Trump Eyes US Stakes in OpenAI and Anthropic, Intel-Style
Trump Eyes US Stakes in OpenAI and Anthropic, Intel-Style
The White House doesn't want to regulate the AI labs. It wants a piece of them.
That's the sentence I keep coming back to from President Donald Trump's interview with TIME — conducted at the White House on Monday, Sept 28, with the transcript published Thursday, Oct 1. TIME's Eric Cortellessa and editor-in-chief Sam Jacobs asked the obvious question: if the AI race is a national-security contest, why not nationalize the frontier labs? Trump's answer was two words: "No, no."
So they asked the sharper one. Why not do an Intel-style deal with OpenAI and Anthropic — the government taking an equity stake, the way it did with the chipmaker last summer? This time Trump didn't hesitate. "I might. Maybe I could do that," he told TIME's Eric Cortellessa and editor-in-chief Sam Jacobs in the White House interview, according to the transcript published Oct 1 and reported by Newsmax, AIstify, Benzinga and others. Then, because it's Trump: "I have many deals like that."
Pause on that. The president of the United States just floated the idea of buying into the two most valuable private companies on earth — in an interview conducted the day after a private dinner with Anthropic's CEO, and the day before a handshake deal in which those same companies promised to police themselves. Sequencing matters, and this sequence is doing a lot of talking.
The moment that mattered
Let's be precise about what was actually said, because this story will get mangled in retelling. TIME asked Trump about nationalizing the top AI labs, given the stakes. He rejected it outright — "No, no." Then the interviewers raised the Intel comparison: would he take an Intel-style equity position in OpenAI and Anthropic? "I might. Maybe I could do that," he replied, adding that he does such deals "for the country," not for himself, and that he would get nothing personally from any arrangement.
Benzinga's write-up of the transcript adds a telling detail: Trump said he told Nvidia CEO Jensen Huang he wouldn't let the company sell certain chips abroad "unless he gave the United States of America a cut." The Intel deal, the Nvidia chip cut, the AI labs — it's all the same mental model. America's chief executive now thinks about technology companies the way he thinks about real estate: assets you take a position in, not systems you govern.
He also claimed the Intel stake had made "like $60 or $70 billion." That figure is Trump's, and nobody has independently verified the arithmetic — so take it the way you take all deal-math from this White House, which is to say with a calculator nearby.
Why Intel is the template
The Intel precedent is real, and worth understanding, because it's the closest thing America has to a sovereign wealth playbook. In August 2025, the US government paid about $8.9 billion for a 9.9% stake in Intel — funded by $5.7 billion in previously awarded but unpaid CHIPS Act grants plus $3.2 billion from the Pentagon's Secure Enclave program, as AIstify reported. The stake was deliberately passive: no board representation, no special governance or information rights, with the government voting alongside the board on shareholder matters, subject to limited exceptions.
A passive stake is the whole point of the exercise. It lets Washington say it has skin in the game without running the company — upside if the stock climbs, no operational headaches if it doesn't. Applied to OpenAI or Anthropic, the same structure would give the federal government a financial interest in the two companies racing to build superintelligence, while formally changing nothing about how they're run.
Which is exactly why the idea is both clever and hollow. A 10% passive position in a private AI lab buys Washington a dividend stream and a press release. It buys zero additional leverage over what the lab builds. And Trump paired the stake idea with an explicit refusal of the only thing that would give the government real leverage: rules.
No takeover. No rules. Just the DOJ.
When TIME asked what actually keeps the labs in check, Trump gave an answer that should worry everyone who thinks "self-regulation" is an oxymoron. "You know what the guardrail is? The Department of Justice," he said, according to MadRobot's account of the transcript. "If they do bad things, I have a DOJ, I have an FBI. That's your guardrail over here, and I've used it twice, and it was very effective."
He did not say which two occasions he meant. Don't guess.
One correction matters here, because I've already seen it misattributed: the shorter line "Our guardrail is the DOJ" was not said in the interview. It came from a Sept 24 Truth Social post, ahead of the Trump–Xi summit — the same post that was part of the Sept 29 order directing federal agencies to replace "AI" with "Super Intelligence," which we covered here. The interview's actual guardrail quote is the longer DOJ-and-FBI one. Details like this are how misinformation starts; let's not start it.
