Palantir Technologies CEO Alex Karp is trying to make the AI market’s most technical billing unit into a boardroom risk. In a Wednesday appearance on CNBC’s “Squawk Box,” Karp criticized OpenAI and Anthropic over token-based pricing, arguing that enterprise customers are paying for AI usage without capturing enough value while exposing intellectual property, operational know-how, and competitive “alpha.”
“I’m going to chillax and waste my time with tokens, I’m going to get no value, and they’re going to get my IP,” Karp said.
That is not just a gripe about invoices. It is Palantir’s clearest attempt to reframe enterprise AI adoption around ownership, not access.
Karp’s argument lands at a moment when AI spending is becoming harder to translate into productivity returns. OpenAI’s API pricing is still explicitly token-metered, with GPT-5.5 standard pricing listed at $5 per 1 million input tokens and $30 per 1 million output tokens, while priority pricing rises to $12.50 per 1 million input tokens and $75 per 1 million output tokens. Anthropic also prices Claude by million-token units, with Claude Sonnet 5 listed at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, before standard pricing moves to $3 and $15.
Sponsored · Altamira Gold Corp.
For AI labs, token pricing is the meter. For Karp, it is the vulnerability. His pitch is that enterprises should not rent intelligence through recurring usage bills if the work involves sensitive workflows, decision logic, proprietary processes, or defense-related systems. Instead, he argued for deploying open-weight models through Palantir, where customers retain more control over compute, data stacks, models, and operational deployment.
That position directly supports Palantir’s own commercial strategy. The company’s Artificial Intelligence Platform has been one of the fastest-growing enterprise AI businesses in the public market. Palantir reported Q1 revenue of $1.63 billion, up 85% year over year. US commercial revenue jumped 133% to $595 million, while US government revenue rose 84% to $687 million. The company raised its 2026 revenue forecast to between $7.65 billion and $7.66 billion.
The numbers matter because Karp is not attacking AI labs from the outside. He is competing for the same enterprise budgets, but with a different story about where AI value should sit.
OpenAI and Anthropic sell access to frontier models. Palantir sells controlled deployment inside complex organizations. The distinction becomes more important as AI moves from chatbots into supply chains, banks, weapons systems, hospitals, and regulated data environments.
Mediaite reported that Karp’s CNBC appearance began around Palantir’s expanded Nvidia (NASDAQ: NVDA) partnership, including integration of Nvidia’s Nemotron models into Palantir’s Sovereign AI platform for secure enterprise and government deployments.
Karp’s comments also fit Palantir’s recent “AI sovereignty” messaging. Business Insider reported that Palantir’s nine-point manifesto criticized “tokenmaxxing” and warned institutions against transferring strategic data too freely.
OpenAI says that, by default, it does not use business data to train models unless a customer explicitly opts in. It also says customers retain rights to their inputs and own eligible outputs. Anthropic says it does not use inputs or outputs from commercial products such as Claude for Work, the Anthropic API, and Claude Gov to train models by default.
That means the real fight is not only whether AI labs train on customer data. It is whether enterprises trust external model providers enough to place mission-critical knowledge inside their systems at all.
Karp made that trust problem explicit.
