in 1917 and which waited more than forty years for a working device, and also on decades of optics and materials work. Magnetic resonance imaging reads the quantum spin of protons and depends as much on superconducting magnets and computing. The Global Positioning System needs quantum atomic clocks and Einstein's relativity together: its satellite clocks run fast by about thirty-eight microseconds a day, and without the correction, navigation would drift kilometers within hours. An optical fiber is itself largely a triumph of classical electromagnetism, glass chemistry, and manufacturing; what is unmistakably quantum are the lasers and photodetectors at each end that launch and read its light. The blue light-emitting diode and the flash memory that stores charge by quantum tunneling are more directly quantum devices. Photolithography, which prints circuit patterns finer than a wavelength of light, is mostly wave optics, chemistry, and precision engineering — but its whole purpose is to fabricate the quantum-dependent transistors that fill a smartphone or a data center. The semiconductor fab, and the racks of chips in the buildings that now train and serve AI, are where the abstractions of 1927 are finally cashed out at industrial scale. Even quantum cryptography, which uses nonorthogonal quantum states — and, in some protocols, entanglement — to expose interception, is engineering layered on the physics.

Running through all of it is a pattern worth stating plainly, because it is one of the least intuitive facts about how technological societies actually work. The advances that most reshape ordinary life are often the delayed payoff of abstract questions asked decades earlier by people who could not have named the application. The lag from pure idea to visible use is routinely thirty to seventy years. This is not a failure of efficiency; it is a feature of depth. Understanding is general precisely because it is not aimed at a use, and its generality is what lets it seed uses no one imagined. The awkward corollary is that basic research is chronically underfunded, because whoever pays for the question is almost never the one who collects on the answer. Ernest Solvay paid for a conference on electrons; the return went to a semiconductor industry two generations away.

What the room suggests

It is tempting to draw crisp maxims from all this, and the crispest ones are usually the least defensible. A few sturdier observations remain.

Fundamental questions and incremental improvement are different activities, and a healthy field needs both. Refining a working system compounds real value and should not be sneered at; it rarely changes the base of what is possible. Reorganizing what counts as a question — as the Born rule and the exclusion principle did — changes that base, and little else does. Most effort sensibly goes to refinement; the rarer institutions that also protect room for the base-changing kind are the ones still spoken of a century later.

The frontier is bottlenecked less by the number of clever people than by shared context and honest friction, which is why the productive unit is usually a small core rather than a crowd — though, as Bell Labs and Fairchild show, that core almost always sits inside a larger institution that makes its work possible. Most of what keeps a modern life running, meanwhile, is invisible settled science, load-bearing and unnoticed: the fate of every deep idea that succeeds is to become reliable enough to be forgotten into the foundations. And there is a matter of humility about timescales. The uncertainty principle scandalized its first audience and is now a homework problem. The frontier's slow work is to turn what is currently unthinkable into what is eventually routine, and the fact that a genuinely frontier argument sounds baffling or beside the point at the time is not evidence against it. It is closer to the normal case.

If the photograph were taken today

If we tried to stage the equivalent portrait for our own moment — the small world now arguing about a powerful artifact whose behavior has outrun the theory of it — the exercise would immediately reveal how much the structure of such work has changed.

The people you would want in the frame were, in the first half of the 2010s, spread across two organizations and a few university labs. Google Brain, started in 2011 by Andrew Ng, Jeff Dean, and Greg Corrado, was where deep learning became industrial. DeepMind, founded in London in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman and bought by Google in 2014, stated the ambition of general intelligence most baldly. Geoffrey Hinton, whose long insistence on neural networks had just been vindicated by his students' 2012 image-recognition result, joined Google in 2013; one of those students, Ilya Sutskever, would help start OpenAI in 2015; Dario Amodei joined it the following year, from Google. And in 2017 eight researchers at Google — Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin — published the paper introducing the transformer, the architecture nearly every large language model since has been a variation on.

<figure> <img loading="lazy" src="/images/modern-solvay-concept-2026.jpg" alt="A synthetic illustration arranging fifteen modern AI researchers in a group portrait loosely echoing the 1927 Solvay photograph." /> <figcaption>Figure 2 — An AI-generated conceptual illustration (2026), not a photograph of a real meeting. Fifteen figures evoking a "modern Solvay" across Google Brain, DeepMind, OpenAI, Anthropic, universities, and the 2017 transformer collaboration; the likenesses are illustrative, not an authoritative identification.</figcaption> </figure>

The first thing the exercise reveals is that no such photograph exists, and could not, because the modern structure is not a conference but a network that promptly fissioned. The transformer's authors scattered to found or lead new labs — Cohere, Character.AI, and others. Sutskever helped build OpenAI and then left to start his own company; Amodei left OpenAI to found Anthropic; Suleyman went from DeepMind to Inflection to Microsoft; Karpathy passed through OpenAI and Tesla and back. The diaspora is the story, much as Fairchild's departures were the story of Silicon Valley. The concentrated talent no longer holds still to be photographed; it moves through institutions, spins out new ones, and increasingly spills into independent researchers and open-source collectives working outside any single lab.

One connection between the two images is no longer merely rhetorical. In 2024 the Nobel Prize in Physics went to John Hopfield and Geoffrey Hinton for the foundations of machine learning with neural networks, and the Chemistry prize went, in part, to Demis Hassabis and John Jumper for using such networks to predict protein structure. Two figures who belong in any modern portrait already hold Nobels, in physics and chemistry — the same two prizes that saturate the 1927 picture. The descendants of the room are being honored with the room's own awards.

The easy parallel is the wrong one. That version — quantum mechanics was mysterious and neural networks are mysterious too — flatters the analogy by blurring the kinds of mystery. Quantum mechanics had exact mathematical structure and made precise, repeatedly confirmed predictions; its puzzle was ontological, about what a well-defined theory meant for the nature of reality. The uncertainty around neural networks is of another kind: about their learned internal representations, about when and why they generalize, about how to evaluate them, and about their effects on the world. Mechanistic interpretability rhymes with the old search for a meaning behind a working formalism, but it is not a foundational crisis in physics, and "this generation's Born rule" is a provocation rather than an equation.

The resemblance that does hold is narrower and more useful. Both episodes combined a new theoretical object — the wavefunction then, the trained network now — with rapidly improving tools, an unusual concentration of young talent, a few large institutional patrons willing to fund it, and a live argument about whether spectacular predictive success amounts to genuine understanding. That last question is the real through-line from Brussels to the present. In 1927 it read: does a formalism that predicts every spectral line tell us what an electron is? Today it reads: does a system that predicts the next word well enough to reason and converse understand anything? Neither field has settled its version, and in both the argument has done more work than the consensus.

One difference between the two settings is not flattering to ours, though it should not be drawn too sharply. Solvay was an invitation-only meeting, closed to the public — but its formal reports and discussions were published, converting part of an elite private gathering into a durable public record. The frontier labs of the 2020s — OpenAI, Google DeepMind, Anthropic, Meta, Microsoft Research, xAI — have all published influential work, and their openness varies by era, organization, and subject; the trend, though, as the stakes and the capital have risen, is toward disclosure only in a short "system card," with the sharpest internal disagreements kept as trade secrets. The value of publishing was never that it settled things — the interpretation of quantum mechanics is unsettled to this day — but that the confusion became public, and so could be worked on by everyone able to see it. Whether a field can come to understand its own most powerful artifacts while the groups who understand them best compete and largely stay quiet is a real and open question.

Can the conditions be designed?

Which returns us to the question under the whole essay, and to a distinction the legend blurs. It is easy to say a meeting of thirty people reshaped civilization; it is truer, and more interesting, to be exact about what the meeting did and did not do. Cancel the 1927 conference and matrix mechanics, wave mechanics, and the transistor almost certainly still arrive — most of the foundational work was finished before the participants reached Brussels, and much of the decisive work came after. What the meeting did was compress a distributed revolution into a shared space: rival formalisms became mutually legible, weaknesses were exposed in days rather than years, and the disagreements entered the permanent record. A gathering of this kind changes history less by manufacturing ideas than by raising the rate at which existing ideas collide, combine, and harden into common infrastructure. The people reshaped the world, and so did the wider network they belonged to; the conference synchronized and recorded that network. That is a real effect, and a more modest one than the myth.

So: was Solvay designed, or did it simply happen? Some of both, and the proportion is the useful part.

The conditions were designed, and they are repeatable. A curator chose participants by contribution rather than rank. The size was kept small enough to keep the friction productive. The problem was single, shared, and genuinely open. The funding carried no product and no deadline, so the room could pursue understanding rather than advantage. And the format put rivals in one place and then published what they said, turning private insight into common property. None of that requires genius to arrange; it requires taste, money, and patience.

What could not be designed was the arrival of the ideas. You cannot schedule a Heisenberg or commission a Schrödinger equation. So the realistic aim is not to manufacture breakthroughs but to build the conditions in which breakthroughs, whenever they happen to come, are amplified rather than wasted. That is a modest-sounding goal and a powerful one.

The part most within deliberate control is the funding, and its source has shifted every generation while its function stayed the same. Solvay was private philanthropy. Bell Labs was the research dividend of a regulated monopoly, whose guaranteed telephone revenues insulated parts of its research program from immediate product pressure. The early internet and much early semiconductor work were government and defense. Fairchild's formation and diaspora helped shape the Silicon Valley venture-capital model. Today's frontier runs on venture money and the balance sheets of a few very large firms. Each era's version of the room was paid for by that era's surplus, spent by whoever was willing to fund the question rather than the product. The worry for our own moment is that the surplus is vast but aimed almost entirely at a product race, while the two institutions that best funded open, patient, published inquiry in the last century — the philanthropic convening and the monopoly lab with a mandate to publish — are both weakened. There is also a plainer point the 1927 picture makes unavoidable: twenty-eight men and one woman. Whatever else it records, it records a talent pool drawn from a sliver of the people who might have filled it. The contributors the age simply never looked for represent one of the largest recurring wastes in the history of ability, and widening that pool is not a decorative afterthought to building better rooms but among the highest-return moves available.

The photograph, again

Return to the picture on the steps. To almost anyone it is an old group portrait, indistinguishable from a thousand others. Read against what came after, it is closer to a diagram of much of the modern world, drawn before any of it existed. The people in it were not, in the moment, aware of doing anything grander than arguing about ψ; some contemporaries spoke as though the basic framework of physics was nearing completion. They were instead near the start of something, and the part of what they were doing that mattered most did not look like much: a small number of people who understood a hard problem deeply, disagreeing in one another's presence, with much of the formal argument later published, and no machine anywhere in the frame.

That is the reason to keep an eye on the unglamorous version of the same scene now. Somewhere a comparable group is doing comparable work, and the record of it will be a phone photo at an offsite, or a list of names on a paper, long before anyone calls it historic. The people in it may not yet know what they have. Most of the rest of us will look past it, the way a passerby in 1927 would have walked past twenty-nine formally dressed academics without a second glance. The photographs that turn out to matter rarely announce themselves. The most we can do is get a little better at guessing which ones to keep.

Notes and sources

This essay follows mainstream history and physics while trying not to convert interpretation into settled fact — a temptation the subject invites. A few points where the popular telling misleads, corrected in the text:

The Born rule fixes the probabilities of measurement outcomes; it does not by itself establish that ψ is physical, that particles lack positions before measurement, or that the world is indeterministic. Those are interpretive questions, still open.
Bell's theorem and the experiments that followed exclude local hidden-variable theories under standard assumptions, not determinism as such; the nonlocal, deterministic pilot-wave (de Broglie–Bohm) theory survives.
Quantum mechanics was not "finished" in 1927: its nonrelativistic formalism had largely taken shape, but Dirac's relativistic equation (1928), quantum electrodynamics, and quantum field theory came later.
The vivid breakfast-by-breakfast Einstein–Bohr narrative rests largely on later recollection; the official 1927 proceedings record little of it, and the photon-box episode belongs to the 1930 council.
Figure 2 is a synthetic illustration, not a photograph of any real meeting or team.

The conference, its venue, and its reconstruction

Nobel Prize photo gallery, 1927 Solvay Conference — the original photograph and a full identification key for all twenty-nine participants. Figure 1 reproduces a later digital colorization of Benjamin Couprie's original (colorization source unknown).
Guido Bacciagaluppi & Antony Valentini, Quantum Theory at the Crossroads: Reconsidering the 1927 Solvay Conference (Cambridge University Press, 2009) — a full English translation of the proceedings and a corrective to the myth that the "Copenhagen interpretation" simply triumphed at Solvay.
Institut International de Physique Solvay, Électrons et photons (Gauthier-Villars, 1928) — the official proceedings, with the reports by Bragg, Compton, de Broglie, Born & Heisenberg, and Schrödinger; International Solvay Institutes.
On the Institut de Physiologie in the Parc Léopold as the venue: Université libre de Bruxelles institutional history.

Foundational figures and their prize-winning work (linked to the Nobel Prize record)

Max Planck — the quantum of action (1900).
Albert Einstein — light quanta (1905) and stimulated emission (1917), the laser's theoretical root.
Niels Bohr — the quantized atom (1913) and complementarity (the Como lecture, 1928).
Louis de Broglie — matter waves (1924 thesis).
Werner Heisenberg — matrix mechanics (1925) and the uncertainty relations (1927).
Erwin Schrödinger and Paul Dirac — wave mechanics (1926) and the relativistic electron (1928).
Max Born — the probability rule (1926; the |ψ|² correction added in a footnote in proof).
Wolfgang Pauli — the exclusion principle (1925).

