Back
Palantir CEO Alex Karp attacks AI token costs, backs open models

Palantir CEO Alex Karp attacks AI token costs, backs open models

Karp said U.S. government and enterprise customers are shifting toward open-source or open-weight AI in sovereign environments, reflecting rising focus on cost control, data sovereignty and operational control.

Fact Check
Every element of the claim is confirmed by primary sources. CNBC ('Palantir's Karp bashes token-based AI model as completely wrong') documents Karp attacking OpenAI/Anthropic token pricing and endorsing open-weight/proprietary approaches. Palantir's official X manifesto explicitly urges institutions to keep their data and control their model weights and criticizes tokenmaxxing. The NVIDIA newsroom blog and Palantir's Business Wire announcement confirm Palantir and Nvidia deploy Nemotron open models in secure sovereign/air-gapped environments with full weight ownership. No conflicting evidence found.
Summary

Palantir CEO Alex Karp criticized OpenAI, Anthropic and other frontier AI labs for pursuing “tokenmaxxing,” arguing that enterprises and U.S. government customers are increasingly frustrated by rising usage-based costs, wasted spending and risks to proprietary value and intellectual property. Karp, Palantir and supporter David Sacks argued that AI safety and sovereignty depend on keeping data, model weights and compute under an organization’s control, a view reinforced by Palantir and Nvidia’s June 29 plan to deploy Nvidia Nemotron open AI models in sovereign environments for U.S. government and critical infrastructure clients. The reported shift by government buyers toward open-source or open-weight AI also matters beyond defense-related stocks because public-sector procurement can influence broader enterprise technology standards and software adoption.

Terms & Concepts
  • AI sovereignty: Keeping AI systems, data and related assets under an organization’s own control and jurisdiction.
  • model weights: The parameters that determine how an AI model behaves and generates outputs.
  • sovereign environments: Secure or locally controlled computing setups designed to keep sensitive data and AI systems under a customer’s own jurisdiction and control.