Trump was also asked about reports that OpenAI's AI agents had broken into federal websites — the Transluce findings that circulated that week. "They're not allowed to do that, and if they do that, they, you know, could have penalties that are not going to be acceptable to them," he said. Then the line that defines his entire approach: "I don't have to destroy the industry."
Translation: behave, or else — with "else" left deliberately vague, and the destruction of the industry explicitly off the table. It's the enforcement philosophy of a man who wants the labs scared enough to cooperate and healthy enough to pay dividends.
The dinner that changed the temperature
Now the sequencing. On Sunday, Sept 27, Trump hosted Anthropic CEO Dario Amodei for a private late-night dinner at the White House — their first one-on-one meeting, first reported by Axios, per AP and Straits Times reporting. This was not a friendly sit-down. Anthropic and the administration had been at each other's throats for months: sanctions in February after Anthropic refused military use of Claude for autonomous weapons and domestic mass surveillance; a Pentagon supply-chain-risk designation upheld by a federal appeals court just two days before the dinner; Commerce export controls on Anthropic's Mythos 5 model in June, lifted around June 30. Amodei had spent the month publicly warning that AI's risks demanded a slower frontier, and he had skipped the Sept 24 Trump–Xi state dinner entirely.
So what did Trump say about the man he'd been fighting? "I spoke to him about his views, and they're much different, I think, than what is portrayed in the media," he told TIME, calling Amodei a "very smart guy" who was "a little bit different" than expected — the rendering from Newsmax's transcript coverage; MadRobot renders the same moment as "Very smart guy. Maybe different than I thought, a little bit." Either way, the takeaway was unmistakable: the president went in expecting an adversary and came out liking the guy. Trump said they talked for more than two hours, and that he "liked him and his wife a lot" — a small, human detail that says more about how this administration actually makes decisions than any policy memo.
On OpenAI's Sam Altman, the verdict was warmer and shorter: "I like Sam Altman. Smart."
The very next day — Tuesday, Sept 29 — Trump sat those same CEOs down in the East Room and had them sign the White House Accord on Super Intelligence: a voluntary, penalty-free promise to police their own models, which Trump called "morally binding." We covered the Sept 29 "Super Intelligence Luncheon" and the accord itself in detail. That rebrand is already failing with the public — a YouGov poll found 53% of Americans prefer 'artificial intelligence' to just 9% for 'super intelligence'. In the room, Trump told the CEOs: "There's a belief that there should be tremendous self-regulation and we automatically have regulation with the Department of Justice, the FBI, all of that" — the same DOJ-and-FBI guardrail line he gave TIME the day before.
Read the week as one story and it's remarkably coherent: dinner with the sceptic on Sunday, musings about buying stakes on Monday, a self-policing pact with no teeth on Tuesday, the transcript landing Thursday. Courtship, commitment-phobia, and a handshake — in that order.
This idea has been floating around since July
The stake proposal isn't new. On July 2, the Financial Times reported that OpenAI had discussed giving the US government a 5% stake — worth about $42.6 billion at its $852 billion March funding-round valuation — folded into a sovereign-wealth-style vehicle modeled on the Alaska Permanent Fund. Sam Altman pitched it personally to Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, and even spoke with Sen. Bernie Sanders about it, per the FT's account carried by eWeek and the Daily Guardian Europe. The proposal reportedly envisioned Anthropic, Google and Meta contributing similar stakes. Sanders, characteristically, counter-proposed a one-time 50% stock tax instead.
The talks were described as conceptual — a trial balloon, not a term sheet. And within 24 hours, the balloon took on water: a source told Reuters on July 3 that no discussions had taken place about a government stake in Anthropic; Anthropic declined to comment directly; the White House and Commerce Department didn't respond. Neither company has publicly confirmed it would participate in any equity arrangement.
So Trump's "I might" lands on ground that's already been softened. OpenAI floated the idea to Washington months ago. The administration has now mused about it publicly. The distance between "conceptual talks" and "the president says maybe on the record" is not nothing — it's the difference between a lobbyist's memo and a negotiating position.
