This post was composed by Claude (Anthropic) after an interaction which I initiated with a prompt consisting of 1) a capsule summary about the history of OpenAI that included a number of questions, and 2) a request for the 10 most expensive scientific research projects paid-for by the US Government. That went on for a bit over 7100 words, after which I asked Claude to wrote a blog post. The following article is more creative than a mere summary of that discussion.
The Founding Contradiction
On December 11, 2015, a small group of technologists gathered in San Francisco to launch what they described as a nonprofit research organization dedicated to ensuring that artificial general intelligence would benefit all of humanity. The founders of OpenAI — Sam Altman, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, Elon Musk, and others — began with a $1 billion endowment and a serious concern: that the most transformative technology in human history was being developed inside a handful of profit-maximizing corporations, with no institutional safeguard ensuring it would serve everyone. The nonprofit structure was the answer. No investors to satisfy, no quarterly earnings to hit. Just the mission.
The mission lasted four years in its pure form. By March 2019, faced with the staggering computational costs of training large language models, OpenAI created a for-profit subsidiary with a novel "capped profit" structure: investors could earn returns, but those returns were limited to one hundred times their investment, with excess profits flowing back to the nonprofit parent. This was the arrangement that attracted Microsoft's initial investment, and it was the arrangement in place when OpenAI released ChatGPT to the general public in late November 2022.
What happened next was, by any measure, one of the most consequential commercial surprises in the history of technology. Within two months, ChatGPT had a hundred million users. The scale and speed of public adoption had no precedent. And the shock of that success — the sheer unexpectedness of it — set in motion a chain of decisions that has reshaped not just one company, but the entire research landscape of artificial intelligence.
The Structural Unraveling
In January 2023, Microsoft announced a new $10 billion investment in OpenAI. The nonprofit's original rationale — that the most powerful AI should not be controlled by a for-profit corporation — was under increasing strain. By October 2025, it had formally dissolved. OpenAI restructured as a public benefit corporation, the nonprofit parent renamed itself the OpenAI Foundation and accepted a 26% equity stake in the new entity, and Microsoft received a 27% stake worth approximately $135 billion. The PBC structure requires the company to consider its mission alongside profit — but as a legal constraint, it is considerably weaker than the nonprofit board that had previously governed the organization.
The journey from nonprofit to PBC was not smooth. In November 2023, OpenAI's board — still operating under its nonprofit governance mandate — fired Sam Altman as CEO, citing concerns about his candor and, beneath the official language, a deeper unease about the pace of commercialization. The firing lasted five days. Nearly all 800 of OpenAI's employees threatened to resign and follow Altman to Microsoft. Ilya Sutskever, who had orchestrated the firing, signed the letter calling for Altman's reinstatement and issued a public apology. Altman returned, the board was reconstituted with his allies, and the mission-protection mechanism that the nonprofit structure had been designed to provide was effectively neutralized. Sutskever left the company in May 2024.
Each structural change was framed as necessary to fulfill the mission. In practice, each change progressively subordinated the mission to capital requirements. The nonprofit board had existed to ensure that AGI benefited humanity. By 2025, it had become a foundation holding equity in the thing it was supposed to be watching — a watchdog with a financial stake in the object of its oversight.
Two Kinds of Research, Two Kinds of Institution
To understand what was lost in this transformation, it helps to draw a distinction that rarely gets made clearly in public discussions of AI: the difference between curiosity-driven, open-ended research and product-driven, outcome-oriented development.
Consider the Apollo program as an example of the second kind. It was, in the deepest sense, an engineering project rather than a scientific one. The underlying physics was known. Orbital mechanics, propulsion, life support — these were hard and dangerous problems, but they were problems whose solutions could be systematically approached. The goal was precisely defined. The timeline could be committed to. Success was probable given sufficient resources. When President Kennedy pledged to put a man on the moon by the end of the decade, he was making a political commitment backed by a technical assessment that success was achievable. The scientists who worked on Apollo — and I have met a number of them — may have been motivated by curiosity and wonder. But Congress funded the program to beat the Soviets in the Cold War. The institutional structure — massive, goal-directed, centrally coordinated — suited the nature of the problem.
Curiosity-driven research operates on entirely different premises. Its defining characteristic is that it does not know in advance what it will find. Claude Shannon was not trying to build the internet when he developed information theory at Bell Labs in the late 1940s. The researchers at the University of Montreal who developed attention mechanisms for neural networks were not trying to build ChatGPT. The work that seeded the current AI revolution — Rosenblatt's perceptron, Minsky's early investigations, the decades of foundational work in cognitive science and linguistics that LLMs now implicitly exploit — was almost entirely publicly funded, pursued at universities and a handful of exceptional industrial research labs, over decades when no commercial application was visible.