The debate and its experimental resolution

Einstein to Born, 4 December 1926 ("…He does not play dice"), in The Born–Einstein Letters (Macmillan, 1971) — find at the Internet Archive.
Niels Bohr, "Discussion with Einstein on Epistemological Problems in Atomic Physics," in P. A. Schilpp, ed., Albert Einstein: Philosopher-Scientist (1949) — find at the Internet Archive.
Einstein, Podolsky & Rosen, Physical Review 47, 777 (1935).
J. S. Bell, "On the Einstein Podolsky Rosen Paradox," Physics Physique Fizika 1, 195 (1964).
David Bohm, Physical Review 85, 166 (1952) — the deterministic, explicitly nonlocal completion of de Broglie's pilot wave.
On quantum key distribution — BB84 uses nonorthogonal states, with entanglement-based protocols also possible: Shor & Preskill, Physical Review Letters 85, 441 (2000).
The 2022 Nobel Prize in Physics (Aspect, Clauser, Zeilinger), on the exclusion of local hidden variables.

The compounding chain

The transistor — Nobel Prize in Physics 1956; the integrated circuit — Nobel Prize in Physics 2000; and the Nobel history of twentieth-century physics on the quantum basis of band gaps and transistors.
Claude Shannon, "A Mathematical Theory of Communication," Bell System Technical Journal 27 (1948).
Krizhevsky, Sutskever & Hinton, ImageNet Classification with Deep Convolutional Neural Networks (AlexNet, NeurIPS 2012); Vaswani et al., "Attention Is All You Need" (2017); Brown et al., "Language Models Are Few-Shot Learners" (GPT-3, 2020).
On Fairchild and the venture-capital model: Computer History Museum, "The Next New Thing". On the internet's evolution beyond ARPANET: Internet Society, "Brief History of the Internet".
The 2024 Nobel Prizes: Physics — Hopfield & Hinton; Chemistry — Baker, Hassabis & Jumper.

Also cited

Max Planck, Scientific Autobiography and Other Papers, trans. Frank Gaynor (Philosophical Library, 1949) — the passage that a new scientific truth "triumphs because its opponents eventually die." Find at the Internet Archive.

Trust Architecture in the Digital Age

Sat, 01 Nov 2025 00:00:00 GMT

We didn't lose trust in 2008.

We discovered it was in the wrong place.

I. One Weekend in September

At 1:45 on the morning of Monday, September 15, 2008, lawyers acting for Lehman Brothers Holdings filed a Chapter 11 petition in the Southern District of New York. The document listed $639 billion in assets, which made it the largest bankruptcy in American history — bigger than WorldCom, bigger than Enron, bigger than both combined. It had been assembled in roughly the time it takes to plan a dinner party.

The dinner party itself had happened two blocks away. Over that weekend, the chief executives of every major firm on Wall Street had been summoned to the fortress of the Federal Reserve Bank of New York on Liberty Street, where Treasury Secretary Hank Paulson delivered a message none of them had ever heard from a Treasury Secretary before: there would be no bailout. Find a private solution or watch the fourth-largest investment bank in America die in public.

One man was conspicuously not in the building: Dick Fuld, Lehman's chief executive, the longest-serving CEO on Wall Street, a man who had spent forty years inside the firm and eight months insisting it was fine. He spent the weekend at Lehman headquarters on Seventh Avenue, working the phones, waiting for news of the rescue he was certain would come. On Sunday it came, briefly: Barclays would buy the firm. Then British regulators declined to bless the deal on a Sunday afternoon's notice, and it un-came. "The British screwed us," Fuld reportedly told his colleagues. It was a characteristic reading of events. The British had merely declined to catch a falling knife that Wall Street, the Federal Reserve, and the United States Treasury had all, in turn, declined to catch first.

What happened next is usually described as a panic. It is more precisely described as an epistemic collapse — the sudden, system-wide discovery that nobody knew anything.

Consider the week as a tick-tock.

Monday, September 15: Lehman files. Merrill Lynch, which understood it was next in line, has already sold itself to Bank of America in a deal negotiated in approximately forty-eight hours. The Dow falls 504 points.

Tuesday, September 16: The Federal Reserve — which had just let Lehman die on the principle that governments do not rescue investment banks — rescues an insurance company. AIG receives an $85 billion credit line, because a unit of a few hundred people in London has written credit protection on half the financial system and can't pay. The same day, something happens that frightens the professionals more than either headline: the Reserve Primary Fund, the oldest money-market fund in America, "breaks the buck." Its shares, engineered and marketed as never being worth less than a dollar, are declared worth 97 cents. The fund had held Lehman paper. Money-market funds were the corner of finance designed to be boring — the mattress under the mattress. Within days, investors pull hundreds of billions of dollars out of them.

By midweek, the interbank lending market — the circulatory system through which banks lend to each other overnight, the plumbing so reliable no civilian had ever needed to learn its name — freezes solid. The spreads that measure whether banks trust each other blow out to levels no model had contemplated. Banks with billions on deposit at the Fed decline to lend to other banks overnight. Overnight. The bankers are not being irrational. They are being precise. Each of them has just looked at their own balance sheet, seen what is actually inside it, and drawn the reasonable inference about everyone else's.

On every trading floor in the world, the same question, whispered and shouted in the same breath: Who can we trust?

The answer terrifies them: almost no one.

Banks don't trust banks. Investors don't trust the ratings agencies — those triple-A stamps were supposed to be bedrock, and they crumble like paper. Regulators don't trust their own models; the risk frameworks designed to detect systemic collapse cannot see the systemic collapse occurring in real time, on television. And the public, watching all of it, arrives at the obvious conclusion: trust none of the above.

I had a seat for that week, though I didn't know it at the time. In the autumn of 2008 I was an entrepreneur in Turkey, several thousand miles from Liberty Street, with no visibility into what was seizing up inside Wall Street's plumbing. What I had was a signed deal: the funds investing in us had committed — contracts executed, the closing dinner already held, the toasts already made. And then, that same season, all of them — every single one — stopped taking our calls. No renegotiation. No explanation. Just phones ringing into silence. Chuck Prince of Citigroup had said, the year before, that as long as the music was playing, you had to get up and dance. That autumn I learned what the sentence actually meant. When the music stops, it doesn't sound like a crash. It sounds like a phone no one answers.

The season left something behind, though. The people who went through it with me are still the people I build with — some of the best minds I have ever worked with were forged in those months, Murat Bayraktar above all, my partner through the hardest of those days. Years later, when Ben Horowitz, the co-founder of Andreessen Horowitz, published The Hard Thing About Hard Things — his account of building a business when there are no easy answers — we didn't read it as a business book. We read it as recognition. Someone had finally written down what those days felt like from the inside.

Nietzsche had put it down long before, in Twilight of the Idols: Was mich nicht umbringt, macht mich stärker — what does not kill me makes me stronger. A century of overuse has worn the line smooth, but that autumn gave it back its teeth. We walked out of the season stronger. We have walked out of every hard season since the same way.

II. The Wrong Question

Nearly every account of that autumn — and there are shelves of them — treats it as the story of trust dying. Congressional testimony mourned it. Ten thousand op-eds eulogized it. The Edelman Trust Barometer began its long ritual of confirming it, year after year, with financial services cementing itself at the bottom of every sector it measures, below oil companies, below social media firms — an achievement, if you think about it.

And yet here is the thing nobody's model predicted.

In the fifteen-plus years after the crisis of trust in finance, people did not retreat from finance. They accelerated into it. They invested through apps built by companies younger than their phones. They lent money to strangers through platforms. They paid each other through smartphones. They stored value in digital wallets, in code, in assets that did not exist in 2008. The volume and velocity of financial transactions run by ordinary people, through institutions nobody had heard of, reached the highest levels in human history — precisely while measured trust in financial institutions sat at record lows.

If you believe trust died in 2008, this is a paradox.

It is not a paradox. It is a migration.

Trust didn't vanish. It moved. From institutions to systems. From people to protocols. From opacity to transparency. From human judgment to mathematical verification. Everyone stood watching the front door for trust's return, while it quietly left through the back and took up residence somewhere else entirely.

The story of modern banking — and the story of whatever banking becomes next — is the story of that move. To understand it, you have to first understand the strange machine trust lived in for three hundred years.

III. A Machine That Ran on Not Looking

For centuries, financial trust moved through a simple transaction: you trusted the institution, and the institution made the risk disappear. The machine had three components. Two of them appeared in the brochure. The third was the actual machine.

The first was reputation. Banks accumulated trust the way wine accumulates age — slowly, expensively, and mostly by not being disturbed. The Bank of England was founded in 1694 to finance a war against France, and its core product, refined over three centuries, was never really returns. It was endurance. A bank that had survived Napoleon, two world wars, and a dozen panics was making an argument no startup could counter: we will still be here. Longevity was the credential. You cannot raise a Series A for having existed since 1694.

The second was regulation. Governments wrote rules — capital requirements, reserve ratios, lending limits, deposit insurance — that locked bank behavior into acceptable boundaries. Regulation was a state-backed guarantee that said: this institution will not blow itself up, and if it does, you will be made whole. The genius of the arrangement was that it removed character from the equation. You didn't need to trust the banker. You trusted the apparatus supervising him.

The third component was the one nobody said out loud: opacity.

This sounds like an accusation. It is actually a design specification. Banking, as practiced for three centuries, required information asymmetry. The bank knew what the customer didn't: which borrowers were creditworthy, what the portfolio actually held, how much leverage was hiding inside the structures. And the system's architects understood — correctly — that this was a feature. Because if depositors could see everything, in real time — the true leverage ratios, the derivative exposures, the counterparty webs — they would not wait politely for a crisis to arrive. They would run at the first ugly quarter. Opacity was the only thing standing between an informed public and a permanent, rolling bank run.

Read that again, because it is the perverse truth at the center of old finance: the system was stable because you couldn't see it. Trust was a function of not looking. The most successful confidence machine in history ran on a carefully maintained absence of information, and it worked — for three hundred years, through its design parameters, it genuinely worked. Institutions were mostly competent. Regulation was mostly adequate. The opacity was mostly benign.

Then conditions left the design parameters.

By 2008, the opacity that cushioned normal times had been compounded into something new. Mortgage-backed securities bundled thousands of loans into instruments so intricate that the banks holding them could not price them — not wouldn't, couldn't. Credit default swaps wove webs of obligation connecting institutions in patterns no regulator, no rating agency, and no counterparty had ever mapped. The people running the machine had lost the ability to see inside it, which meant the not-looking was no longer a choice. When Lehman failed, the cascade tore through channels that were invisible because they had been built invisible.

The system didn't fail despite the opacity.

It failed because of it.

Three centuries of trust architecture, exposed in a single week as beautiful in normal times and catastrophic outside them.

IV. The Nine-Page Rebuttal

Forty-six days after Lehman filed — on Halloween, of all days — a nine-page PDF appeared on an obscure cryptography mailing list read by a few hundred people. The author signed it Satoshi Nakamoto, a name attached to no face, no institution, no résumé, and, to this day, no confirmed human being. The title was almost aggressively boring: "Bitcoin: A Peer-to-Peer Electronic Cash System."

The timing was not subtle, and in case anyone missed it, the author made it permanent. When the Bitcoin network's first block was mined that January, embedded in it was a newspaper headline from that morning: The Times 03/Jan/2009 Chancellor on brink of second bailout for banks. A timestamp and an indictment, welded together forever into the foundation of the thing.

Strip away everything that came later — the manias, the crashes, the laser-eyed profile pictures — and the white paper's core proposition was a direct answer to the question echoing across the trading floors that September: Who can we trust? Satoshi's answer: no one, and it doesn't matter. Here is a financial system in which trust is placed not in institutions but in mathematics. Proof-of-work: a consensus mechanism making it computationally infeasible for any single party to rewrite the ledger. Open code. Public rules. Transactions anyone can verify. You don't need to trust a banker, a regulator, or a government.

You need to trust math.

It is hard to overstate how philosophically strange this was. In five thousand years of money — debt tablets, coins, notes, wires — every financial system ever built had required trust in some human institution at its center. This one didn't. Whether Bitcoin itself ends up as currency, as digital gold, or as a cautionary tale is genuinely beside the point. The deeper insight escaped immediately and cannot be recaptured: trust can be engineered into a protocol rather than vested in a person.

While the cryptographers were rebuilding trust from first principles, a less romantic revolution was running the same experiment with better user interfaces.

Venmo let you split a dinner check with a tap; you trusted the app, not a bank. Robinhood gave you commission-free trading; you trusted the interface, not a broker. TransferWise — now Wise — sent money across borders and did something banks had spent decades making sure nobody could do: it showed you the actual exchange rate, next to the actual fee, next to what your bank would have quietly taken. Its marketing was, in essence, a single move — here is exactly what was being hidden from you — repeated until it had moved billions.

Notice what the fintechs were selling. Not longevity — most were younger than the phones they ran on. Not regulatory pedigree. They traded the old architecture's entire asset base, centuries of institutional reputation, for one thing: you can see everything we do. The old model said, "Trust us — we've been here two hundred years." The new model said, "Trust us — watch."

DeFi then extended the Satoshi principle from money to finance itself. Aave. Compound. Lending protocols with no loan officers, no credit committees, no relationships — terms encoded in smart contracts, collateral locked algorithmically, interest rates adjusting in real time to supply and demand. The protocol is trustworthy not because anyone vouches for it, but because anyone can audit it.

Which reveals the master principle of the entire migration. The old architecture ran on a proposition so familiar nobody noticed how strange it was: you trusted the institution because you couldn't see inside it. The new architecture inverts it precisely: you trust the system because you can.

Transparency is not a feature of the new trust architecture.

It is the foundation.

V. Trust as Infrastructure

There is a structural difference between the old trust and the new, and it matters more than the technology.

