Palantir Technologies CEO Alex Karp criticized the token-based pricing models used by OpenAI and Anthropic during a CNBC interview on June 10 [1].
The dispute highlights a growing tension between frontier AI laboratories and the corporate clients who implement their technology. As costs for these services rise, enterprises are increasingly questioning whether the current billing structures align with actual business value, or simply serve to inflate the revenue of AI startups.
Karp said that the current state of AI pricing is unsustainable for many companies. He accused the labs of "tokenmaxxing," a term suggesting that these companies prioritize the maximization of token usage over the operational needs of their customers [2]. According to Karp, this approach has led to skyrocketing costs that force corporate customers to prioritize efficiency over the expansive capabilities offered by frontier labs [3].
"Something has gone completely wrong," Karp said [4].
He indicated that the friction caused by these pricing models is driving a shift in the market. Karp said that enterprises are becoming unhappy with frontier labs and believe those labs only care about maximizing token use [5]. This dissatisfaction is reportedly pushing businesses to move away from proprietary models and toward open-weight alternatives, which may offer more predictable cost structures and greater control over deployment [2].
"Businesses think that the startups do not understand their operations and only care about maximizing token use," Karp said [6].
While Karp focused on the operational frustrations of his clients, Palantir has seen significant market growth. Shares of the company have gained nearly 2,700% since the start of 2023 [7]. This growth occurs as Palantir positions itself as a bridge between complex AI models and the practical requirements of large-scale enterprise operations.
Karp concluded that the disconnect between how AI labs price their products and how businesses actually function is creating a barrier to widespread adoption. He said that enterprises now believe the labs are more interested in their own metrics than in helping customers achieve specific operational goals [5].
“"Something has gone completely wrong."”
This conflict signals a transition from the 'experimental' phase of generative AI to the 'operational' phase. While token-based pricing allowed AI labs to scale rapidly and cover massive compute costs, it creates unpredictable overhead for corporations. If frontier labs fail to introduce more stable, outcome-based, or flat-fee pricing, they risk losing enterprise market share to open-weight models that companies can host internally to control costs.