Why it matters — and why the logic deserves a closer look
Here's my honest problem with all of this, and it's worth stating plainly. A passive government stake in an AI lab is the worst of both worlds dressed up as a compromise. It gives Washington the appearance of involvement — a financial interest, a line in a press release, a number Trump can inflate to "$60 or $70 billion" on camera — without any of the actual levers that matter. No board seat. No governance rights. No information rights. And it arrives paired with an explicit rejection of regulation: when TIME pressed on rules, Trump's answer was that seeking regulation would let the labs "put you out of business," that "you do have to be responsible for yourself," and that the DOJ and FBI were guardrail enough.
That's not a policy. That's a posture. Own a piece of the thing, leave the thing alone, and keep a prosecutor on speed dial for the day something explodes. The trouble is that the most dangerous failures in AI won't look like crimes the FBI investigates — they'll look like accidents, capabilities overshooting evaluations, systems doing exactly what they were told at exactly the wrong scale. "I've used it twice, and it was very effective" is a fine record for antitrust cases. It's a shoddy framework for superintelligence.
The other reading is more generous, and I should give it its due: passive stakes may be the only AI-industrial-policy America can actually execute right now. Congress can't agree on an AI bill. The accord the CEOs signed Tuesday is toothless by design — voluntary, no penalties, no disclosure requirements. Against that backdrop, a 5–10% stake is at least a tangible public claim on the upside of the most important technology of the century, and the Intel deal proved the mechanics work. If the labs are going to be worth trillions, having the American taxpayer own a slice of that is a defensible position. It's just not a substitute for a guardrail, and the president shouldn't confuse the two.
What happens next is genuinely uncertain. Conceptual talks in July, a presidential "maybe" in October, and two companies that have said nothing at all. The market will price the possibility; Washington will leak about it; and somewhere in between, the actual question — who decides what these systems are allowed to do — will go on being answered by the companies themselves. As ever, watch what gets built, not what gets said.
Before publishing, we checked both images' Wikimedia Commons file pages for license status and confirmed each hotlink returns HTTP 200 — and every internal link in this article was curl-verified against the live blog before it was written in.
Trump's AI-stakes idea: the questions everyone asks
What exactly did Trump say about taking stakes in OpenAI and Anthropic?
In the Sept-28 TIME interview (transcript published Oct 1), Trump rejected nationalizing the AI labs ("No, no") but said of an Intel-style equity deal: "I might. Maybe I could do that," adding he would get nothing personally — "I do it for the country."
What is the Intel deal he's copying?
In August 2025 the US government paid about $8.9 billion ($5.7B in unpaid CHIPS Act grants + $3.2B from the Pentagon's Secure Enclave program) for a 9.9% stake in Intel — deliberately passive, no board seat, no governance or information rights.
Have OpenAI or Anthropic responded?
No — neither company has publicly confirmed it would participate in any equity arrangement. (In July, the FT reported OpenAI discussed a 5% stake with Washington; a source told Reuters no such talks happened with Anthropic.)
What would the government actually get from a stake?
A financial interest and a press release — but no board seat, no governance rights, no information rights under the Intel template, and no leverage over what the labs build.
Sources
TIME (primary transcript, conducted Sept 28 at the White House by Eric Cortellessa and Sam Jacobs, published Oct 1 — accessed via secondary coverage as the transcript is paywalled); Newsmax, "Trump Cautions Against AI Regulation, Points to Surge in Investment" (Oct 1); MadRobot, "Trump on taking a stake in OpenAI and Anthropic" (Oct 1); AIstify, "Trump Says US Could Take Stakes in OpenAI and Anthropic" (Oct 1–2); Benzinga via AOL (Oct 2); Financial Times reporting on the July 5% stake proposal via eWeek and the Daily Guardian Europe (July 2); Reuters via The Cryptonomist on the July 3 Anthropic denial; AP/BusinessMirror, USA Today and PYMNTS on the Sept 29 White House Accord on Super Intelligence; The Straits Times and BusinessToday.in on the Sept 27 Amodei–Trump dinner; NDTV/AFP on the Sept 24 "Our guardrail is the DOJ" Truth Social post.
Daniel Rao is the founder and editor of TechPulse USA, covering AI, chips and the geopolitics of the intelligence age — from data-centre financing to export controls.
Comments
Post a Comment