Palantir CEO Alex Karp raised concerns about large enterprises potentially losing intellectual property and seeing little return on investment when using frontier AI models from companies like OpenAI and Anthropic. Speaking amid news of a Palantir-Nvidia partnership focused on sovereign AI infrastructure, Karp suggested that current token-based pricing models offer little value and risk exposing sensitive business data.

While acknowledging that some companies are indeed worried about AI ROI and token costs, the article argues Karp's claims are largely overstated. Many businesses are actively seeking value from AI, particularly in software development and customer service, and those struggling often haven't re-engineered workflows. Karp’s suggestion for AI vendors to charge a percentage of value derived, rather than per token, mirrors Palantir’s own pricing strategy.

The analysis disputes Karp's assertion that AI labs are broadly "stealing alpha." Leading AI vendors typically have policies preventing direct access to customer data for model training unless explicitly opted-in. However, the article concedes a narrow exception: companies with "design partner" relationships, like Anthropic's past collaboration with Figma, may face heightened risks if AI labs leverage early access to inform competing product development, as suggested by a dispute where Figma withdrew from a joint product launch.