Schneier Calls For US Nationalization Of AI Giants

The official account, as offered by Mr Schneier and Mr Sanders, runs thus: artificial intelligence is a technology of such consequence that its ownership cannot safely be left to markets alone, and should the markets reject OpenAI or Anthropic - should the money simply stop coming - the American government must be ready to take the companies into public hands, as one would a bank whose failure threatens the whole system. The machinery, however, works rather differently, and it is worth being precise about how.

A government does not nationalize a thing by wanting to. It nationalizes a thing when it already, in practice, controls the conditions of that thing’s survival, and the formal transfer of ownership merely dignifies an operational reality that preceded it. This is the old lesson of the Bank of England, dressed in new clothes: the central bank became the lender of last resort not by statute first but by habit, by the accumulated expectation that when the system wobbled, it would act, until the acting became indistinguishable from its constitution. The proposal to nationalize an OpenAI or an Anthropic assumes the state would be stepping into a vacuum. It would not. It would be stepping into a boardroom already crowded with Microsoft’s compute contracts, Amazon’s cloud arrangements, and the informal but very real understanding - reached across an unglamorous number of closed meetings in Washington - that these firms are, for national security purposes, already treated as instruments of the state whenever it suits the state to treat them so.

Here is the gap, then. The dignified proposal is democratic control in the event of market failure. The efficient reality is that the American government has never needed to own these companies to direct their most consequential behavior; it needs only to control export licenses for advanced chips, set the terms of federal procurement, and reserve the right to declare a frontier model a matter of national security. Nationalization, in this light, is not a rescue plan for a crisis that has not yet occurred. It is a redundant ceremony for a control that already exists in substance, waiting for a moment when it becomes convenient to exist in form as well.

Consider, too, who is left out of Schneier and Sanders’s proposal and why that omission is instructive. Google and Meta are named among the interested parties but are curiously absent from the nationalization scenario, and this is not oversight but structure: they are diversified enough, and cushioned enough by advertising revenue, that they are the last firms likely to face the market rejection the proposal contemplates. It is OpenAI and Anthropic, dependent on continuous capital injection and unable to fall back on a search monopoly, who are structurally exposed to the scenario Schneier and Sanders describe. The proposal, in other words, is not really about artificial intelligence as a category. It is about the fragility of a particular financing model - enormous capital expenditure against speculative future revenue - dressed in the vocabulary of civilizational risk.

The convention that actually governs this situation is the one Britain learned in Threadneedle Street a century and a half ago: the state will act at the last moment, decisively, and will afterward insist it had been planning to do so all along. Nobody in Washington need announce today that OpenAI is too important to fail. They need only wait until the moment it might fail, at which point the same officials who spent years insisting on the sanctity of private enterprise will discover, with impressive speed, doctrines of essential infrastructure they had somehow neglected to mention earlier. This is not duplicity. It is how confidence is actually maintained - by never committing to the rescue until the rescue is unavoidable, so that markets continue pricing risk as though the rescue might not come.

What the official account obscures, then, is that the interesting question is not whether the government should nationalize a failing AI company. It is how much unowned control the government already exercises through chip export rules and procurement contracts, such that formal ownership would change less than either its advocates or its opponents suppose - a change of title on a house the state already keeps the keys to.