Institutional trust was personal. You trusted your bank, your advisor, your branch manager. It was relational — accumulated slowly, over years of handshakes and repaid loans — and it was non-transferable. Move to a new city, start over. Your bank fails, your trust dies with it. Trust was a bilateral asset, expensive to build and impossible to move.

Systemic trust has different physics. You trust the protocol, not the operator. The code, not the company. If the protocol is sound, it does not matter who deploys it. If the smart contract has been audited, the deployer's reputation is irrelevant. Trust stops being a relationship and becomes a property of the architecture itself — transferable, verifiable, and available to strangers.

Run that change through the problem of financial inclusion and watch what it does.

Roughly 1.4 billion adults on Earth have no bank account. The standard explanations involve poverty and geography, but the mechanism is simpler: the old model required institutional trust as the price of admission. Credit history. Collateral. A banking relationship. The unbanked had none of the three, so the institutions' answer was no — not occasionally, but structurally, forever. The system was not failing to reach them. It was working as designed, and the design excluded them.

I did not learn this from a World Bank report. In 2012, in Istanbul, we founded ininal on the philosophy Y Combinator had made famous — pick one small, unglamorous problem and solve it completely for people nobody else wants — and the problem we picked was this: millions of people in Turkey no bank would open an account for. Not risky customers; invisible ones. Students, cash workers, the young, the informal — the structurally excluded. We gave them a prepaid card you could buy at a corner shop and load with cash. No branch, no credit history, no relationship manager, no questions asked. The institutions' trust said no, so we routed around it: we put trust on a shelf, next to the chewing gum, and sold it for the price of a card.

More than fifteen million of those cards were sold. And four million people did something more telling than buying one: they paid to keep one. ininal was the market's first subscription business, which meant that people every bank had refused were now not merely using trust — they were subscribing to it, renewing it, the way you renew a phone plan. The niche turned out not to be a niche. It was the opening move of an entire sector — the wallets, the payment companies, the digital banks that followed in Turkey all trace a line back to the same discovery: that trust could be packaged as a product instead of granted as a privilege. Between that card and today runs, compressed into a single market and a single working life, the whole migration this essay describes.

Under the new model, the price of admission is a smartphone and a signal. A person in Lagos can reach a lending protocol without knowing anyone at any bank, without a credit history, without collateral — without anything except an internet connection and the same mathematical guarantees extended to everyone else on the network. The question is no longer "Who will trust you?" The question is "Can you reach the protocol?"

That is what it means for trust to become infrastructure. Not a privilege accumulated. A utility accessed.

And yet the years that taught me trust could be a product also taught me that its oldest form still outperforms. Almost every weekend of the ininal days, I worked through the business plan with Deniz Devrim Cengiz — a man of rare character, finely educated, relentlessly self-improving, and my partner in everything but the cap table. Not for lack of trying: I offered stock options, advisory shares, anything that would put his name where his weekends already were, and the policy of the institution he worked for ruled out all of it. The most I could ever pay him was dinner. When Colendi was founded, he invested on day one. He carried what those weekends had built to places no plan of ours had drawn — two more banks, in Kuwait and across the region — and in 2022, when I asked him to come build ColendiBank, he did not say no. He joined the family as my co-founder.

Most of this essay argues that trust migrated from people to protocols. It did. But I cannot let the argument stand without recording its great exception: the best-performing trust position of my career has been a person, compounding quietly for more than a decade — acquired at the price of a few dinners.

VI. The Trilemma

Here is where it gets hard. Because the migration has one more stage, and it is the one nobody has architecture for yet.

AI agents are beginning to make financial decisions autonomously. Not screening applications for a human to review — deciding. And the trust problem they create is categorically different from anything in the story so far.

Hold the two versions of a mortgage side by side. A human loan officer approves yours: whatever else is true, you know a person reviewed the file, weighed your circumstances, exercised judgment. The judgment may be flawed, biased, or lazy — but it exists inside a chain of accountability you can grab at any link. Appeal to the supervisor. Complain to the regulator. Sue. Somebody signed.

Now an AI agent approves it — or declines it. You cannot appeal to its emotional intelligence; it has none. You cannot assume it understood your situation beyond what the data encoded. And here is the genuinely new problem: you may not be able to learn why it decided — because with large language models, the reasoning isn't fully transparent even to the people who built them. The fintech era's bargain was "trust us — watch." What do you do with a system you can watch but not understand?

This produces a trilemma. You want an agentic financial system to be three things: autonomous, so you get the efficiency that justified building it; transparent, so decisions can be understood and challenged; and accountable, so someone is responsible when it fails.

Any two are easy. Autonomy plus transparency, minus accountability: a system that operates in the open and answers to no one. Autonomy plus accountability, minus transparency: someone signs for decisions nobody can explain — accountability as theater. Transparency plus accountability, minus autonomy: the current system with better documentation.

All three at once is the governance problem of the next decade, and the solution is not to abandon a corner. It is to build a trust architecture sophisticated enough to hold all three — and it has four load-bearing components.

Embedded governance: compliance rules, risk limits, and ethical constraints encoded into the agent at the foundation, so the agent doesn't check compliance after the fact but is compliance-aware by construction. Continuous auditability: every decision logged with full provenance — inputs, reasoning chain, outputs, which constraints were binding. Human escalation: the system recognizes the boundary of its own competence and surfaces the hard cases to people before deciding, not after. And explainability: the agent articulates, in natural language, why it decided — a genuine trace of reasoning, not a post-hoc press release.

Together these form what I call the trust layer: a dedicated stratum of the agentic bank whose sole purpose is to keep the system safe, accountable, and trustworthy. In the seven-layer architecture for agentic banks that I develop in Banking.io, trust and governance sit at Layer 2 — above raw infrastructure, beneath everything else — so that every capability built on top operates inside constraints the trust layer defines. In the old architecture, trust was the byproduct of reputation. In the agentic one, it is a component you engineer — the load-bearing wall everything else hangs on.

VII. The Claim on Society

In 1900, the German sociologist Georg Simmel published The Philosophy of Money and buried inside it an observation that reads today like a specification: money is a claim upon society. Not upon a person. Not upon an institution. A claim on the entire social system's agreement that this thing has value. Money works because we agree it works.

Money, in other words, is crystallized trust.

Follow the logic one step further than Simmel could: if money is crystallized trust, then the architecture of trust determines the architecture of money. Trust vested in institutions gives you institutional money — bank deposits, government currency, regulated securities. Trust vested in protocols gives you programmable money — smart contracts, stablecoins, tokenized assets whose trustworthiness derives from code. And trust vested in intelligent systems gives you something newer still: agentic money. Not just programmable but managed. Not just executable but reasoned about. Not just conditional but adaptive.

The full arc, compressed: trust the banker, then the bank, then the software, then the protocol, then the intelligent system.

Each stage keeps the ones before it — we still need institutions, still need regulation, still need human judgment, and 2008 is a permanent lesson in what happens when any of them fails. But the center of gravity moves, and it moves in one direction only: from human judgment to systematic architecture, from personal relationships to transparent protocols, from institutional reputation to demonstrable trustworthiness.

The old social contract of money was two words: Trust us.

The new one is also two words: Verify us.

Trust — not because we say so, but because you can read the code, audit the trail, inspect the reasoning, and challenge the decision. Not because we have survived centuries, but because the architecture itself is engineered for trustworthiness, and the engineering is on display.

On the trading floors, on the morning of September 15, 2008, the question was: Who can we trust? It took fifteen years to see that it was the wrong question. The right one — the one Satoshi answered forty-six days later, the one the fintechs answered with interfaces and the protocols answered with math, the one agentic banking now has to answer at a higher level of difficulty than anyone before it — was never who.

It was what.

This is not the death of trust. It is trust, rebuilt — for a digital, programmable, agentic world.

Trust the system, not the institution.

Still Day One

Mon, 07 Jul 2025 00:00:00 GMT

This spring, on the fifth of May, ColendiBank accepted its first customers. After years of building, a real person opened a real account at a bank we built from nothing. The hardest number in the life of any company is not a million. It is one.

Peter Thiel calls this move zero to one—the leap from nothing to something that exists, which is categorically harder than going from one to many. In a regulated bank the leap is unusually literal. You cannot fake your way to the first customer. Every layer beneath that account has to be genuinely real: the capital, the core systems, the settlement rails, the risk and compliance functions, the operating license—and then, at the very top, one human being trusting you with their money.

The timeline is worth stating plainly, because it is the opposite of a growth-hack story. Turkey's digital-banking regulation was published on the 29th of December, 2021—the first day a bank like this was legally possible at all. We spent the following year preparing the application; our documents were filed and accepted through the end of 2022. The BDDK approved our establishment in August 2023, granted the operating license on the last day of October 2024, and we welcomed our first customers—the alpha cohort—on the fifth of May 2025. Something over three years from the day the rules existed to the day someone opened an account. For a bank built from scratch that is fast; that it still took three years is the honest price of building inside the rules rather than around them. That price is also the moat. Anyone can announce a neobank. Very few will do the unglamorous, multi-year work of becoming a licensed one.

What went live is the bank we described when we had only permission to build it: an AI-native, purely digital deposit bank—underwriting, service, and risk designed around models, with a person accountable for every automated decision. The thing that was a thesis is now holding deposits.

And this is the part we most want to hold onto: it is still Day One. Zero to one was the conceptually hard part, and we have done it. One to millions—earning trust at scale, keeping the magic legible, being worthy of a banking license every single day—is the actual work, and it has barely started. The first customer is not a finish line. It is the moment the real company begins.

Eight years ago we set out to give an invisible borrower a fair decision. Today someone we have never met keeps their money in a bank we first imagined and then, slowly and under real scrutiny, made real. From here the standard only gets harder—which is exactly as it should be.

To the team, our investors, and to our first customers: thank you for the trust. And one thank-you that is personal—to Deniz Devrim Cengiz, my partner and co-founder from the very first day: none of this happens without you. As we always say, “Niyet, gayret, kısmet.”

It's still Day One.

Money as Code — Why Programmable Money Changes Everything

Sun, 01 Jun 2025 00:00:00 GMT

For five thousand years, money has been a noun.

A coin. A note. A balance. A number on a screen. Something you have, something you hold, something you move from one place to another.

Money, in all its historical forms — shells, cattle, gold, paper, electronic digits — has been fundamentally passive. It sits. It waits. It does nothing until you command it.

Now money is becoming a verb.

Programmable money doesn't just move value. It executes economic relationships. It enforces conditions. It operates without you. This is the most significant monetary innovation since the invention of credit itself.

And it changes everything about what banking can be.

The Four Epochs of Money

To understand why programmable money matters, you need to understand what it supersedes.

Epoch One: Commodity Money

Gold. Silver. Salt. Cowrie shells. Objects with intrinsic value (or perceived intrinsic value) that served as media of exchange.

Commodity money solved the double coincidence of wants problem. Instead of needing someone who had what you wanted and wanted what you had, you could use a universally accepted medium.

But it was heavy. Indivisible. Insecure. Scarce.

It worked for millennia because nothing better existed.

Epoch Two: Credit Money

This was the revolution most people misunderstand.

David Graeber's anthropological research revealed the truth: credit, not barter, was the original form of exchange. The Mesopotamian clay tablets recording debts predate coined money by thousands of years.

A banknote was not gold. It was a promise of gold.

A bank deposit was not cash. It was a claim on cash.

Credit money abstracted value from physical objects. The entire modern financial system is built on promises layered on promises, obligations creating obligations. Genius and danger coexist here: you decouple value from physicality, but you concentrate trust in the institutions managing those promises.

Epoch Three: Digital Money

Banks computerized their ledgers.

SWIFT connected international payments.

Online banking replaced branch visits. Mobile apps replaced checks.

Money became electronic. Faster. More accessible. Available 24/7.

But — and this is the critical point — digital money did not change what money is. A digital dollar is still a claim on a bank. A mobile payment still settles through the same interbank clearing systems that processed paper checks.

Digital money is traditional money with a better user interface.

The noun unchanged. Only the adjective changed.

Epoch Four: Programmable Money

Now we are entering the fourth epoch.

Unlike the transition from physical to electronic, this is not a change of medium.

It is a change of nature.

I can date my own entry into the fourth epoch precisely. In 2018 we published the first Colendi technical paper — a protocol for decentralised credit scoring, an attempt to write the prerequisites of money as code: identity and trust. In April 2019 the apps went live, and a person could carry a blockchain identity in their pocket and generate a credit score from the digital exhaust of their own life — their phone, their purchases, their patterns — a score that belonged to them, not to a bureau. We did not call it programmable money then. It was programmable trust, and it taught us the fourth epoch's first lesson early: before value can execute, identity and trust have to compile.

What Makes Money Programmable

A piece of money becomes programmable when it can embed and execute logic.

The Old Way

You instruct your bank to transfer $1,000 to a supplier.

The bank debits your account. Routes the payment through an interbank network. Credits the supplier's account.

The money moved. But the money itself did nothing. The banks, the networks, the clearing systems performed the work.

The money was passive cargo.

The New Way

You encode a condition into the money itself: "Transfer $1,000 in USDC to Supplier X when the shipment tracking system confirms delivery at Warehouse Y."

The money sits in escrow — not at a bank, but in a smart contract. It evaluates the condition continuously. When delivery is confirmed, it executes the transfer automatically.

No bank intermediary. No manual confirmation. No three-day settlement period.

The money did something. It evaluated a condition and acted.

This is the simplest possible example. The design space is enormous.

Streaming Payments

Instead of paying an employee $5,000 at the end of the month, value flows continuously. $0.11 per minute. Every minute. In real-time.

The employee has access to earned income the moment it is earned, not weeks later.

