"If the US is smart, it will catch the companies as they fall." — from an essay written with Nathan E. Sanders and originally published in The Guardian
June IPO filings and falling valuations
The essay notes that in June OpenAI and Anthropic each filed for IPOs and briefly generated "buzz about trillion-dollar valuations." That optimism has already collided with market turbulence: the piece points to public backlash around AI datacenters, Nvidia’s slumping stock, and SpaceX’s newly minted stock price tanking weeks after its IPO as signals that investor sentiment can shift quickly. The authors argue the leading labs now face "strong headwinds" and that there are real questions about whether they will ever be sustainably profitable.
The economic case: training costs, rapid depreciation, and commoditization
The core economic critique in the essay is threefold. First, "frontier AI models are both expensive to train and depreciate within months, when a newer model appears," compressing the payback window for investors. Second, enterprise customers are "getting smart about minimizing AI token usage," and models are becoming commodities: the "best ones largely perform and behave similarly," which the essay says depresses prices. Third, free and open-source and Chinese competitors—"lagging only a few months behind"—give away models that the leading labs sell, and many of those free models "can be run locally" on private clouds, high-end servers, laptops, or even cellphones. Together, these dynamics lead the authors to conclude the unit economics of the current private-lab model may not sustain high valuations.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleTwo-part transition: national labs for innovation; utilities for compute
The essay proposes a concrete restructuring: separate each company into "product innovation" and "compute operations." The innovation arm could become a publicly managed national-lab–style entity under congressional oversight, reclaiming a mission orientation rather than venture-style incentives. Compute and datacenter operations, by contrast, could be managed "as a commodity resource, like public electrical or water utilities," with local or regional ownership, nationwide distribution, and strict regulation balancing ratepayer fees and infrastructure funding. The authors point to the US's existing $200bn R&D portfolio as precedent and frame frontier AI development as a current gap in that public R&D mix.
International precedents: Switzerland, Spain, Singapore, Germany and Australia
The essay stresses that national public AI efforts are not unprecedented. It says other countries, "including Switzerland, Spain and Singapore, are already operating public AI labs," and that Germany and Australia already run national supercomputing centers providing public access for running AI models. The authors use these examples to argue that public ownership and access models for compute and AI research have existing analogues that the US could draw from.
What this means for employees, Congress, and the public
- Employees: The essay acknowledges staff and researchers would "sacrifice hypothetical billions in equity" if labs were nationalized but frames the tradeoff as a return "to their roots and to their core mission" of safe, public-interest AI. Compensation, the authors argue, should be "aligned to the civil service" and not include "golden parachutes" or "outlandish pay rates."
- Congress and policymakers: The authors propose congressional oversight of publicly managed innovation labs and note Congress already manages a "$200bn R&D portfolio," implying legislative authority and budgetary mechanisms exist to create and govern such entities.
- The public: Under public ownership, models could be "open, transparent and responsive to the demands of the public rather than private shareholders," the essay says — trained on "only appropriately licensed data," refusing advertiser money, and focused on "maximizing the usefulness of AI to society" rather than pursuing speculative "artificial general intelligence" as a profit story.
Across the argument the authors repeatedly return to a single conditional: if the market reassesses and decides these firms "offer nothing of financial value," the public interest may be best served by taking them out of the private-equity cycle. The essay does not call for preserving current executive pay or investor windfalls; instead it frames nationalization as a rescue that realigns resources toward democratic oversight, scientific cooperation, and reduced environmental and resource waste from duplicated, hype-driven training runs.
Whether or not that contingency arrives, the essay leaves a clear test on the table: if "the bubble bursts," the authors say, the United States should be prepared to "catch the companies as they fall" and convert their capabilities into publicly governed national resources.
