Alex Karp has spent two decades selling software to governments that guard their secrets. On September 17 he told CNBC that the frontier AI labs may be carrying too much to hide to ever go public.
The Palantir chief executive argued that the leading AI companies bear liabilities so large that no public market could absorb them, and that the only exit is for government to take them over. His comments, made on CNBC’s Squawk on the Street, circulated widely this week.
The exchange turned sharp when an interviewer asked how such risks would be written into an S-1, the registration document a company files before listing. “You’re assuming that there will be an S-1,” Karp replied. “Okay. Maybe there will be.”
He went further, laying out what he sees as the only workable path. “The only way to deal with this kind of liability is to go to the government and say, ‘Nationalize us, please,'” he said. “A lot of times in business, people don’t say what they want. They’re leading you to the conclusion.”
Karp’s argument is that the labs themselves are steering toward state ownership because it would cap the liability they carry. If a model built with a customer’s data does harm, he reasoned, the company’s own clients would sue. Only the government, he said, could absorb risk on that scale, perhaps by taking a stake of around half the company.
He put the size of the risk in concrete terms. A company that proposes it could destroy a tenth of the world, he argued, is an unacceptable bet for investors, and government is the only backstop large enough to hold it. No other industry, he said, gets to shrug off the risks it creates and insist that regulation is the answer.
The claim carries a commercial edge. Palantir sells software that lets enterprises and governments keep control of their own data, and Karp said his customers are angry about their proprietary information ending up inside models that a competitor can then query. The nationalization thesis is also a pitch for Palantir’s alternative: keep the data, and the liability, out of the shared model.
Palantir, founded in 2003, went public through a direct listing in 2020 and has since become one of the more valuable software companies in the country, selling to intelligence agencies, militaries and large corporations. Karp’s warnings about the labs come from a company that built its name guarding information the labs would like to ingest.
The timing gives the comments extra weight. OpenAI confirmed in June that it had confidentially filed a draft S-1 with the Securities and Exchange Commission, and Anthropic made its own confidential submission a week earlier. Neither has made a public filing, and both have kept their financials private.
Then the schedules slipped. Sam Altman said in an interview published in mid-September that now would be an ill-advised moment to go public, ruling out a listing in 2026. The Wall Street Journal reported that OpenAI’s offering has been pushed from October to November, with a target valuation around $2 trillion.
The delays predate Karp’s remarks but land in the same direction. A company that cannot decide how to describe its worst-case risks to investors is a company that finds going public complicated. Karp’s point is that the risks may be indescribable in a prospectus at all.
This is not a new position for Karp. He has said he spent six months privately warning AI executives about nationalization, and told an interviewer in June that the momentum sits with people who want to nationalize the labs. He acknowledged he does not know whether the White House would pursue any of it.
No lab has asked to be nationalized, no government stake exists, and no public S-1 has been filed. Karp was describing a route he expects others to take, not one anyone has taken yet.
Skeptics note that Karp is a competitor. His customers would be on the plaintiff side of the liability he describes, and a collapse in frontier-lab valuations would do Palantir no harm. His description of rivals angling for state cover is a guess about their intentions, stated as conviction.
Still, the question he raised will not go away. Two of the most valuable private companies in the world have promised to list and then pushed the date back. Karp’s answer to the S-1 question, “You’re assuming there will be one,” turns a scheduling problem into a structural one.
Whether the labs ever reach public markets, the debate over what they would have to disclose to get there has already begun. Karp has drawn the hardest line: that the risk is too big for any market to price, and that the only institution big enough to hold it is the state.