Conditional Credit

A loan where the interest rate adjusts in real-time based on the borrower's cash flow.

Revenue strong? Rate drops. Borrower accelerates repayment.

Revenue dips? Rate adjusts downward. The burden lightens.

The terms are not negotiated once and frozen. They are dynamic. Embedded in the money itself.

Parametric Insurance

A farmer's crop insurance that pays out automatically when rainfall falls below a threshold.

No claim form. No adjuster. No dispute.

The contract observes the data. It executes. The farmer receives payment within hours of the drought, not months later.

Composable Financial Instruments

Stack a stablecoin yield protocol on top of tokenized Treasury bills on top of a liquidity pool.

Each layer interacts with the others programmatically, optimizing returns based on market conditions, rebalancing automatically.

The financial product is not designed by a banker and sold to a customer. It is assembled from composable primitives by an algorithm.

The Qualitative Leap

It is tempting to see programmable money as merely an improvement.

Faster payments. More efficient lending. Lower costs.

This misses the point entirely.

The transition from digital to programmable money is not quantitative. It is qualitative. It is the difference between a photograph and a camera.

A photograph captures a moment. A static representation of reality.

A camera acts. It perceives light. Focuses. Adjusts exposure. Creates.

Digital money is the photograph: static representation of value.

Programmable money is the camera: an active system that perceives conditions and creates outcomes.

The Return of Quality

Georg Simmel wrote in 1900 that money is the instrument that reduces all quality to quantity. Everything becomes measurable. Comparable. Exchangeable. Money flattens and equalizes everything it touches.

Programmable money reverses Simmel.

It reintroduces quality into money.

A programmable dollar is not the same as every other programmable dollar. It carries conditions. Purposes. Constraints. Behaviors.

A dollar earmarked for healthcare spending that can only be spent at approved providers is qualitatively different from a dollar in a savings account earning yield.

A dollar in a smart contract that releases to a contractor upon project completion is qualitatively different from a dollar in a wire transfer.

Programmable money creates what Viviana Zelizer called "special monies" — except the specialness is not cultural or social. It is encoded.

The Policy Implications

Central bank digital currencies could implement targeted monetary policy.

Imagine: stimulus payments that can only be spent (not saved) within 90 days. Only at approved merchant categories. Only in designated geographic areas.

Whether this is desirable is a political question.

That it is now technically possible is an economic revolution.

The Three Properties of Programmable Money

Programmable money has three properties.

Individually, each is significant.

Together, they are transformational.

Property One: Conditionality

Money can be made contingent on external events. Data feeds. Logical evaluation.

If-then-else logic applied to value transfer.

This eliminates entire categories of financial intermediation. Escrow agents. Claims adjusters. Custody services. Compliance officers.

These jobs exist largely because traditional money cannot enforce its own conditions.

Programmable money can.

Property Two: Composability

Programmable money instruments can be assembled and recombined like software libraries.

Take a stablecoin. Wrap it in a yield-generating protocol. Attach it to a lending pool. Connect it to an insurance contract. All through standardized interfaces.

Each component interacts with the others programmatically.

This is what the DeFi ecosystem demonstrated in 2020 and 2021. Hundreds of novel financial products were assembled from composable primitives by developers with no banking licenses. No regulatory approval. No relationship with traditional financial institutions.

Some were reckless.

Many were innovative.

All demonstrated the creative potential of composable finance.

Property Three: Autonomy

Programmable money can operate without human intervention.

Once deployed, a smart contract executes its logic regardless of whether anyone is monitoring it.

It does not take holidays.

It does not make errors of attention.

It does not succumb to conflicts of interest.

It does what it was programmed to do. Relentlessly. Precisely.

This autonomy is both the great strength and the great risk of programmable money. Managing that tension is the central governance challenge of the agentic banking era.

What Programmable Money Means for Banking

If money can enforce its own conditions, the role of the bank changes fundamentally.

Why Banks Exist

Traditional banking exists because of a structural problem: transacting parties cannot trust each other directly.

You don't hand cash directly to a stranger for a house.

You use a bank that holds the escrow. Verifies the title. Processes the mortgage. Ensures settlement.

The bank's value proposition is not the money itself.

It is the trust. The process. The middleman standing between two parties who don't know each other.

The New Architecture

Programmable money encodes trust and process into the money itself.

The escrow is a smart contract.

The verification is an oracle that checks the title registry.

The mortgage terms are embedded in a self-executing loan.

The settlement is atomic — it either completes in full or does not execute at all.

The bank's traditional functions — custody, escrow, settlement, compliance enforcement — become features of the money rather than services of the institution.

From Intermediary to Infrastructure

This does not mean banks disappear.

It means their role shifts from intermediation to infrastructure.

The bank provides the regulated on-ramp (converting fiat to programmable money). The compliance framework (ensuring that programmable instruments meet regulatory requirements). The balance sheet (backstopping liquidity and absorbing risk).

But the high-margin, customer-facing functions — the advisory, the product structuring, the relationship management — migrate to platforms and AI agents that operate on the programmable money layer.

The Transport Layer, Not the Destination

Programmable money is the transport layer of the new financial stack.

Not the application layer.

It is enormously important — you cannot build the agentic bank without it.

But it is not the final destination.

The final destination is intelligent systems that use programmable money to execute increasingly sophisticated financial operations.

The Risks We Cannot Ignore

No honest assessment of programmable money can avoid its risks.

The Terra/Luna Catastrophe

May 2022: An algorithmic stablecoin collapses.

Money that was supposed to maintain a dollar peg through code rather than reserves loses its peg. Spirals into a death loop.

Approximately $40 billion in value destroyed.

In 72 hours.

The code did exactly what it was programmed to do.

The programming was wrong.

Smart Contract Vulnerabilities

DeFi hacks of 2020-2023 revealed a fundamental problem: smart contracts, once deployed, are immutable attack surfaces.

A bug in a smart contract is not like a bug in a mobile app. You cannot push a patch. Cannot roll back the update.

The code runs. The exploit executes. The funds are gone.

Over $3 billion was lost to DeFi exploits in 2022 alone.

Surveillance and Control

The privacy implications are equally sobering.

When money carries logic, it also carries information.

A fully programmable monetary system could enable surveillance of every transaction. Every financial relationship. Every spending decision.

China's digital yuan experiment has already raised concerns about the state's ability to monitor and control citizen behavior through programmable currency features.

The Governance Imperative

These risks are real. They demand a governance architecture as sophisticated as the technology itself.

Programmable money without governance is a weapon.

Programmable money with governance — embedded compliance, risk guardrails, human oversight, transparent audit trails — is a revolution.

The Inevitability of Programmable Money

Despite the risks, programmable money is not a speculative future.

It is an accelerating present.

The Evidence

Stablecoins — the most widely adopted form of programmable money — represent a market capitalization exceeding $150 billion. They process more transaction volume annually than PayPal.

BlackRock, the world's largest asset manager, has launched a tokenized money market fund on blockchain infrastructure.

JPMorgan has built Onyx, an institutional platform for programmable payments and tokenized assets.

The European Central Bank is developing a digital euro.

Brazil's central bank has launched Drex, a wholesale CBDC designed for programmable financial operations.

Over 130 countries are exploring central bank digital currencies. Most with programmable features.

I am not reading this evidence from the outside. In 2022 Colendi acquired SETL — the London ledger-technology company that was one of only two providers, alongside Digital Asset, chosen to build the Regulated Liability Network proof of concept with the Federal Reserve Bank of New York's Innovation Center, working with BNY Mellon, Citi, HSBC, Mastercard, PNC, Swift, TD Bank, Truist, U.S. Bank, and Wells Fargo. The published conclusion of that exercise deserves to be more famous than it is: 24/7 atomic settlement in dollars is achievable with legal finality under existing law — no new statute required. One number from the report stayed with me: moving settlement from Federal Reserve business days to around-the-clock adds 117 days of settlement availability a year. A third of a calendar, unlocked by code.

That codebase — productised as LedgerSwarm, already inside the UK government's DLT pilot and Singapore's cross-border experiments — now runs under our roof. ColendiBank, Türkiye's first AI-native digital deposit bank, holds the BDDK licence, sits on the SWIFT network, and operates as an endpoint on Circle's Payments Network. And we have put a reference architecture on the table with the Central Bank of the Republic of Türkiye for the Digital Turkish Lira sandbox.

Programmable money stopped being a thesis for me somewhere along that road. Today it is a licence, a BIC code, and a sandbox deadline.

The Transition Is Underway

The infrastructure is being built.

The standards are emerging.

The regulatory frameworks are forming.

The only question is how quickly — and how wisely — the financial system makes the transition.

For five thousand years, money has been a noun.

A thing to hold. A number to move. A balance to check.

The most radical transformation in the history of finance is not the invention of a new kind of money.

It is the transformation of money from object to instrument.

From something that is to something that does.

When money can think, banking must evolve.

When money can act, intermediaries must transform.

When money can compose, assemble, and execute, the entire architecture of financial services must be reconceived from the ground up.

This is not a fintech trend.

This is a civilizational shift in the nature of economic infrastructure.

And it has already begun.

When money can think, banking must evolve.

The Death of Traditional Banking

Fri, 01 Mar 2024 00:00:00 GMT

"Banks don't die from competition. They die from irrelevance."

September 15, 2008. A Monday.

Lehman Brothers — the world's fourth-largest investment bank, 158 years old, 25,000 employees, $639 billion in assets — ceased to exist. Not gradually. Not over quarters of decline. Over a weekend.

The phone calls had started on Friday. By Saturday evening, the Treasury Secretary was in a conference room at the New York Fed, surrounded by the CEOs of every major Wall Street bank. By Sunday night, it was over. No buyer. No bailout. No rescue.

Monday morning, Lehman employees walked out of 745 Seventh Avenue carrying cardboard boxes. Some were crying. Some were stunned. Some were already calculating how long their severance would last.

Within months, the global financial system was on life support. Governments spent trillions to keep banks alive.

And somewhere in the wreckage, a pseudonymous programmer named Satoshi Nakamoto published a nine-page paper describing a peer-to-peer electronic cash system that required no banks at all.

The message embedded in Bitcoin's first block was not subtle: "The Times 03/Jan/2009 Chancellor on brink of second bailout for banks."

The 2008 crisis did not just break banks. It broke the idea that banking requires banks.

The Three Pillars

Traditional banking rested on three structural advantages that made it, for centuries, almost impossible to disrupt.

The first was physical infrastructure.

Banking required buildings. Vaults to store gold. Branches to serve customers. Trading floors to execute orders. This was not just operational convenience — it was the architecture of trust itself. You trusted your bank because you could walk into it. You could see the marble floors, the steel vault, the suited professionals behind mahogany desks.

The building was the brand.

In 1994, there were over 95,000 bank branches in the United States. A new competitor needed billions in capital just to establish a physical footprint large enough to be taken seriously.

The second was information asymmetry.

Banks knew things that customers and competitors did not. Proprietary credit data — payment histories, income verification, default records — gave them exclusive insight into who was creditworthy and who was not. This informational advantage was the foundation of the lending business. Without access to the data, no outsider could compete.

Knowledge is power. And banks had a monopoly on financial knowledge.

The third was regulatory moats.

Banking licenses. Capital requirements. Compliance regimes. Deposit insurance schemes. Barriers to entry that protected incumbents for decades. Obtaining a banking charter required years of regulatory engagement, tens of millions in legal fees, and enough capital to satisfy reserve requirements.

The regulations designed to protect consumers simultaneously shielded existing banks from competition. It was a fortress. And the regulators — inadvertently — were the guards.

Together, these three pillars made banking one of the most durable business models in human history. Banks survived wars, revolutions, depressions, and hyperinflation.

They seemed permanent.

They were not.

How Technology Dissolved Each Pillar

Each pillar was hollowed out by a specific technological shift. Not simultaneously. Not catastrophically at first. But with a compound force that, by the mid-2020s, left traditional banking structurally weakened.

The smartphone killed the branch.

It did to bank branches what the automobile did to the horse stable: it did not make them illegal — it made them unnecessary. When you can open an account, transfer money, apply for a loan, and invest in securities from a device in your pocket, the marble lobby becomes a cost center, not a competitive advantage.

In the United Kingdom, bank branches fell from 20,583 in 1986 to fewer than 8,000 by 2023. In Brazil, Nubank — a bank with no branches at all — acquired over 100 million customers faster than any traditional bank in history.

The building, once the symbol of banking, became its anchor.

Brett King saw this before almost anyone. In 2011 he founded Moven — the world's first mobile bank account with in-app debit card signup — and later gave the era its slogan in the title of Bank 4.0: banking everywhere, never at a bank. The features Moven pioneered — the real-time spending receipt, the financial-health gauge on the home screen — were dismissed as gimmicks by incumbents, then quietly copied into nearly every banking app on Earth within a decade. And the ending is the most instructive part: pressed by the pandemic, Moven closed its consumer bank in 2020 and pivoted to selling its technology to other institutions. Even the first challenger became infrastructure. The features survived; the brand did not. Remember that pattern — this essay returns to it.

Data ubiquity killed information asymmetry.

The explosion of alternative data — social media behavior, mobile phone usage patterns, e-commerce histories, geolocation data, utility payments — destroyed the bank's monopoly on creditworthiness assessment. Ant Financial's Sesame Credit scored 450 million users using e-commerce and social behavior data that no traditional bank possessed.

Jack Ma had announced the intention in 2008, with the crisis still smoldering: "If the banks don't change, we'll change the banks." It sounded like bravado from an e-commerce salesman. It was a roadmap.

When everyone has access to data, the bank's unique advantage evaporates.

Policy innovation breached the regulatory moats.

