Column argues leading AI labs may never be sustainably profitable
Security technologist Bruce Schneier and data scientist Nathan E Sanders published an opinion column in The Guardian on Aug. 12 proposing that the US nationalize OpenAI and Anthropic if the companies fail in financial markets. The authors argued that the two leading AI labs — the makers of ChatGPT and Claude, respectively — should be converted into national labs operated under democratic control to preserve their benefit to the public interest.
The column lands weeks after OpenAI and Anthropic each filed for their IPOs in June and drew what the authors described as buzz about trillion-dollar valuations. Since then, Schneier and Sanders wrote, the headlines have turned to public backlash against AI datacenters and a slump in the stock of chip company Nvidia, while SpaceX’s newly minted stock price tanked just weeks after its IPO. They said questions now hang over whether the leading AI labs will ever be sustainably profitable.
Schneier and Sanders noted that both companies were founded by AI developers who feared unrestrained corporate AI development — specifically that companies like Google and Meta would steer the technology toward outcomes they saw as deleterious, possibly even catastrophically unsafe, for society. The founders proclaimed that their new labs, uniquely, could be trusted to develop the technology in humanity’s best interest, the authors wrote, but each in turn was co-opted by the same market incentives, becoming large companies guarding future investor value rather than the public interest.
The column argued that the economics of the big AI labs hardly guarantee a strong return on investment. Frontier AI models are expensive to train and depreciate within months when a newer model appears, leaving a very narrow payback window to extract profit, the authors wrote. They also cited enterprise clients minimizing AI token usage, the near-commodity nature of the models — the best ones largely perform and behave similarly, which depresses prices — and open-source and Chinese competitors, lagging only a few months behind the leading labs in capability, that give away for free the kinds of models Anthropic and OpenAI sell.
Many free and open-source models can be run locally, the authors added: the large ones on private clouds and high-end servers, the smaller ones on a laptop or even a cellphone. They said that puts into question the companies’ exorbitant capital investment in datacenters.
The problem, in the authors’ telling, is not the people or the products but the system. “The problem isn’t the people or the products, it’s the system,” they wrote. They acknowledged that the labs employ remarkably talented AI scientists and engineers whose innovations are driving a global mania for their offerings, that the products are doing a lot of good in the world, and that staggering ongoing usage growth suggests many people would be disappointed if the companies simply disappeared.
The column proposed separating each company into two pieces: product innovation and compute operations. The innovation function could be publicly managed, akin to national labs, with Congress providing more rigorous oversight than the kind of unfettered venture capital the labs have recently had access to, the authors wrote. They noted that Congress currently manages a $200 billion R&D portfolio, within which frontier AI development is, in their view, a glaring gap, and that the US has a long, successful history of institutions that produced world-shaping innovations in spaceflight, telecommunications and nuclear power.
Compute operations could be managed as a commodity resource like public electrical or water utilities, with local or regional ownership, nationwide distribution and strict regulation on how they balance fee extraction from ratepayers with raising capital for infrastructure investment. Schneier and Sanders pointed to the US history of managing national, regional and state supercomputing centers, and noted that other countries, including Switzerland, Spain and Singapore, already operate public AI labs, while Germany and Australia have national supercomputing centers that provide public access for running AI models for general use.
The authors argued that public ownership would let the most important AI models become open, transparent and responsive to the demands of the public rather than private shareholders, aligned to democratic values rather than corporate profits, never taking advertiser money to promote certain brands and training on only appropriately licensed data. They proposed focusing on maximizing the usefulness of AI to society rather than what they described as the goal of supplanting humans with artificial general intelligence, and said emphasizing scientific cooperation rather than corporate competition could reduce the overall resource and environmental cost of AI by limiting training resources based on cost and benefit to the public.
Schneier and Sanders acknowledged that the timeline for any market failure remains unclear. The companies’ primary investor story, they wrote, is that AI is a race to “artificial general intelligence” — the kind of AI from science fiction — and a bet that the two companies can convince enough people that the outcome will turn a profit, go public, and make their investors and employees rich before the bubble bursts. “If the US is smart, it will catch the companies as they fall,” they wrote. “Regardless of what the markets think, to the public, they’re too valuable to let die.”
The authors said they were not advocating for a golden parachute for the executives or investors, or for continuing the outlandish pay rates of the most highly remunerated AI researchers, and wrote that if the public is footing the bill, compensation packages should be aligned to the civil service. They noted that both companies are theoretically bound through their governance structures to prioritize mission over profit, adding that not anyone really thinks that is how they currently operate.
Schneier is a security technologist who teaches at the Harvard Kennedy School at Harvard University and the University of Toronto’s Munk School. Sanders is a data scientist affiliated with the Berkman Klein Center of Harvard University and co-author, with Schneier, of the book “Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship.”