Palantir Criticizes AI Token Model as Costs Surge

Palantir CEO Alex Karp has criticized the token-based pricing model used by AI labs Anthropic and OpenAI, arguing that it leads to wasteful spending. In a CNBC interview, Karp stated, 'Something has gone completely wrong,' as enterprises increasingly prioritize return on investment over token-based AI services. Meanwhile, Palantir has published a 9-point manifesto advocating for 'AI sovereignty,' urging companies to retain control over their data and avoid outsourcing to third-party AI providers.

Part 1: Immediate Action & Core Facts

• Palantir CEO Alex Karp criticized the token-based pricing model of OpenAI and Anthropic, calling it wasteful amid rising AI costs.
• Palantir published a manifesto advocating for 'AI sovereignty,' warning against 'tokenmaxxing' and urging companies to retain control over their data.

Part 2: Deeper Dive & Context
The Critique of Token-Based AI Models
Karp argued that enterprises are shifting away from 'tokenmaxxing'—excessive spending on AI tokens—toward more cost-effective solutions, such as open-weight models and Chinese AI alternatives. He claimed that token-based models incentivize disposable scripts over robust software, leading to 'false progress.'

Palantir’s Manifesto on AI Sovereignty
The manifesto emphasizes the importance of data retention and controlling AI weights, arguing that outsourcing AI needs relinquishes institutional control. Key points include:

1. AI sovereignty is essential for long-term institutional success.
2. Data retention is critical for competitive advantage.
3. Tokenmaxxing undermines institutional intelligence by prioritizing short-term usage over long-term value.
4. Controlling AI weights ensures that companies retain their proprietary knowledge.

Industry Shifts and Concerns
As AI costs rise, enterprises are exploring alternatives to token-based models. Open-weight models and Chinese AI advancements are gaining traction, raising concerns about U.S. competitiveness in the AI space. Palantir’s stance reflects broader debates over AI ethics, cost efficiency, and data control.