Regulators themselves — often portrayed as guardians of the old order — began opening the gates. The EU's PSD2 directive mandated that banks share customer data with licensed third parties via APIs. Open banking regimes spread from Europe to Australia, Brazil, India, and beyond. Regulatory sandboxes allowed startups to operate under lighter supervision while they proved their models.

Banking-as-a-service providers enabled any software company to embed financial services without obtaining its own charter.

The fortress did not fall. It was opened from the inside.

The Trust Inversion

But the deepest blow was not technological. It was psychological.

For centuries, banking operated on a simple premise: you trust the institution. You trusted that your deposits were safe. You trusted that the bank's lending decisions were sound. You trusted that the system was managed by competent, honest professionals acting in your interest.

The 2008 crisis shattered that premise.

Not because every banker was corrupt — most were not. But because the system's complexity had outgrown its accountability. Mortgage-backed securities. Collateralized debt obligations. Credit default swaps. Off-balance-sheet entities. The financial system had constructed an architecture of risk so intricate that even its architects could not assess it.

The people who were supposed to understand the system were as surprised as everyone else when it collapsed.

The aftermath was devastating — not just economically, but epistemically. Financial services became the least trusted industry globally. A Bain study found that 71 percent of American consumers would trust a financial product from Amazon — a company that had never held a banking license — over an equivalent product from their own bank.

Think about that.

Trust did not disappear. It migrated. From institutions to technology. From bankers to algorithms. From regulation to transparency.

This is the deepest structural change in the history of finance. Not the shift from paper to digital, or from branches to apps. The relocation of trust itself — from human institutions to technological systems.

When a person trusts a transparent algorithm over an opaque committee, the fundamental social contract of banking has been rewritten.

What Remains Is Infrastructure

Strip away the branches. Strip away the information monopoly. Strip away the regulatory moat. Strip away the trust premium.

What is left?

Infrastructure.

The rails on which money moves. The ledgers that record transactions. The compliance frameworks that satisfy regulators. The settlement systems that finalize payments.

This is not nothing — it is enormously valuable. But it is a different business from what banking used to be.

It is the difference between being a hotel and being a building. The hotel has a brand, a guest experience, a loyalty program, a relationship. The building has walls, plumbing, electricity, and an address. Both are necessary. Only one captures the premium.

Traditional banks are becoming the buildings of finance. The guest experience — the interface, the intelligence, the personalization, the trust — is migrating to new entrants: fintech platforms, super-apps, embedded finance providers, and AI-driven systems that make the bank invisible.

This is not a prediction about some distant future. It is a description of the present.

When you use Apple Pay, you interact with Apple. The bank behind it is invisible. When you buy now and pay later with Klarna, you interact with Klarna. The lending institution is a white-label service in the background. When a small business in Kenya receives working capital through M-Pesa, the customer's relationship is with Safaricom, not the bank.

The pattern is consistent. The platform captures the customer relationship. The bank provides the regulated infrastructure. The platform earns the margin. The bank earns a utility fee.

The Absorption Thesis

Banks will not be disrupted in the traditional sense. They will not go bankrupt en masse or be replaced by startups.

They will be absorbed.

Into larger platform ecosystems. Becoming invisible components of a financial infrastructure stack.

Consider cloud computing. In the early 2000s, companies built and maintained their own data centers. The data center was a strategic asset. Then AWS, Azure, and Google Cloud emerged. Within a decade, the data center went from strategic asset to commodity utility. Companies did not stop needing compute. They stopped needing to own it.

Banking is following the same trajectory. Companies and consumers do not stop needing financial services. They stop needing to interact directly with the institutions that provide them. Banking becomes embedded. Invisible. API-accessible.

The bank is there. You just never see it.

This is not the death of financial services. It is the death of banking as an identity — as a brand, a relationship, a building you walk into, a name you trust. What replaces it is something more powerful and more distributed: a programmable financial infrastructure that anyone can build on, and that no single institution controls.

What Comes Next

Bill Gates said it in 1994: banking is necessary, banks are not. For three decades the line lived on conference slides as a provocation. What this decade did was quietly convert it from provocation to operating description.

The question is no longer whether traditional banking will survive in its current form. It will not.

The three pillars are gone. The trust has migrated. The infrastructure is being commoditized.

The question is what replaces it.

A fundamentally new architecture of finance. One built not on institutions but on programmable infrastructure. Not on human judgment but on artificial intelligence. Not on products but on platforms.

An architecture where money itself becomes code — executable, conditional, autonomous. Where financial services are assembled from composable layers rather than delivered by monolithic organizations.

This is not a fintech story. Fintech, at its best, has only digitized the surface of traditional banking — replacing the branch with an app, the paper form with a web form, the teller with a chatbot. The underlying model remains: a centralized institution intermediating between savers and borrowers.

The transformation I am describing is structural. It changes what banking is, not just how banking looks.

The death of traditional banking is not a tragedy. It is a metamorphosis.

What emerges is something more powerful, more accessible, and more aligned with the true nature of money — which has always been, as Georg Simmel understood more than a century ago, not a thing but a relationship.

The relationship is changing. The architecture must change with it.

The question is not whether traditional banking will survive. It is what replaces it.

Permission to Build a Bank

Sat, 05 Aug 2023 00:00:00 GMT

"Any sufficiently advanced technology is indistinguishable from magic." — Arthur C. Clarke

Today Turkey's banking regulator, the BDDK, granted permission to establish ColendiBank—a digital deposit bank, to be founded with two billion lira of capital and the ambition of competing in the global league of digital banks. After seven years of building around banks, we have been given permission to build one. This is establishment approval, not yet an operating license—the beginning of the hardest stretch of work, not the end of it.

<figure> <img loading="lazy" src="/images/colendi-bddk-2023.jpg" alt="A framed 'Never Stop Hacking' print, a red WIRED magazine issue titled 'the end of code', and a white Colendi mug on a marble desk." /> <figcaption>Cade Metz's WIRED cover—"the end of code," soon we won't program computers, we'll train them—beside a Colendi mug. The bank we're building sits on the other side of that sentence.</figcaption> </figure>

It is worth remembering where this started. When we founded Colendi in 2017, the question was narrow and stubborn: how do you give a person a fair credit decision when the system has never written them a credit file? We answered it with alternative data, machine learning, and an identity a person could carry. But every honest answer to that question runs into the same wall—the decision is only half the system. To act on it you need the other half: accounts, deposits, settlement, the regulated core where money is actually held and moved. We acquired a piece of that core last year, when SETL became part of Colendi. Today we are permitted to build the rest.

We have said for years that banking is necessary and banks are not—that the function matters and the incumbent does not. That is still true, and it is exactly why the license matters. You cannot rebuild banking as software from outside the perimeter forever; at some point, to own the layer where value is held and trust is regulated, you have to stand inside it, accountable to the same rules as everyone else. A charter is not a trophy. It is a responsibility, and a permission to be judged by a harder standard.

The bank we intend to build is not a traditional one with an app bolted on. It is designed to be intelligent from the first line—underwriting, service, risk, and segmentation built around models rather than branches, with a person accountable for every automated decision. Clarke's line is the promise and the warning at once: a bank that decides through machines can feel like magic to a customer, and our whole job is to keep that magic legible, fair, and answerable to someone. Advanced technology should be indistinguishable from magic in how it feels, and completely distinguishable from it in how it is governed.

There is a longer arc here I would rather not flatten into a press release. The distance from a credit score for the invisible to a bank of our own is the distance this whole body of writing keeps trying to describe: money becoming programmable, trust migrating into software, and the institutions that matter becoming the ones that control the layer where it all settles. Permission to build a bank is a milestone on that road, not the destination.

To the team that carried this for years, and to a regulator willing to let something new be built under real rules: thank you. Now the harder part begins.

Acquiring SETL, and a Pilot at the New York Fed

Tue, 15 Nov 2022 00:00:00 GMT

Today the Federal Reserve Bank of New York's Innovation Center announced a proof-of-concept our team has been quietly proud of for months: the Regulated Liability Network U.S. Pilot. Its technology partner is SETL—a company that, in June, became part of Colendi.

The acquisition surprised some people. Colendi built its name on credit scoring and embedded finance for people the traditional system could not see; SETL builds wholesale settlement infrastructure for the institutions at the very center of that system. Founded in London and led by Anthony Culligan and Philippe Morel, SETL had spent years building distributed-ledger settlement systems capable of moving regulated money between banks and market infrastructures. To us the two were never far apart. They are the two ends of the same stack: one decides who can be trusted with credit, the other moves and settles the money once the decision is made. A financial system that is becoming programmable needs both, and we would rather own the rails than rent them.

The New York pilot is where that infrastructure meets its most demanding test. Over twelve weeks, the New York Innovation Center and a group of the largest U.S. banks—BNY Mellon, Citi, HSBC, Mastercard, PNC, TD Bank, Truist, U.S. Bank, and Wells Fargo—will explore whether central bank money and commercial bank deposits can be represented as tokenized "regulated liabilities" on a shared distributed ledger, and settled between institutions in a single, always-on, programmable network. SWIFT is supporting interoperability; Sullivan & Cromwell is advising on the law; Deloitte on the design. SETL, together with Digital Asset, is providing the technology sandbox.

It is worth being precise about what this is and is not. It is a research proof-of-concept, run in a test environment with simulated data. It is not a product, not the launch of a digital dollar, and not an endorsement by the Federal Reserve of any company or design. What it is, is serious: some of the most important financial institutions in the world sitting down to ask whether the plumbing of regulated money can be rebuilt on a common, programmable substrate—and testing the idea on infrastructure our team now owns.

That is the through-line of everything we write here. Money is becoming something that executes, and the institutions that matter will be the ones that control the layer where value actually moves and settles. Colendi began by trying to give an invisible borrower a credit file. Acquiring SETL, and standing behind a table at the New York Fed, is the same conviction carried to the other end of the system: banking is necessary, banks are not, and the settlement layer is worth building well.

Congratulations to Anthony, Philippe, and the entire SETL team.

Fintech Will Eventually Eat the Banks: Every Company Will Become a FinTech Company

Mon, 03 Aug 2020 00:00:00 GMT

In today's world, most of the companies and startups even those that do not have anything to do with financial services, have been increasingly taking the provision of alternative financial services to the individuals as a mission. Therefore, it would not be wrong to claim that every company will be a fintech company in the not-too-distant future. The distinguishing characteristic of the current period is the appearance of more choices, better products, and lower prices for consumers all around the world via the interplay between finance and technology sectors under the name of "fintech." In other words, digital transformation has been for a while changing the way existing finance and banking industry operates.

Over the last ten years, there have been new entrants to the banking ecosystem in the form of either competitors or enablers, and one of the main aims is to leverage innovative technologies to change the way financial services are provided to the low-income and unbanked populations.

Every company will be a FinTech company! Why?

Today, financial institutions are required to adapt to the fintech wave, or they have to die. The notion of adaptation is ranging from digitizing existing products and services to a complete digital transformation of processes and the introduction of new products and services.

To understand the reason behind such large-scale changes in the financial ecosystem in the digital age, first, it is of paramount importance to be aware of the deficiencies in the traditional banking industry. For example, according to the survey conducted by the World Economic Forum, only 28 percent of the millennial and Gen Z generations trust their banks to be fair and honest, which shows that the established banks are not successful enough to provide delightful products to their customers.

Furthermore, it is also possible to talk about a gradual reduction in investors' confidence in the banks as a result of slowing growth in volumes and top-line revenues in the market. For example, there was only 4 percent increase in the loan growth in 2018 — the lowest in the past five years and a good 150 basis points below nominal GDP growth, as can be seen in McKinsey company report.

Moreover, there has been a dramatic reduction in the global return on tangible equity (ROTE) from 20.0 percent in 2013 to 14.1 percent in 2018 because of the "digital disruption", which shows the difficulty for the banks to strengthen their productivity and manage the risk costs.

At that point, it is obvious that today, people all around the world need to get access to the alternative financial services and products, while ensuring the privacy and security of their data in the digital age. Until today, there have been fewer options, and those offerings are much more expensive. In that sense, the thorny question is as follows; why has the status quo survived until today, despite the extreme levels of customer dissatisfaction?

The answer to that question can be understood, considering the difficulty of adapting innovative technologies to existing financial services. Established financial institutions have been around for more than 100 years with a large brick and mortar retail footprint, which in turn prevents the businesses from cutting costs and introducing new products quickly. At the same time, many of these institutions have already prefer to invest their billions of dollars, and IT budgets in maintaining their existing systems, and to comply with the highly regulated industry by a wide range of regulators across state and federal, leaving almost no room for innovative product development process. The existence of compliance requirements and complex infrastructure creates both challenges and opportunities for many startups trying to enter the market, which is a paradox of today's financial ecosystem.

Nevertheless, it would be fair to be optimistic about the future without ignoring such challenges both traditional institutions and new entrants mostly face.

Why should we be optimistic?

When looking at ten years ago, the people trying to build a product for the market were required to go through many phases, just like buying the physical servers, software licenses, writing some codes for a database e.g. However, today, a software company can be started through only a credit card or a laptop with the help of infrastructure as a service system developed by Amazon Web Services which can bring all of these phases into one step. More importantly, this system reduces the cost and complexity of launching a software business, and of enabling the financial services companies, which in turn unleashes thousands of experiments for the future of banking.

In other words, the notion of "AWS for fintech" shows itself in the banking industry. For example, before the infrastructure as a service system, the cost of running compute and storage for a business could be around $150,000 for a month, yet it now costs approximately $1,500 for a month. And, as mentioned previously, today, any company can more easily and less costly start or enable financial services. In that sense, consumer apps across a wide range of categories are working as de facto banks.

