NextFin News - Palantir CEO Alex Karp said on CNBC on Wednesday, June 10, 2026, that enterprise customers are “unhappy” with how the frontier AI labs are operating, and that many of the public projects those labs discuss are actually “running on Palantir.” He made the comments in New York during an interview with CNBC’s Sara Eisen.
The remarks fit a line Karp has pushed for years. The Palantir co-founder has repeatedly argued that his company’s software is the right way to bring artificial intelligence into government and corporate workflows, while questioning whether large model developers can turn technical advances into usable business products. His CNBC appearance was not a neutral readout on the market. It was a direct sales pitch from a chief executive arguing that application software, rather than model labs, will capture the most durable enterprise value.
Karp’s criticism lands on a real divide in enterprise AI buying. Frontier labs including OpenAI, Anthropic and Google DeepMind have improved model capability quickly, but companies buying AI need more than model access. They need data governance, permissions, audit trails, workflow integration, security controls and systems that let decisions be traced to a business owner. Palantir has tried to position itself in that part of the market, and Karp’s argument is that the frontier labs are still weak there.
His specific complaint was that customers think the labs do not understand their businesses and are too focused on “tokenmaxxing,” his term for burning through AI tokens to signal activity rather than solve concrete problems. That complaint mirrors a common frustration among enterprise buyers: AI pilots can look polished in a demo, then run into internal data problems, compliance rules and legacy systems once companies try to deploy them. Karp was pressing a competitive advantage, but he was also describing a procurement problem many large companies face.
He was sharper when discussing Anthropic. Karp said that “most of the things they talk about in public are running on Palantir,” suggesting that a significant share of high-profile customer deployments marketed around frontier models still rely on Palantir infrastructure underneath.
That claim deserves caution. It came from Palantir’s chief executive, not from customer disclosures, independent audits or a broad survey of enterprise buyers. Many organizations could be using both frontier models and Palantir products in different parts of the stack. That would not necessarily mean one depends on the other, or that model labs are losing the enterprise market. It does suggest the line between model provider and application provider is less clear than either side’s marketing implies.
For investors, the distinction matters. Karp’s comments point to where Palantir sees its advantage: not in building the best foundation model, but in controlling the operational layer companies use after the initial excitement fades. The enterprise AI market may end up split between model creators that drive broad usage and software vendors that make money from workflow control. Public markets have recently rewarded companies that can show clearer AI monetization, and Palantir has benefited as investors put more weight on contracts, deployments and recurring revenue than on conference-stage promises.
There is also a straightforward counterargument. Frontier AI labs are no longer just model vendors; they are building enterprise products, developer tools, governance features and partnerships with consultants and cloud platforms to make adoption easier. Their scale, research budgets and brand recognition still give them leverage, especially with companies that want to experiment broadly before settling on a vendor. Palantir may win some deployments and still face pressure if the labs keep improving the enterprise layer and reduce the need for intermediaries.
Karp’s statement is best read as a competitive thesis, not a final market verdict: enterprise customers want AI that behaves like enterprise software, and many are still dissatisfied with how frontier labs package and sell their tools. The clearest fact remains the one he put on the record on June 10: he told CNBC that enterprise customers are unhappy with the frontier AI labs, and that many of the projects those labs promote publicly are running on Palantir.
Explore more exclusive insights at nextfin.ai.
