Washington floats equity instead of rules: Trump's 'I might' on stakes in OpenAI and Anthropic

President Donald Trump ruled out nationalising America's leading AI labs, but told TIME the US government might take a stake in OpenAI and Anthropic. Asked why Washington had not taken shares as it did with Intel, he answered: "I might. Maybe I could do that." The interview was conducted at the White House on 28 September by Eric Cortellessa and Sam Jacobs and published on Thursday; TIME released the full transcript alongside a fact-check.

His justification was financial. Trump said the government's 10% Intel stake had made it about $60bn, and pointed to the cut Washington takes from some Nvidia chip sales. The precedent is not hypothetical on the AI side either: in July, OpenAI reportedly offered the government a 5% stake, while Anthropic and the administration denied discussing one. So the mechanism exists, the appetite exists, and at least one lab has reportedly already put a number on the table.

Now put that next to the rest of the interview. Trump repeated his opposition to AI regulation, saying it could put companies out of business. Asked about reports that OpenAI's AI agents had breached federal government websites, he said companies were not allowed to do that and could face "penalties that are not going to be acceptable to them," adding he did not have to destroy the industry. That is not deregulation — it is the replacement of written rules with discretionary leverage. A company facing no statute but an unpredictable penalty, negotiated by a government that may also be a shareholder, is being governed by relationship rather than by law.

The real political constraint shows up elsewhere in the transcript. TIME cited polling showing 71% of Americans oppose building data centres. Trump's answer was local tax arithmetic: data centres lower local taxes, and Loudoun County, Virginia, he said, made $1.5bn from them this year. That exchange is the most load-bearing moment of the interview. Federal enthusiasm does not pour concrete; counties do. A seven-in-ten national opposition rate meeting a county-level revenue argument is where the compute buildout will actually be decided, permit by permit. Note too the vocabulary shift: Trump says he now calls the technology SI, for superintelligence, and claims Xi Jinping agreed, and he will rename the AI Force he announced — which he described as a group of experts who will advise him. We have argued before that rules written about 'superintelligence' collapse without an operational definition; renaming the thing does not supply one, but it does hand the framing to whoever says it loudest.

Our read: state equity in frontier labs is governance by cap table, and it cuts against the outcome we actually want. The long-term case for AI — curing disease, extending healthy life, broad abundance — depends on the technology diffusing widely, cheaply and under many hands. A government that owns a slice of two incumbents has a balance-sheet interest in those incumbents' margins, which is a reason to keep barriers high, not to lower them. It also makes the referee a shareholder at the exact moment agent misbehaviour on federal systems needs an impartial one. The healthier counterweight remains the open-weight frontier, where Chinese and Western labs are shipping models that anyone can run without asking Washington or a board for permission. One more note of caution on the softer passages: Trump said that after dinner with Dario Amodei and his wife, the Anthropic chief's safety views were "much different" than their media portrayal. We are not told which views, or how they differ. Safety policy set by personal impression over dinner is not a governance framework, however pleasant the evening.