For example, Distributed Core Banking Platform was established through the collaboration between the Ant Financial Services and Hoperun Information Technology to offer alternative banking products to people with the help of customer-oriented business models. Such a new solution is facilitated by Business Platform as a Service product that was introduced by ANT Financial in September 2018. Through such an innovative business platform, financial institutions can take the advantage of Ant Financial's capabilities in the areas of product and asset management, capital verification and full-link pressure tests that are combined with Hoperun's experience in the finance industry.

All of these developments sound promising. However, how can the globalization of fintech wave become possible?

As we know, while B2B companies are traditionally making money by selling the products to the customers, B2C companies gain their revenues by selling advertising to customers as well as selling products. However, both B2B and B2C companies have started to implement new business models based on the white-labeling financial services with the help of rising fintech wave.

For example, ride-sharing companies such as Uber joined neo-banks to provide customers with debit cards and checking accounts. Adapting the financial services to their business plans has contributed to the roadmaps of such companies in two ways. First, there was a high cost resulting from the employment of the drivers for these companies. After that, the employees were required to compensate that cost through margin on rides. However, such a compensation process has become faster through margin on banking services. At the same time, the driver is more likely to continue to work with the company because of the provided financial services.

In addition to the ride-sharing companies, there have also been many other companies across multiple sectors that started to support fintech business models and infrastructure to offer specialized services to their customers. In that sense, as an attractive opportunity for investing, the infrastructure companies are highly supporting the implementing financial services for both consumers and B2B products. Considering some examples, just like Shopify, Mindbody e.g., whose almost half of their revenues are provided by the financial services such as offering websites for a monthly subscription fee to any merchant, the dimensions of the fintech explosion with the rise of "as a service infrastructure" might be understood to a certain extent.

As another example, CollectAI, digital receivables management firm, puts the business-to-business-to-consumer (B2B2C) white-label solution for the digital payment reminders in the segments of credit cards and consumer credits, and importantly, such sophisticated solution is deployed by Hanseatic Bank, a Germany-based subsidiary of Societe Generale. Through the collaboration between the traditional bank and the fintech company, while the collection rate of the bank pointed up from previous results, the expenses of the bank for receivables management reduced. Michel Billon, CEO of Hanseatic Bank, states:

"Thanks to collectAI's solution, we have increased the efficiency of our account receivables management and optimized customer retention with higher satisfaction. Further key results were overall higher cash flow, a faster execution as well as reduced communication costs. Customer-centric collections are the key to our digital transformation strategy."

In the highly regulated banking industry, applying for a license can take years, and most of the new startups and companies have been increasingly trying to find a sponsor bank, to be a partner with the banks, and borrow a license by "open banking" innovation. However, as well as such partnership, the startups are required to get the core system, just like the large database to log where the money of their customers is, and how it is moving around. And then, it is necessary to integrate this database with a series of payment systems to enable the customers to take money out of their accounts.

As we know, to provide the loans to the customers, there is a need for the borrowers' credit history and some personal information obtained by the credit bureaus following the KYC (know your customer) process. And, because of the necessity of protecting the private data of the customers, there is also the need for more software programs. Therefore, the open banking partnership or banking as a service system has been increasingly preferred by both banks and startups because of its contribution to the KYC process and customer-centric business model.

At that point, it can be said that through the "infrastructure as a service" system enabling card issuing, regulatory compliance, and money movement, financial services have been increasingly being part of almost every company's roadmap day by day. That is to say, the way the banking industry operates is fastly changing in the digital age. As mentioned above, startups have also emerged as the actors contributing to such large-scale transformation of the financial landscape by introducing new, better, and cheaper products and lower prices for consumers.

In that sense, Colendi company can also be regarded as one of these new entrants to the finance and banking industry. As a credit scoring and micro-credit platform, at the beginning, Colendi project was developed to evaluate the creditworthiness of the consumers by leveraging a wide range of data sources about users. The main goal was to provide the un/underbanked individuals with financing opportunities with the help of machine-learning-based credit scoring mechanisms that enable the Colendi company to evaluate complementary and distributed data segments of the users.

In that sense, Colendi company was developed as a protocol aiming to understand the real needs and solutions of the problems the consumers mostly face. And then, the protocol has been tested in many used cases both through users and developers globally until this time. At that point, Colendi can now be called "as a service" platform whose B2B2C model aims to provide white-label solutions to any merchant, utility company or consumer company in today's Fintech ecosystem. Today, Colendi has more than one thousand merchants in the market in Turkey and is working with more than three big companies through B2B2C white label product model with the goal of creating big scalable model for utility, telecom companies and most of the other merchants trying to create their own credit market places.

Now, let's get to know Colendi company better, focusing on its primary functions. Through Colendi tools, for paying their shopping, users can receive a monetary loan from e-commerce networks, retail shops, or any merchant with the agreement that the loan will be repaid. In other words, users can do shopping through installments offered by the Colendi contracted merchants. In the first stage, users are required to sign up to Colendi after downloading Colendi APP, and then, there is a need for each user to create their ID and scores through data access permission. Thus, borrowers can use the Colendi application at check-out to receive microcredit if their Colendi scores are sufficient. After that, they have a chance to repay their installments via Colendi platform. All of these transaction details shape the credit score of the users as data dimensions.

One of the main aims is to team up with reputable organizations that currently work on enhancing and expanding microcredit opportunities for customers. Whether they are non-profit or for-profit, Colendi prioritizes signing a partnership to provide finance to the unbanked legitimately and efficiently. In that sense, Colendi offers alternative credit scoring mechanism, therefore, the chance to be financially included in the market to previously un/underbanked people through its fintech business model.

Due to the failure of traditional credit scoring mechanisms to evaluate people's eligibility for microcredit, for a long time, most of the global population has been wholly deprived of banking services according to international organizations' data. In the Colendi platform, the credit score of users is shaped by user-owned data, smartphone, and social media data, and tertiary data sourced from data partners in the ecosystem. In addition to these data segments, Colendi credit history shaped by the transaction and repayment performance of the borrowers is also the main factor in determining their financial scores. Thus, it is important for borrowers to repay their debts on time to have a high credit score, and thereby, eligibility for taking micro-credits.

As mentioned above, the users can make their payments for shopping through the personal wallets the users create in their Colendi mobile application. First, users are required to register to the application by creating a pin code and giving their telephone number. And then, personal verification code is sent to the mobile phones of the users. After that, users can start to make shopping from the stores taking place in Colendi merchant list according to the credit amount the users can determine and then, see at the top of the screen of their applications. After users determine the credit amount for shopping, they also have an opportunity to choose the payment system including the installment plan, total amount of repayment, e.g. In addition to the desires of the users, the credit amount is also determined by their current credit scores.

At the same time, we can talk about the Colendi Card as a payment method that can be obtained physically, and connected to users' Colendi ID and scores by users. The users of Colendi Card can improve their Colendi scores after each transaction and through regular repayment schedule. These can be efficiently handled without the risks carried by typical credit cards, such as excessive credit and high-interest rates, and the users can transact with the merchants in the Colendi platform or anywhere a MasterCard®️, and Troy®️ is accepted. Thanks to Colendi Card usage, the aim is to enhance the financial reputation of the users.

As we know, the term "utility" referring to the triangulation of the ability to store value, to move money and access credits has become one of the buzzwords to identify today's fintech ecosystem. In that sense, as Colendi, we are proud to be a part of that ecosystem, offering the real solutions to the problems not only consumers but also businesses mostly face in the market.

Disclaimer: The author is the Co-Founder at Colendi.

Sources

The World Bank Features Colendi's Approach to Credit Scoring

Thu, 02 Apr 2020 00:00:00 GMT

Today the World Bank Group and the International Committee on Credit Reporting published their Credit Scoring Approaches Guidelines. The report includes Colendi as a use case for decentralized credit scoring and cites our technical paper.

For our team, this is meaningful because it brings attention to the problem we set out to solve: conventional credit scoring works best for people who already have a formal financial history. Those without bank accounts, credit cards, or previous loans may be economically active and reliable, yet remain difficult for the system to evaluate. A missing credit file is too often treated as evidence of missing creditworthiness.

We built Colendi around a different idea. With a user's permission, a scoring model can learn from a broader and more current picture: smartphone and telco activity, purchases, transaction patterns, retail relationships, and repayment behavior. These signals do not replace careful risk assessment. They make it possible to begin an assessment where traditional data provides little or nothing.

The World Bank guidelines describe Colendi as combining machine learning, identity checks, alternative data, and blockchain-based infrastructure. Our technical architecture is designed to keep an important boundary intact: the information used to calculate a score should not automatically become information exposed to every lender or network participant. Users authorize access. Sensitive records are encrypted or anonymized. Computation takes place separately from the public ledger, and a lender receives the resulting assessment rather than a copy of a person's private digital history.

This distinction matters. Financial inclusion cannot be built by asking people to surrender control of their data in exchange for opportunity. A more inclusive credit system must also be a more responsible one—clear about permission, proportionate in the data it uses, and designed to protect the individual behind the score.

The report is not an endorsement of one company or one model. It is evidence that alternative-data credit scoring has become part of a serious global conversation about the future of credit reporting. We are proud that Colendi's work is included in that conversation, and even more determined to prove that better technology can make access to credit broader, fairer, and safer.

Everything You Ever Wanted to Know About Fintech 3.0 and Its Impact on the Banking Sector

Sun, 20 Oct 2019 00:00:00 GMT

The world of banking shaped by FinTech

Banking is always one of the business sectors that has leveraged the technologies of the time to improve its products, services, and processes. According to Facebook's report of Millennials + Money: The Unfiltered Journey, 92% of millennials, the largest single generation ever with an average age of 26.5, don't trust traditional banks.

This shocking observation can be attributed to the 2008 financial crisis and its effects on the global population. Moreover, during the global financial crisis, the profitability of the banks was found to undergo a dramatical decrease because of the slowing revenue growth.

One of the most surprising development in the post-crisis period, was the rise of start-ups, companies and big technology firms offering financial technologies (FinTech) to provide different methods to the clients worldwide, especially new ways of payment services. In other words, the actors of the financial landscape in the business environment were not only the banks anymore. There has been an attention-grabbing shift away from established banks toward new players of financial services.

Given that today, financial technologies are growing at an unprecedented pace, the most thorny questions to answer are as follows; how can the banking sector be adapted to the recent trends in FinTech for the next generation? Is the rise of FinTech a threat to the traditional banking industry or is it an opportunity that the banks should utilize to achieve better functionality in a new competitive landscape?

To understand the extent of the challenge for banks to be adapted to the rapidly emerging FinTechs, first, let's take a look at the evolution of modern banking throughout the 20th century. As the phrase goes, before the flood, it was possible to have a customer record only by filling out a physical card in a specific bank branch. And, since the bank account details of customers were stored on these cards, people could not move from one branch to another without opening another account. However, after Electronic Record Machine for Accounting (ERMA) was built by MIT for the Bank of America in 1953, people could use new unique bank account numbers for the first time, rather than being defined by their name and surnames. In the mid-1990s, with the introduction of internet, we met with ATM Machines, which could contribute to the emergence of self-service banking. Considering the evolution of banking until the 2000's, we can properly state that the world of banking has successfully harnessed the technology and applied it on a comprehensive scale after a fifty years of dormant period.

Our experience today tells us a different story from what it was about 30 years ago. Game-changing technologies are now coming into our lives once every 2-3 years, which means that the global banking industry is challenged by the necessity of adapting to these rapid changes in FinTech. As stated by Brett King, Co-Founder of Neo-Bank Moven, you can bank anywhere, any time in today's world. As the new channels for banking such as mobile wallets, digital payment platforms, and social media have rapidly emerged, the banks are no longer those old branches that you have to pay a visit but something you do because you can now do 7/24 from anywhere through prepaid credit cards and mobile payment systems without the need for a traditional bank account.

Today, this new wave of technology is unprecedently altering the conventional definition of banks and banking. With the rise of innovative technologies over the past decade, people became aware of how their lives were reshaped by the "blessings" of technology. However, there are still some areas that are underdeveloped such as the idea of non-human beings such as technological tools are not allowed to follow the changes on a large scale by themselves without the individuals who can contemplate and put these tools into practice depending on their temporal demands and expectations. Therefore, it should be kept in mind that to understand better the main reasons behind the transformation in many areas of our lives in conjunction with the technological developments, it is necessary to be aware of people's changing needs and demands in time to have the technical knowledge of game-changing innovations.

At this point, we can rightly state that in today's banking world, the critical issue should be the utility of services embedded in the functionalities of emerging technologies. The term utility is accompanied by the combination of three core elements; the ability to store value, the ability to move money, and the ability to access credits. In other words, channels and devices in the banking ecosystem are now preferred according to their ability to provide easiness and, the speed for the customers who are trying to open and access their accounts, conduct transactions, and get service issues resolved. Therefore, established banks must meet the requirements of their customers through easier and more convenient services to continue their businesses and thereby, survive in a new banking landscape. As stated by Anne Boden, Founder and CEO of Starling Bank;

"Banks were so focused on getting rid of their bad loans and reducing their staff headcount that they'd forgotten about customers. Customers had changed — but banks hadn't noticed."

FinTech can help the banking industry to create a personalized user experience based on customers' needs and demands. The concept of FinTech can be identified as the innovative use of technology in the design and delivery of financial services. Therefore, financial technologies have the potential to transform the world of banking through technological innovations such as equity crowdfunding platforms, mobile payment systems, peer-to-peer lending, big data, robo advisors, blockchain, artificial intelligence, and machine learning.

