Palantir Technologies CEO Alex Karp said Wednesday that the current method of selling artificial intelligence is fundamentally broken [1].

Karp's critique targets the economic structures of the AI industry, suggesting that the current pricing models create perverse incentives for developers and enterprises. He argues that the focus on token costs forces companies to prioritize narrow efficiency over the broader utility of the technology.

During an interview on CNBC’s “Squawk Box” program, Karp said, “Something has gone completely wrong with how AI is sold” [1]. He specifically criticized the prevalence of soaring token-cost models, which charge users based on the volume of data processed by the model [1].

Karp urged a departure from what he termed “tokenmaxxing” — the practice of optimizing AI interactions solely to reduce token expenditure [1]. He suggests that this approach hinders the development of truly effective AI systems by restricting how they are deployed and scaled.

To address these issues, Palantir released a nine-point manifesto [4]. The document decries the current reliance on token-based pricing and instead champions the concept of AI sovereignty [4]. This approach emphasizes a company's ability to control its own AI infrastructure and data, without being beholden to the pricing whims of external model providers.

When an interviewer noted that Karp sounded angry during the discussion, the CEO said he maintained his stance on the necessity of a systemic shift [2]. He argued that the industry must move toward a model that supports the strategic goals of the user rather than the profit margins of the token provider [1].

““Something has gone completely wrong with how AI is sold.””

The tension between token-based billing and enterprise utility reflects a broader struggle in the AI industry to find a sustainable business model. By advocating for AI sovereignty, Palantir is positioning itself as an alternative to the 'pay-as-you-go' ecosystem dominated by companies like OpenAI and Anthropic, signaling a potential shift toward localized, fixed-cost AI infrastructure for large corporations.