On Wednesday, Palantir Technologies Inc. (PLTR) CEO Alex Karp launched a sharp critique of the token-based model employed by AI firms like Anthropic and OpenAI, citing spiraling costs.
"I'm not disparaging them, but that model is absolutely broken," he stated. "The general perception in corporate America is: I might as well sit back and just waste my time burning tokens."
As AI expenses continue their steep climb, with newer model generations proving more costly than their predecessors, businesses are shifting away from a "tokenmaxxing" mindset and increasingly focusing on tangible return on investment (ROI).
This shift is driving some companies toward open-weight models, which can perform comparable tasks at a significantly lower cost. Concurrently, the rapid advancement of models from other regions is raising concerns that AI competitors may soon catch up to leading U.S. labs.
Amid this landscape, numerous enterprises are also moving from utilizing general-purpose large language models to constructing and training their own, more efficient proprietary tools.
Earlier this week, Palantir announced an expansion of its partnership with NVIDIA (NVDA), leveraging the chip giant's AI tools to build custom models for U.S. government agencies.
Karp suggested that open-weight models present a potential solution for CEOs who are frustrated with traditional AI labs.
"The attitude of the most technically sophisticated players right now is: 'I want something I own. This is my business,'" he said.
Comments