So, why should the banking business harness such "sophisticated" technologies? The answer to that question can be found, considering the new gap between the traditional banking system offers customers and what people came to expect to meet their needs and demands. To survive in today's financial ecosystem, banks must actively listen to their customers and have a user-centric perspective. The impact of innovations mentioned above that are central to fintech on shaping the banking industry in many aspects such as mobile, online and retail banking, e.g. are covering more area every day.

The Future of Banking in the Digital Age

Rather than visiting the high street bank branches, people have long preferred to use the mobile banking system to make deposits, account transfers and, monitor their expenditure and earnings. According to Business Insider Intelligence's Mobile Banking Competitive Edge Study, 89% of the participants in the survey said that they actively use mobile banking. As the number of people using mobile banking grows in popularity, we can observe the growth in mobile application downloads worldwide as well.

The main reasons behind the growing interest in mobile banking are related to its ability to offer easiness, speed, and security through smartphones that allow customers to make purchases online or in-store without the need for any physical credit card or visiting a branch. As people also become more agitated about large scale data breaches of leading financial institutions, they increasingly rely on technology companies with their digital data more than their traditional banks to eliminate their concerns about transaction security. With the help of some unique properties of mobile devices, mobile banking apps have the potential to provide consumers with more robust measures against cybercriminals such as identity thieves than what established banks can offer in the online environment, according to Aite Group's research.

According to Business Insider Intelligence experts, the popularity of online banking is strikingly surpassed by that of mobile banking. As their data have shown, mobile banking is growing at five times the rate of online banking, and half of all online customers are also mobile banking users. As also figured out by Eric Wilson, Co-founder of Xinja; "The next evolution of online banking is to be a last-century business model built by specifically smartphones."

However, despite the popularity of mobile banking, some banks still cannot satisfy the demand for some mobile tasks such as paying the bills and reward redemption, and they push users to use online banking. Although mobile banking and online banking are mostly viewed as interchangeable concepts, there are some differences between these two terms. While mobile banking refers to the opportunities provided by banks that enable customers to make their transactions through short message services, mobile application or websites, via internet banking, customers can conduct their transactions over the bank's website on their computers. More importantly, in mobile banking, fund transfers can be made through the National Electronics Funds Transfer System (NEFT), and Real-Time Gross Settlement (RTGS). On the other hand, in online banking, it is possible to make fund transfers from one bank or branch to another with the help of Immediate Payment Service (IMPS) as well as NEFT or RTGS.

As another payment method, today, cryptocurrencies increasingly started to receive attention from the financial ecosystem. For example, U.S. Bank J.P Morgan Chase's launch of JPM Coin, and Facebook's Libra go beyond the basics of today's commerce industry. As a stable coin representing fiat currency, JPM Coin is launched in February 2019 for the transfer of payments on a decentralized blockchain network between the institutional clients. Libra is also created by Facebook to provide those without access to financial services through the integration of digital wallet capabilities into Facebook's existing messaging services. David Marcus, the head of Facebook's blockchain subsidiary Calibra, has often emphasized that:

"The very people who say they lack the money to open a bank account are actually not saying that they have no use for modern financial services. They're just saying they can't afford to access the system."

As well as mobile and online banking, retail banking has also been reconceptualized by the "disruptive" technologies over the past few years. Traditionally, retail banking provides consumers with some financial services such as savings and checking accounts, credit and debit cards, and loans. With the rise of digital technologies, the operating efficiencies of retail banking considerably improved. According to Business Insider Intelligence, 39% of retail banking executives accept the impact of new technologies on reducing costs and increasing transparency, while 29% say that new technological trends are more effective in improving customer experience.

Another game-changing technology with a strong impact in retail banking is the blockchain. According to the Matt Higginson, the management consultant in McKinsey, because of the cost pressure in the retail banking market, banks have started to consider blockchain-based projects. Blockchain infrastructure enables financial transactions to be verified by each computer in a decentralized network without the need for any central authority, which means that there is no single point of failure, an effective solution to data breaches. Because of such transparent and decentralized nature of this technology, retail banking started to leverage the blockchain. Santander Bank, for example, launched a new international money transfer service, known as Santander One Pay FX, using blockchain technology in cooperation with California-based Ripple in 2018. After the launch of the blockchain-based service, Ana Botín, Executive Chairman of Banco Santander, stated that:

"One Pay FX uses blockchain-based technology to provide a fast, simple, and secure way to transfer money internationally — offering value, transparency, and the trust and service customers expect from a bank like Santander."

Specifically, the three specific areas retail banking prefers to integrate blockchain technology into its roadmap are remittance payment processing, fraud prevention, and risk scoring system.

As we all know, cross border payments take a long time and have a much higher cost than local payments. However, fintech plays the role of a new "panacea" in reducing the costs of remittance services for a while now. According to World Bank report on global trends for remittance prices published in June, 2019, the global average cost for sending remittances was 6.84%. However, the cost of sending remittance from low and middle-income countries such as in South Africa are still somewhat upwards of 20% level. However, as stated by McKinsey authorities, with the help of distributed Blockchain ledger, up to $4 billion could be saved in a year in cross-border payment costs. Blockchain-based digital currencies can create value by fixing certain inefficiencies since they are not required to be controlled by a central regulating body, which means that through decentralized blockchain, the payments can be made in a matter of seconds.

Therefore, many banks have, for a while, started to implement blockchain technology in their current remittance systems. For example, Visa Europe and BTL Group's Interbit Platform are collaborating to provide cross-border settlement solutions through cryptocurrency BTL acting as a bridge currency to create instant and nearly free global money transfers of any size. Similarly, we can talk about the partnership between blockchain-based platform, Digital Trade Chain, and a few major European banks, including such as Spain's Santander, along with IBM to manage open account trade transactions globally. As we mentioned before, Santander is also one of the first banks in the UK harnessing blockchain-based Ripple for international payment transfers through mobile applications. Furthermore, UBS, Royal Bank of Canada, and the National Bank of Abu Dhabi have, for a while, used Ripple's payment protocol and exchange network to provide real-time affordable money transfers for their customers.

Second, the problem of identity fraud can also be solved with the distributed blockchain ledger enabling retail banks to conduct their businesses with greater transparency and immutability since no data cannot be possibly altered or attacked without the "consensus" of majority in the decentralized blockchain network. Identity fraud has been a real challenge for the banks because it leads to an approximately $20 billion loss a year. For protecting the customer data, retail banks now mostly attempt to combat fraud, prevent money laundering, and introduce real-time information sharing based on the predictive models. However, such efforts of the retail banks to provide their customers with ID protection cause longer onboarding times and higher costs which can be resolved with blockchain. For onboarding or account opening, customers have a chance to use the digital fingerprint as a unique identifier to prove their identities globally since these fingerprints can be stored on a distributed blockchain ledger and referenced by any bank in the network. As data clearly show in CB Insights report, blockchain-based solutions can reduce the annual costs resulting from identity fraud, which can amount to as high as $7 billion to $9 billion by cutting the risk of financial crimes, and thus improving efficiency.

In that sense, Bluzelle networks, a blockchain-based data storage startup, has started to work with HSBC, OCBC, and Mitsubishi UFJ Financial Group in Singapore in 2017 to try their proof of concept for a platform for KYC, Know Your Customer. As another example, IBM and the National Bank of Canada, Scotia Bank, and TD have collaborated with the Secure Key, Canada-based FinTech firm, to develop a digital identity and authentication service that provides their customers with an easier access to online services in the scope of digital banking.

Third, the credit decisions are mostly made by the banks with the limited data of potential borrowers, which means that some parts of society still cannot get access to most of the financial services such as opening a bank account, taking out a loan, e.g., due to the deficiency of traditional credit scoring mechanisms of banks in assessing potential borrowers' credit-worthiness. Blockchain technology can offer a new solution for collecting data of borrowers from a wide range of sources such as social media, smartphones, etc. to fairly measure borrowers' eligibility for bank credits. As detailed in the article of McKinsey company, the process is as follows;

"…Data carried on a distributed ledger could be accessed without at the time permission (customer consent can be granted via pre-programmed smart contracts.) Banks could theoretically view data that has been uploaded by any bank in the network. The result should be faster decisions, more efficient processes, and the potential for a more informed credit allocation process…"

Latest Trend in Banking: Partner Bank Model

As can be understood from the developments mentioned above, many startups and companies started to take place in the global finance industry besides established banks. As the list of fintech startups such as Paypal, Zelle, Venmo, Stripe, and Coinbase, e.g., increasingly grows, the fintech unicorns are mostly viewed as a "threat" to the dominant position of the traditional banks in the sector. In fact, these new fintech companies were further boosted with the help of open banking rules of the EU that led them to gain access to data and accounts of clients who authorize it.

On the other hand, as mentioned at the beginning of the article, we have witnessed the rise of big technology firms, known as "tech fins" that provide alternative financial services for customers globally from the US to China in addition to the small fintech startups. Examples to the concept of tech fin include Google, Amazon, Facebook, and Apple (GAFA) in the U.S and Baidu, Alibaba, & Tencent (BAT) in China. These tech giants provide new alternatives for "financial inclusion" especially for the unbanked, with the new digital wallet services through their both massive customer bases and vast cash reserves. For example, today, Alibaba's affiliate company, Ant Financial, approximately serve 1.2 billion customers around the world (data of June 2019) through its digital payment platform, known as Alipay. Another tech giant Tencent has reached a whopping 30% CAGR (combined annual growth rate) in customer acquisition since 2016 with the help of its mobile payment platform, WeChat.

However, as a common point of "fintechs and techfins," both of them are doing exceptionally well in collecting and analyzing massive data sets, improving digital engagement in real-time, and focusing on consumer needs and demands more than the traditional banks ever could do. In this framework, the claim that there is a "tug of war" between the banks and new entrants to the banking market is becoming stronger. However, contrary to what is believed, companies and banks are increasingly open to collaboration to gear their services towards consumers' voices and thereby, to expand their footholds under the name of "the partner bank model."

For example, by Banking as a Service, the companies can provide their customers with financial services such as branded banking products, and payment solutions, which in turn can make it easier for banks to deliver their products and services over the web. That is to say, as a subset of open banking, Banking as a Service (BaaS) allows third party companies to get access to banking services. A smart partnership between the banks and fintechs can improve banking revenue pools with the help of innovative technologies. While fintechs are more agile and innovative, established banks are viewed as the source of large distribution networks with their familiar brands. Therefore, the collaboration between banks and companies can be considered the best path to take for long-term growth.

Within the scope of the partner bank model, Cross River Bank, for example, lets other companies provide banking products and services for their customers via its application programming interface known as APIs. Gilles Gade — President & CEO, Cross River Bank stated that:

"We are positioning ourselves as a new bank play, somewhere between what a bank should be and what fintech aspires to be with the help of new trends in the technology sector."

One of these trends in banking technology that contribute to the collaboration between the banks and fintech companies is the opening up of banks' API to third parties. As we mentioned before, the "open banking" process allows companies to get access to available data provided by the banks, if there is a consumer's permission to access to it. Therefore, it can be properly said that banking APIs allow third party developers to build fintech applications and services based on the data provided by financial institutions. By this means, all consumer financial data is now callable through a single API, while the data was centrally held by the bank in the past.

One of the most remarkable developments is happening in Africa. There has been for a while various initiatives to foster collaboration between South Africa Reserve Bank and new fintech startups through the "regulatory sandbox" system that was designed to harness innovative technologies in the financial sector in relaxed, but still safeguarded regulatory environment, which in turn can contribute to developing open APIs standards.

It is not so likely to close our eyes to such exciting developments in the banking ecosystem. However, there are some key steps that both banks and companies are required to follow to bring off the partner bank model without being distracted. First of all, we can talk about the need for a continuous dialogue between the partners to have a strategic agreement. Then, as stated by Martijn Hohmann, Co-founder of Five Degrees, to have a robust partnership between banks and companies, fintech partners should be kept up with the developments related to the compliance, regulation, licensing requirements. And lastly, banks and fintechs are required to harness big data models and automated decision making processes, while testing the resilience of their businesses.

As demonstrated by looking at the last breakthroughs in the banking industry, there has been a dramatic change in the nature of financial ecosystem over the past decade. However, today, what the crucial one is to be able to keep up with the developments, and to establish strategic partnerships depending on the necessities of the time, as we mentioned in that article. In the new digital era, the needs and demands of customers are focus for many institutions in especially financial landscape. In that point, it is important not to exclude potential partners such as established banks by labelling them as "the actors who are behind the times" since for a while, there have been various examples of smart partnership between the banks and fintechs companies on the basis of customer-centric understanding.

Building New Generation Digital Banks with Programmable Money and Trust

Sat, 16 Mar 2019 00:00:00 GMT

Banking is one of the sectors that most successfully implements technological developments in their products, services and processes. However, the actors of the transformation started with digitalization in finance are not only these banks anymore. According to Facebook report of Millennials + Money: The Unfiltered Journey, 92% of millennials, the largest single generation ever with their average age of 26.5, don't trust traditional banks and this is no doubt shaped by the 2008 Financial Crisis.

The most striking result was the new products and services which have for a while started to appear in the market and extend their reach. For example, in recent years, tech giants, Alibaba and Tencent, Ping An, Baidu and more that have both massive customer bases and vast cash reserves have provided new choice of payment for especially many unbanked customers with their digital wallet services replacing cash and credit cards in both local and global market. When the integration of technology and innovation into the financial sector is considered, it's possible to say that despite its late start compared to its Western "counterparts", the most important one, US, China was currently enjoying a "late-mover" advantage in fintech sector along with the liberalization of its economy.

According to MIT Review January 2019 The China Issue, China's mobile payment market has reached a whopping $15.4 trillion magnitude, about 41 times that of the US. The growth and adaptation difference between the two countries demonstrate how different approaches, investments and supports by the general population can create two sets of trends.