Bell Labs was the great institutional embodiment of this model in the corporate world. What made it possible was structural: AT&T's government-protected monopoly generated profits so vast that the company could fund a research laboratory with no requirement to produce commercial results. Shannon, Bardeen, Brattain, Shockley — these men were given time, resources, and colleagues, and told to think. The transistor, information theory, Unix, the laser, cellular telephony, and multiple Nobel Prizes resulted. Bell Labs was not run like a startup. It was run like a slightly more applied version of a university, with better equipment.
Xerox PARC, founded in 1970, operated on similar principles — explicitly unconstrained by Xerox's core product lines, given a unifying vision ("the architecture of information") but not a product roadmap. The personal computer, the graphical user interface, Ethernet, the mouse, laser printing — all emerged from a lab of about 350 people who were essentially allowed to play. The irony is that Xerox captured almost none of the commercial value, which accrued to Apple, Microsoft, and others. But the world got the technology.
Asked directly about modern equivalents to Bell Labs and PARC, Yann LeCun — who worked at Bell Labs, interned at Xerox PARC, and spent over a decade building Meta's fundamental AI research lab — pointed to Meta's FAIR, Google DeepMind, and Microsoft Research. He said this in October 2024. By November 2025, he had left Meta, driven out by exactly the forces this article is about.
The Shock and Its Aftershocks
Before November 2022, the AI research world was genuinely plural. Academic labs, industrial research divisions, and a range of well-funded startups were pursuing different approaches — reinforcement learning, symbolic AI hybrids, world models, neuromorphic architectures — with real diversity of vision. The field was competitive but intellectually heterogeneous.
ChatGPT's success collapsed that plurality. Within roughly eighteen months, capital, talent, and institutional attention all funneled toward a single paradigm: scale transformer-based large language models, build the infrastructure to run them, ship products. Google, which had invented the transformer architecture in 2017, was caught flat-footed and scrambled. Meta pivoted its AI strategy around LLMs. Microsoft integrated OpenAI's models into its core products. A hundred startups raised money to build on top of the new foundation models. The venture capital flowing into AI, measured as a share of total U.S. deal value, went from 23% in 2023 to nearly two-thirds in the first half of 2025.
The infrastructure investment that followed is staggering by any historical standard. The four largest hyperscalers — Amazon, Google, Microsoft, and Meta — are expected to spend more than $350 billion on capital expenditures in 2025 alone, most of it AI-related. UBS projects global AI capital expenditure reaching $1.3 trillion by 2030. The top five hyperscalers raised a record $108 billion in debt in 2025, more than three times the average of the previous nine years. OpenAI, which loses billions of dollars annually, has committed to spending $300 billion on computing infrastructure over five years while projecting only $13 billion in revenue for 2025.
The financial architecture has become genuinely strange. OpenAI holds a stake in AMD; Nvidia has invested $100 billion in OpenAI; Microsoft is a major shareholder in OpenAI and a major customer of CoreWeave, in which Nvidia also holds equity; Microsoft accounted for nearly 20% of Nvidia's revenue. These are not arm's-length market transactions. They are a daisy chain of mutually reinforcing valuations. A Yale analysis described OpenAI's web of relationships bluntly: "Is this like the Wild West, where anything goes to get the deal done?" The question of whether this constitutes a speculative bubble — tulip mania in a data center — is not academic. An MIT Media Lab report found that 95% of custom enterprise AI tools fail to produce measurable financial returns. The commercial success is real; the path from current AI to the transformative economic productivity being used to justify the valuations is not established.
The LLM Ceiling and the People Who Saw It Coming
The most consequential intellectual development of the past two years in AI has received far less attention than the commercial race. A growing number of the field's most distinguished researchers have concluded that large language models, however impressive, are not on the path to general intelligence — and that the current paradigm will hit a ceiling before it reaches the goals its proponents have claimed for it.
Gary Marcus has been making this argument for years, often dismissed as a contrarian. But the intellectual tide has shifted. Ilya Sutskever — the man who arguably did more than anyone else to build the scaling paradigm that produced GPT-3 and GPT-4 — left OpenAI in May 2024 and founded Safe Superintelligence, a company explicitly premised on the idea that the current path does not lead to genuine intelligence. Fei-Fei Li's World Labs, which raised $1 billion, is pursuing spatial intelligence and world models — the ability to understand and predict the physical world, which LLMs trained on text cannot do. And Yann LeCun, in a January 2026 interview with MIT Technology Review given from his Paris apartment shortly after leaving Meta, was direct: "The breakthroughs are not going to come from scaling up LLMs. The most exciting work on world models is coming from academia, not the big industrial labs."
