Palantir's Karp Says Enterprises Are 'Fed Up' With Frontier AI Labs

Palantir CEO Alex Karp used a Wednesday interview with CNBC's Sara Eisen to deliver a blunt assessment of the AI companies preparing for public offerings: their enterprise customers are unhappy, and that frustration is steering business toward Palantir.

Karp claimed that every single enterprise customer Palantir works with is dissatisfied with frontier AI labs such as Anthropic and OpenAI. He described those companies as operating on what he called a "hyper religion of hyper optimism" that does not match what their customers actually experience when they try to deploy large language models.

"They believe all problems present, past, and future, including the ones they create but don't acknowledge, are going to be solved by them," Karp said. "Enterprises are fed up because they know this doesn't actually work this way, and isn't working."

The argument behind the bluster

Karp's central claim is that handing a business an LLM is not the same as solving a business problem. The value, he argues, sits in the implementation layer: the data integration, governance, and orchestration work that turns a model into something useful inside a real organization with messy, scattered data.

That is the pitch for Palantir's Foundry platform, which Karp positions as an AI-agnostic data integration system. The idea is that a customer unifies disparate data sources first, then connects whatever LLM they prefer on top. In Karp's framing, the model is a component, not the product.

Whether or not you accept the sales framing, the underlying numbers give the argument some support. AI deployments have been loss makers for many of the companies running them. A recent Gartner estimate cited in the discussion found that only 28 percent of AI use cases fully meet return on investment expectations, and most pilots never reach production. Despite that, business leaders keep funding new projects in search of value that, by Karp's account, only materializes when models are paired with serious infrastructure.

The 'tokenmax' complaint

Karp's sharpest criticism was aimed at how the labs measure success. He accused them of wanting customers to "tokenmax," treating token consumption as a proxy for productivity and usefulness. More tokens burned, in this view, gets read as more value delivered, even when the customer's actual problem remains unsolved.

The complaint is not unique to Karp. Google CEO Sundar Pichai referenced the same dynamic at the company's I/O event last month. And the economics are starting to bite: rising token costs are pushing companies to question their spending, and OpenAI is reportedly weighing a per-token price cut to compete with Anthropic, which Karp described as the "leading frontier firm."

Karp said Palantir leadership has gone so far as to debate paying prospective customers to go talk to the frontier labs directly before signing with Palantir. "People come out of there screaming, saying 'this could never work for me, they don't understand the enterprise, they don't care about my enterprise,'" he said.

Shots at OpenAI's deployment push

Karp also took aim at OpenAI's recent acquisition of UK consulting firm Tomoro, which is being folded into a newly launched OpenAI Deployment Company aimed at helping customers actually generate returns from their ChatGPT investments. He read the move as an attempt to copy Palantir's model and dismissed it as "a complete farce," adding that the labs "don't understand how unlikeable they are."

He was careful to separate the people from the product. Karp said he is friendly with some of the lab leaders and that they are pleasant to talk to, but maintained that "the product doesn't actually work and it's very expensive." He went further, claiming that much of what Anthropic touts publicly succeeds because it is "running on Palantir."

What it actually signals

Strip away the self-promotion and there is a real strategic claim here about where margin lives in the AI stack. "It is not that LLMs aren't crucial for the world, it's just that the implementation is where the value is, certainly in the next 7 years," Karp said. If he is right, the companies that win the enterprise market may not be the ones building the best models, but the ones doing the unglamorous work of connecting those models to data, workflows, and accountability.

Karp's delivery was, as usual, confrontational and self-aggrandizing, and his interest in talking down competitors is obvious. But the enterprise AI return figures are not his invention, and they line up uncomfortably well with his core point: that buying a model is not the same as getting value from one. For organizations still struggling to move AI projects past the pilot stage, that distinction is the one worth sitting with.