If you look back a couple of years ago it's not surprising that while two Chinese technology and payment companies, Tencent's WeChat Pay and Ant Financial's AliPay had respectively reached of 231.9 and 213.4 bn USD capitalization, whereas the US companies, Paypal, Mastercard and Visa were standing at of 47.6, 112.4 and 181.2 bn USD according to 2016 Factset financial data and analytics report.

In another market study done by PWC "Global Top 100 Companies by Market Capitalisation" report published in 2018, there are two new entrants in the top ten, Chinese technology companies, Tencent and Alibaba, replacing US companies Exxon Mobil and Wells Fargo. While the US still dominates Global Top 100 with 54 companies in 2018 compared to 42 in 2009, China holds the second position with 12 companies (two more than last year and three more than 2009). Especially, Chinese technology and payment companies have the second largest absolute increase in 2018 compared to the data in 2017 with the rates of change, respectively 82% and 75% in market cap among 20 top countries across the globe.

So, where does all of the Tencent and Alibaba's value come from?

Let's start with Alibaba. The success of it lies in part, with Alipay, the third-party mobile and online payment platform under Ant Financial, Alibaba's affiliate company that processed payments for 1.1 billion transactions on Alibaba Group's Tmall.com and Taobao.com. Ant Financial has now approximately 450 million active users by providing an ecosystem for Alipay according to data from Analysys International Enfodesk; and this digital payment service has already begun expanding outside Alibaba's reach to provide e-commerce operations for a larger client base in Mainland China.

The second giant company, transforming itself virtually into a digital bank for its customers, Tencent was designed as a popular messaging application used by over 800 million people for daily communication and payment activities. Since 2016, there has been 30% annual growth rate in user acquisition because of the opportunities provided by its flagship product, WeChat and digital payment business WeChat Pay that gives access to financial management, banking transactions, peer-to-peer payments and more.

From Payment to Banking

Hong Kong government has planned to provide digital banking licenses to the China's top technology and payments companies, Ant Financial, Tencent, ZhongAn, Xiaomi for the first time. This is another groundbreaking accomplishment by these tech giants that disrupt the older traditional banking system that we are familiar with. Such an attempt of Hong Kong government can be understood in the context of a policy change based on the aim of transforming the country, as this semi-autonomous region governed by China, is now evolving into a smart city by opening its own market to the online competition.

Until recently, we've come to know only a few banks such as Hang Seng Bank, Standard Chartered and HSBC owning a lion's share of credit cards and retail mortgage loans with approximately 66% in Hong Kong according to Goldman Sachs researches. For example, HSBC ensured its leader position in the market in last year with $1.4 billion profit obtained thanks to its wealth management and retail operation policy for the second quarter. However, we still could see unsatisfied customers with the existing bank models. According to the Accenture researchers' report published in 2017, just over half of customers in Hong Kong are satisfied with their current banks compared to 88 percent in the US. Accenture Asia Pacific managing director Fergus Gordon commented, "There is a large majority of customers in major markets who are willing to do banking with different models, and digital banks have a great opportunity to tap into that."

The rise of the new fintech giants also seems the have started to affect the lending scene. Growth is very significant, both in amount and transactions which is a true indication of the success of their expansions. WeChat's WeiliDai, although started operations in 2014, have seen a massive growth to surpass 133 billion dollars of loan amount in 2017 with 12 million customers. The growth ratios suggest much more growth in the upcoming years. It's not just the loans given out, these two major platforms also are becoming the mainstream methodology for investing funds. Alipay's money-market fund Yu'e Bao, receives investments up to $3,000 yielding an annual return of 4%. The total amount has surpassed $160 billion.

However, we can say that the unsatisfied customers with traditional banking models is not just peculiar for the Hong Kong market. For a long time, the bidders hope to challenge the "oligopoly" of traditional banks in market by offering a digital alternative to customers who are unsatisfied with their current options in most of the markets in global scale in today's world.

Other fields of tech stepping into the banking scene

As the competition grows between the tech giants and traditional banks to satisfy their customers, we've come across a wide range of new tools and platforms such as artificial intelligence (AI), machine learning and Blockchain infrastructures which opened the door to other capabilities. It can be concluded that such technologies disrupt traditional financial landscape and reshape the methods of financial institutions on providing services in accordance with the expectations of "incredulous" customers.

Especially the Blockchain technology, as a disruptive innovation, is shaping up to be a challenge to the existing social, economic and thereby, political relationships in both local and global level. It seems to perfectly combine the transparency of Internet with the security of cryptography. People's expectation for a more "trustful" and democratic world of data is coming to being day by day. The democratic character of Blockchain is not just related to its capacity to enable people to verify their own data but also to its ability to provide an infrastructure for sharing economy that involves all participants in the network.

The Tech giants are aware of the power of Blockchain technology in various fields and they have recently started to integrate the technology into the artificial intelligence mechanisms of their companies. For example, Alibaba, one of the most giant e-commerce company in China, started to use Blockchain technology to accelerate payment process in the international arena. On the basis of Blockchain technology, the users' account balance will be verified through smart contracts. Thus, without the need for more servers, the system will be more cost-effectively worked over Blockchain networks.

Old Players, New Looks

As the most "disruptive" innovation in today's world, the Blockchain has been the "buzzword" and ubiquitous technology of 2018. Today, even messaging app Telegram have conducted the TON project based on the Blockchain technology with the aim of increasing the real-world applications of cryptocurrencies through the idea of first-ever mass-market cryptocoin. Basically, "TON is a decentralized and robust supercomputer transferring high volumes of information or value." The project is expected to bring non-censorship and anonymity through Blockchain decentralized infrastructure. As TON will achieve maximum security with minimum transactional time, the ecosystem is poised to become a Mastercard/VISA alternative. According to TheBlock, a report for investors of the project, TON Virtual Machine is a software enabling smart contracts to be executed is mostly completed at that time.

As blockchain and fintech revolution is happening worldwide, more giants from various countries are getting more eager to jump into the scene. In the last week of February, Japanese "Megabank" Mizuho Financial Group, which is currently standing at $1.8 trillion, announced that they will be launching a new stable coin J-Coin with the hopes to use it in payments and remittance services. This is a second move after Facebook had announced that they were also working on a new coin that would enable Whatsapp users to exchange crypto coins. If we are speaking of giants interested in the blockchain technology, we also have to take into consideration the latest stablecoin project of JPMorgan. As one of the biggest banks of USA, JP Morgan seems to have started to get involved in the blockchain scene and they are now after being a "reliable" institution with their stable currency in a considerably volatile market. These three examples are not the perfect representations of a "decentralized" coin per se, but it is a perfect use case of the blockchain technology deployed by a centralized company.

Race of blockchain adoption

Telegram a direct competitor of Whatsapp is one a few steps closer to the finish line of their vision. The 5 year-old new messaging app already managed to raise 1.7 billion dollars in a private ICO and is fiercely bearing down on the vision to create a cryptocurrency platform based on their messaging app. This is considered to be 2018's boldest attempts to drive mass adaption of blockchain applications.

At this point, it's inspiring to realize the increasing wave of the "alternative economy" channels created not only by the traditional banks that have started to use digital technology and above-mentioned BigTechs transforming into the new digital banks but also by newly emerged Blockchain-based startups with the "shared economy" slogan.

Is Blockchain for BigTechs or against Tech Giants?

Although BigTechs can be considered alternative digital banks for the traditional in recent years, it is paradoxical for massive technology companies to unite Blockchain and AI since such tech giants continue to act as "gatekeepers" for preservation of data. In contrast, cryptographic protocols designed around Blockchain aims to allow users to have more control over their private data. Therefore, as the number of alternative channels aimed at the idea of "self-sovereign" user identity increases day by day, the suspicion toward the efficiency of BigTechs increases in a similar manner.

Therefore, the mission of some Blockchain-based startups can be understood in the context of "uncompleted" shared economy. Their intention is to create a path based on the trade of data instead of currency, and thereby, erasing the hegemony of the technology oligopolies by providing free access to the information. Through decentralized and distributed ledger system of Blockchain technology, the entire process divided between several nodes is now faster and more trustful because of the requirement of licensing that includes legally binding contracts and permissions for commercial purposes. Blockchain enables users to sketch contracts easily using built-in tools and efficiently transfer the rights to their data to other entities. As a result of which, through Blockchains, it can be considered more fairly awarded content creators for the work they do in the context of "sharing economy."

It would be a very optimistic prediction if we say that blockchain technology will have an equal positive effect all over the world. We have already begun to see major differences in approach to these new tokens and innovations. Some countries are already assuming the leader role thanks to their innovative approach and risk appetite. China's increasing trend in technological investments and growth in patent applications favor the country immensely. However, many western countries like Switzerland, Malta, Germany, USA, Estonia and the other Asia giant Japan do not lag following suit. The Swiss city of Zug has already earned the reputation as the Blockchain Valley thanks to their early recognition of the power of this recent technology, which they supported by establishing regulations for the private ventures that intend to use blockchain. The new technology implies decentralization and the world seems to adapt to the idea very quickly, as we are drawing closer to a future where Silicon Valley giants will not be leading the world anymore. With the new lead of these countries and more of them joining in this ultimate technological revolution, the regions adapting the new tech is expected to go through a significant improvement in revenue and growth.

Introducing the Colendi Protocol: Our Technical Paper

Sat, 18 Aug 2018 00:00:00 GMT

Today we are publishing the Colendi Technical Paper, a high-level description of the architecture behind our decentralized credit-scoring protocol and microcredit platform.

When we began building Colendi, we started with a simple question: how can a person prove financial reliability when the traditional system has never created a useful credit file for them? A bank can only evaluate the history it can see. Billions of people remain invisible to that view, even though their daily transactions, mobile activity, retail relationships, and repayment behavior can demonstrate consistency and trust.

Our answer begins with identity. Colendi ID is designed as a self-sovereign financial passport controlled by its owner. We use ERC-725 and its complementary claims model, ERC-735, so identity can become reusable across services without depending on a single institution. This direction also connects with the emerging work on decentralized identifiers in the W3C Credentials Community Group.

Identity alone is not enough. Credit scoring requires data, and financial data demands privacy. The Colendi Protocol therefore separates the system into three layers: identity, secure object storage, and a secure computation environment. Personal information is encrypted or anonymized and stored away from the public chain. Computation is performed without exposing the underlying personal records to a lender. The lender receives the information needed for a decision, not unrestricted access to a person's digital life.

Ethereum provides the shared execution and verification layer. It records approvals, references, and protocol events while the larger and more sensitive data remains off-chain. This design reflects an important reality explained in the Ethereum Foundation's discussion of privacy on the blockchain: transparency is valuable, but private data cannot simply be placed on a public ledger.

The technical paper is not the final word. It is the architecture we are testing, challenging, and improving. But it makes our direction clear: a credit system should give people control of their identity, use broader evidence responsibly, and make financial opportunity available beyond the boundaries of a conventional credit bureau.

That is the protocol we are building.

Hello World! Hello Colendi!

Wed, 24 Jan 2018 00:00:00 GMT

"Banking is necessary, banks are not."

Five years ago, members of the Colendi team launched Ininal — "The Bank for the Unbanked" — with a dream of empowering all of Turkey's citizens with financial tools. Our journey at Ininal touched over 1 million users, and we learned first-hand about the foundational inequalities of global banking and the great need for change. Almost 72% of the world's population has either limited credit data or none at all, and have no access to credit or assessment systems.

Today, both financial institutions and credibility assessment systems have fallen behind the needs of a fast globalizing world. Addressing these long-standing inequalities has became the goal of Colendi — a dedicated technology bringing the digital and accessible means of credit scoring to the entire world.

Colendi is a blockchain-based, decentralized credibility evaluation and microcredit platform, fully compatible with participation banking requirements, as well as traditional financing methods. Colendi aims to provide a unique and global identity and a global financial reliability passport on the blockchain for every person in order to create a more democratic, transparent, inclusive and truly mobile creditworthiness measure.

Through its decentralized structure — built on machine learning algorithms — Colendi gathers and processes all digital data of a user in order to create a universal digital identity and comprehensive credibility image. Smartphone, social media, telco, retail history and many other aspects of a user's data are evaluated in Colendi network in order to bring out the most comprehensive creditworthiness measure as a global financial passport.

Via its cutting-edge technology and global credibility identity, Colendi provides financial reliability information based on users' own distributed data. The Colendi platform uses its decentralized intelligent scoring and fraud detection algorithms by evaluating their users' real-life data to identify truly credible users, whether they are banked or unbanked.

Colendi will enable various financial applications for unbanked individuals as well as an evolution of credibility assessment. Microcredit, installment shopping, P2P financing, and reliability checking will be the primary focuses in regards to financial use cases. Many more applications and systems will be built on top of Colendi protocol and credibility technology as the roadmap develops.

The Colendi journey began in 2016. Here's a summary of what's happened so far:

Dedicated team was formed — Feb 16
Development of Colendi main scoring technology has started — Feb 16
Over 1 million digital wallet users have been scored with Colendi algorithms! — Sep 17
Whitepaper has been published — Nov 17
Patent applications were made — Nov 17
First Retail Chain agreement has been made — Dec 17
First Colendi Scored credits were lent! — Jan 18

Colendi has started its journey to transform credibility in a fast pace and both our team and most valued advisor pool are growing even faster! What will be heard from us is as follows:

Colendi Token Sale! — Coming soon
First Colendi Financial Passport and Credits on Blockchain — 2018 Q3

Now, join our Telegram account to be a part of the transformation and visit our website to learn more about Colendi and whitepaper.

Stay connected!