Palantir CEO Alex Karp Attacks OpenAI and Anthropic's Enterprise AI Model on CNBC

Palantir CEO Alex Karp appeared on CNBC's Squawk Box on July 1 to discuss the company's new sovereign AI partnership with Nvidia. What he delivered instead was a roughly three-minute critique of the frontier AI business model, naming OpenAI's Sam Altman and Anthropic's Dario Amodei while insisting he was not attacking either of them personally.

The line that spread fastest was Karp relaying what he says enterprise customers tell him in private: that they are livid, "paying for tokens that create no value," and watching frontier labs take the weights and alpha of their businesses. When co-anchor Becky Quick told him he sounded angry, Karp called himself the voice of American business. When Andrew Ross Sorkin said the repeated disclaimers about Altman and Amodei sounded like shade, Karp replied that it was reporting, not shade.

The market reacted quickly. Palantir shares rose roughly 9% on the day of the interview while some other AI names dipped.

The argument itself

Underneath the theater is a real thesis about how enterprise AI value gets divided.

Karp's claim is that frontier labs profit from the same customer more than once. They charge per token consumed. They also gain exposure to the customer's proprietary workflows, data structures, and institutional knowledge as it flows through their models. Over time, he argues, that accumulated intelligence erodes the advantages that made those customers competitive in the first place.

His sharpest jab was about pricing. If a model could genuinely make a company a billion dollars, Karp asked, why wouldn't the lab demand a percentage of that value instead of charging by the token? In his telling, consumption pricing is what you charge when you are selling compute rather than outcomes.

It is a clever argument, and a self-serving one. Karp acknowledged on air that he profits from the same anxieties he was naming. Palantir's entire pitch is that enterprises should control their compute, their models, their data stack, and their alpha. A world where CEOs distrust the labs is a world where Palantir's on-premises positioning sells itself. The Nvidia partnership he was there to promote is built on that exact promise.

Why this matters for brand

Set the token economics aside and this is a trust story, and trust is a brand asset.

For two years the frontier labs have had something close to a halo with enterprise buyers. AI transformation was a board-level mandate, the labs were the obvious suppliers, and questions about data usage and IP leakage mostly lived in procurement fine print. Karp pulled those questions onto live television and attached the words "stealing" and "wealth tax" to them.

Whether or not the characterization is fair is a separate question. The major labs' enterprise agreements exclude customer data from training by default, a detail absent from the viral clips. But the accusation is now public, and every enterprise AI sales conversation in the second half of 2026 happens downstream of it. Security and legal teams who were already asking where their data goes now have the CEO of a company worth several hundred billion dollars validating the concern on CNBC.

The risk for OpenAI and Anthropic is not that enterprises rip out their models tomorrow. It is that the default posture shifts from enthusiasm to suspicion. Suspicion lengthens sales cycles, invites competitors into deals, and reframes the labs from partners into vendors to be contained.

For Palantir, the play is also brand-driven. Karp is positioning the company as the champion of enterprise sovereignty: your models, your weights, your alpha, on your terms. Agree with him or not, it is a coherent narrative delivered with total conviction, and the stock's reaction suggests investors think it landed.

Where this goes

Several things are true at once. Frontier models remain the most capable systems available and enterprise adoption keeps growing. Some of the frustration Karp describes reflects deployment failures, pilots that never scaled and use cases that never justified their bills, which are as much the customer's fault as the vendor's. And Karp has an obvious commercial interest in narrating a rebellion.

But he did not invent the anxiety. He branded it. He gave a name and a villain to a diffuse unease that has been building in boardrooms about dependency, data leverage, and who captures the value AI creates.

He closed by daring viewers to test his claim, investors especially, by privately calling two or three enterprise CEOs and asking if they are as frustrated as he says. It is a confident move because it cannot be checked on air, and it is devastating if even half true.

So who owns the enterprise AI relationship when the dust settles? Right now the labs own the capability, the hyperscalers own the distribution, and companies like Palantir are fighting to own the trust layer in between. Karp's CNBC appearance was a land grab for that middle position, executed in prime time. The labs would be wise to answer it with more than a press release.