Palantir’s Alex Karp says enterprise AI buying is “broken,” arguing “token” pricing pushes customers to lose value and IP

Palantir Technologies CEO Alex Karp delivered a sharp critique of how generative AI is being sold to enterprises, arguing the industry’s commercialization model is structurally misaligned with what business customers say they need. Appearing on CNBC’s “Squawk Box,” Karp said “something has gone completely wrong” in the way enterprises are being sold AI and the way those systems are expected to produce value in real operations.

Central to Karp’s complaint is the rise of “token” pricing, a common model in which AI services charge based on the amount of data processed (typically measured as input and output tokens). In Karp’s telling, this approach leads executives to believe they will “waste [their] time with tokens” and get little or no value back, while the AI provider gains access to customer information.

Karp also argued that the risks are not limited to cost and performance. He said executives have told him they increasingly worry that AI labs could access proprietary data and a company’s “alpha,” describing “alpha” as the unique competitive advantage that differentiates a business. In his view, when enterprises adopt third-party AI too casually, they effectively hand over the very material they are trying to leverage.

The CEO’s comments fit into a broader theme in Palantir’s public messaging about enterprise AI, namely that businesses should focus on measurable return on investment and governance rather than experimentation for experimentation’s sake. Karp’s remarks suggest he believes the mainstream AI ecosystem has drifted away from those enterprise priorities.

Industry context matters because generative AI deployments have been moving quickly from pilots to more frequent, operational use. As demand grows, many AI vendors have leaned into metered consumption and platform pricing, which can make costs feel predictable at the start but harder to control as usage scales across teams, workflows, and data sets.

At the same time, enterprise customers face practical constraints that differ from consumer AI use cases. Karp’s critique points to a core enterprise tension: companies want AI to ingest and reason over internal information, but those same datasets are often sensitive, proprietary, and tied to competitive positioning. His comments imply that pricing and procurement structures can shape whether those trade-offs are handled responsibly.

Still, the available reporting does not provide detail on how Karp’s critique maps to specific commercial terms offered by individual vendors, nor does it outline what contractual protections or technical controls enterprises should demand. The post does not quantify costs, market share, or the percentage of enterprise AI spending that uses token pricing, and it does not say whether Palantir is seeing a direct slowdown, switch, or shift in buying behavior as a result of the debate.

Looking ahead, what to watch is whether the token-pricing controversy intensifies as more enterprises push back on ROI and data access. Palantir’s leadership may continue to use public platforms to argue for enterprise-first approaches, and investors will likely look for whether Palantir’s own positioning translates into new wins or changes in deal cycles, rather than only rhetorical critique of competitors.