A short history of US nationalization — and what it means for AI

Summary

Anthropic's dispute with the Pentagon, which labeled the company a supply chain risk, was the first time an American AI firm received that designation. Dario Amodei has since offered a public apology, describing it as one of the most disorienting crises in Anthropic's history. Whatever form government involvement in AI ultimately takes, that standoff is unlikely to be the last deal made between Washington and an AI lab.

Government control across US history

The history of US nationalization runs from light-touch equity stakes to full state operation. Tyler Cosgrove, writing in the TBPN newsletter, maps the terrain across two axes: the strength of nationalization, ranging from a weak minority stake to complete government operation, and the reason, which is either economic distress or national security.

The anchor points run from 1791 to the present. The First Bank of the United States gave the federal government an equity stake in what would otherwise have been a private institution, arguably the earliest example of public-private ownership. The Civil War Railroad and Telegraph Act of 1862 gave the president authority to seize railroads and telegraph lines when public safety required it, and the same logic recurred in World War One. The federal government built and operated the Alaskan Railroad from 1914 until 1985, then handed it to the state of Alaska, establishing a precedent for temporary nationalization that later reverts to private or state control. The Tennessee Valley Authority, created in 1933, had the federal government electrify rural areas where private capital would not go; the TVA still operates today and has appeared in discussions around data center power supply. The Manhattan Project in 1942 was not a takeover of an existing private company but a de novo government-run industrial program spanning an entire sector. During World War Two, FDR seized meat packing facilities and petroleum plants, often due to labor disputes, and some of those seizures extended into the postwar period. Amtrak and Conrail emerged in 1970 after private operators let rail lines deteriorate. The Chrysler rescue in 1979 relied on loan guarantees rather than equity, with the government backstopping the debt without taking formal ownership. The TSA replaced private airport security contractors with a federal agency after 9/11, a clean national-security-driven nationalization that is rarely framed that way in public debate. The 2008 conservatorship of Fannie Mae and Freddie Mac and the GM bailout followed; GM mattered both as a consumer employer and as a defense supplier. The Trump administration's recent equity stakes in Intel and MP Materials involve no reported board seats and sit closer to a grant with upside than a nationalization, motivated by keeping national champions solvent rather than directing their operations.

Where AI fits

Leopold Aschenbrenner's Situational Awareness argument for a government-run Manhattan Project for AI sits in the strong-military quadrant. Amodei's public framing lands differently. He talks about AI's economic disruption, including the potential disappearance of 50 percent of entry-level white-collar work, and the government's role in responding to it, which places him closer to the weak-economic quadrant and nearer to the 2025 Intel deal than to the Manhattan Project.

Amodei's positioning is harder to pin down than that framing suggests. He has compared sending Nvidia GPUs to China to giving nuclear weapons to North Korea, and his most-gifted book is The Making of the Atomic Bomb. He appears to see multiple scenarios as live possibilities, and what he advocates for at any given moment tracks who is in power and what the administration is actually prepared to do.

The AI 2027 forecast adds a third scenario in which two parallel labs, one private (modeled as "OpenBrain") and one government-run, compete rather than merge. That would be novel in US history, closer to the Cold War dynamic between NASA and private aerospace than to either a clean nationalization or a pure market outcome.

Where the definitional lines sit

A grant is not nationalization. Equity-for-funding starts to look like it. The Defense Production Act lets the government jump to the front of any contractor's production queue without taking ownership, which is control without equity. The most extreme scenario, barring private actors from training frontier models entirely, remains theoretical but is the logical endpoint of the strong-military quadrant, analogous to prohibiting private nuclear weapons development.

Amodei's interview with The Economist, notably sponsored by Anthropic, was published the same day he described the Pentagon episode as one of the most disorienting in the company's history. An AI CEO who reads atomic bomb history, compares GPU exports to nuclear proliferation, and has just been labeled a supply chain risk by the US military is now trying to define what a cooperative relationship with Washington actually looks like.