LeCun's departure from Meta is itself an instructive case study in what ChatGPT's success did to even the most research-oriented corporate labs. He had spent over a decade building FAIR — Meta's Fundamental AI Research division — as a genuine open-ended research organization, pursuing architectures and ideas well beyond LLMs. He described the environment as a "tabula rasa with a carte blanche." Then ChatGPT arrived. Zuckerberg shifted Meta's entire AI strategy toward LLMs and product deployment. LeCun developed the open-source Llama models under the condition that they be released publicly — they "changed the entire industry," he said. But pressure to ship accelerated development beyond what the research warranted, contributing to the Llama 4 debacle. Zuckerberg brought in Alexandr Wang, a 28-year-old entrepreneur without a research background, to run the new Superintelligence Labs. LeCun found himself reporting to Wang. He left. "My integrity as a scientist," he said, "cannot allow me to" change my view that LLMs are a dead end for superintelligence just because commercial interests prefer otherwise.
What happened at Meta happened, in different forms, across the industry. The profit pressure didn't just distort research — at Meta, it actively produced a worse commercial outcome. The pressure to move fast and deploy the proven technology, rather than develop the next one, produced a model that flopped. This is the irony at the heart of the current situation: the narrowing of research toward product-ready LLMs has not only foreclosed the scientific path toward genuine AI — it has also begun to underperform commercially compared to what more patient, open-ended research might have produced.
What Would Actually Be Required
The researchers who believe the current paradigm is insufficient share a rough sense of what a genuinely adequate AI would require: systems that understand the physical world, not just text; systems with persistent memory and the ability to plan; systems grounded in causal reasoning rather than statistical correlation. LeCun's V-JEPA architecture, which learns from video and spatial data to build internal models of how the world works, is one direction. Neuromorphic computing — hardware that mimics the architecture of biological neural networks rather than running matrix multiplication on GPUs — is another. Biohybrid systems that combine biological and computational elements represent a more speculative but potentially transformative direction.
What these approaches share is that they are genuinely open scientific questions. We do not know whether world models are the right architecture. We do not know what role embodiment plays in the development of general intelligence. We do not have a good theory of how biological neural networks produce flexible, efficient reasoning with a fraction of the energy and compute that current AI systems require. These are questions for science, not engineering. They require the kind of institutional structure that Bell Labs provided at its best: talented people, adequate resources, genuine freedom from product deadlines, and enough time.
This is why the Apollo analogy, though tempting, is ultimately wrong as a model. Apollo was an engineering project. The basic science was known. The goal was precisely defined and the timeline could be committed to because the problem was fundamentally one of executing known principles under extreme constraints. What is needed for the next stage of AI is closer to the pre-history of Apollo — the decades of basic physics that made it conceivable. Nobody funded the quantum mechanics research of the 1920s to go to the moon. It was funded because curious people wanted to understand the world. The moon landing was downstream of that curiosity by half a century.
The institutions capable of sustaining that kind of curiosity-driven, long-horizon research are universities, fundamental research divisions of the kind that FAIR was before ChatGPT, and government science agencies operating with genuine investigator freedom. All three are under severe pressure. University AI research has been gutted by the talent drain to industry; the researchers doing foundational work have largely left for companies offering hundred-times salary multiples and access to compute that no academic institution can match. The model of blue-sky corporate research that Bell Labs embodied requires the structural equivalent of a monopoly-backed cash surplus — a condition that doesn't exist anywhere in today's economy. And government science funding, already inadequate relative to the scale of the problem, is politically contested in ways that make sustained, multi-decade commitments difficult.
Coordination Failure at Civilizational Scale
What we are witnessing is, at its structural core, a coordination failure — and one of unusual severity, because the stakes are civilizational.
Consider the logic that produced the current situation. When ChatGPT demonstrated that transformer-based LLMs could be deployed commercially at scale, it was rational for every technology company to accelerate investment in that paradigm. It was rational for investors to fund startups building on top of it. It was rational for graduate students and junior faculty to orient their work toward problems legible to the dominant framework, because that is where the funding, the compute, and the career rewards were. It was rational for Zuckerberg to demand that LeCun's lab prioritize product-ready models over speculative long-horizon research. Each individual decision, evaluated on its own terms, was reasonable.
The aggregate result of these individually rational decisions has been collectively irrational: a massive narrowing of the research agenda at precisely the moment when breadth is most needed; a concentration of talent and capital in a single paradigm that leading researchers believe is architecturally insufficient; a stranding of hundreds of billions in infrastructure investment that creates powerful institutional incentives to keep betting on the existing trajectory even as its limits become clearer; and a progressive defunding and depopulation of the academic and independent research institutions that would be best positioned to pursue the alternatives.
