Palantir CEO Slams OpenAI, Anthropic Token Model

Palantir CEO Alex Karp recently criticized the token model used by some U.S. AI labs, including OpenAI and Anthropic. He stated that "something has gone completely wrong" with how these models are sold, highlighting a focus on "tokenmaxxing." Karp noted that many enterprise customers are frustrated by rising token costs. This frustration is reportedly driving interest in open-weight models as an alternative. His comments come as Palantir announced an expanded partnership with Nvidia to build custom AI models for U.S. government agencies. The issue revolves around the economics of per-token billing for hosted large language model APIs. This pricing can significantly increase operational costs for companies with high-volume workloads. Enterprises are now looking at options like reducing model size, moving inference on-premises, or using open-weight models to manage costs and maintain data control. This matters because businesses need predictable costs and data sovereignty when deploying AI, influencing their choice between vendor APIs and self-hosting solutions.