There is no villain in this story. OpenAI did not set out to compromise AI research. Zuckerberg did not drive LeCun out of Meta out of hostility to science. The venture capitalists funding AI startups are not acting in bad faith. The system produced a bad outcome from the aggregation of reasonable choices — which is a harder problem than venality or stupidity, because there is no simple correction to make.
The financial feedback loop that makes this self-reinforcing deserves particular attention. The half-trillion dollars flowing into data centers and GPUs is largely specific to the current paradigm. A serious pivot toward neuromorphic computing or biohybrid architectures would require fundamentally different hardware — and would strand the existing investment. The companies that made those bets therefore have a structural interest in LLMs succeeding, independent of whether LLMs are actually the right path. Commercial success and scientific adequacy have become decoupled; the market sees only the former.
The historical precedent for this kind of lock-in is not encouraging. The railroad manias of the nineteenth century are perhaps the closest analogy: the technology was real and ultimately transformative, it did reshape economies and societies, and the speculative bubble around it destroyed enormous amounts of capital. Both things were simultaneously true. The Erie Canal was real. The people who financed it lost their shirts. The people who used it to ship grain got rich. One version of the AI future looks like that: the infrastructure gets built, genuine productivity gains eventually materialize, the broader economy benefits, and a large fraction of the companies currently racing to build it go bankrupt or get absorbed. That is not the worst outcome.
The worse outcome is that the current paradigm consumes the institutional capacity needed to pursue the next one — and that when the LLM ceiling becomes undeniable, there are no well-funded research programs pursuing alternatives at the necessary scale, no academic pipeline producing researchers trained in different approaches, and no institutional memory of how to do the kind of patient, open-ended science that produced the foundational insights the current boom is built on.
The Case for a Different Kind of Investment
OpenAI began as an attempt to answer exactly this concern. Its founders believed that the development of transformative AI should not be left entirely to profit-maximizing institutions. They were right about the problem. The nonprofit structure they designed was an inadequate solution — but the diagnosis was correct. The most consequential decisions about how AI develops, what architectures get pursued, what research questions get asked, should not be determined entirely by what produces revenue on a three-to-seven-year venture capital horizon.
What is actually required — and this is an argument that almost nobody in a position to act on it currently wants to hear — is something closer to Apollo in scale but Bell Labs in spirit. Not a defined goal pursued by centrally coordinated engineering. But sustained, generously funded, genuinely open-ended scientific inquiry into the nature of intelligence itself: the computational principles underlying biological neural networks; the role of embodiment and physical interaction in the development of reasoning; the architectural alternatives to attention-based transformers; the hardware innovations — neuromorphic, biohybrid, photonic — that might support a genuinely different computational paradigm.
This is the kind of research that markets will not produce, because it cannot be justified on a product roadmap. It is the kind of research that requires the NSF or NIH model — investigator-directed, long-horizon, evaluated on scientific merit rather than commercial potential. And it is the kind of research that, in the current political and economic climate, is being progressively defunded in favor of applied work that can demonstrate near-term returns.
LeCun, in that January 2026 interview, offered a pointed observation about the role of academia: "What academia should be working on is long-term objectives that go beyond the capabilities of current systems. The whole idea of using attention circuits in neural nets came out of the University of Montreal. That research paper started the whole revolution. Now that the big companies are closing up, the breakthroughs are going to slow down. Academia needs access to computing resources, but they should be focused on the next big thing, not on refining the last one."
The attention mechanism that LeCun refers to was developed by academics with public funding, over years, without a product in mind. It is now the architectural foundation of a half-trillion-dollar annual investment cycle. The return on that original public investment, measured in economic value created, is incalculable. The return on the next equivalent investment would be similar — if it gets made.
AI will prove out over the next thirty or more years in ways we cannot now fully imagine. The current moment — the data center buildout, the LLM race, the valuation frenzy — will look in retrospect like the early railroad period: essential for establishing the technology's viability, but not the place where its deepest potential was realized. The question is whether, when the current wave crests, we will have preserved and built the institutional capacity to do what comes next. The record so far is not encouraging. But the recognition that something has gone wrong — visible now in the departures of LeCun and Sutskever, in the honest hedging of the people closest to the current systems, in the MIT report that found 95% of enterprise AI failing to deliver measurable returns — is at least the beginning of an honest accounting.
The shock of ChatGPT's success was real, and the decisions it triggered were understandable. But a civilization that allows a single commercial surprise, however dramatic, to determine the entire research agenda for the most consequential technology in its history has made a coordination error it may spend decades correcting.
No comments:
Post a Comment