"I'm not throwing shade at them, but something has gone completely wrong," he told CNBC's Squawk Box. "The basic view among enterprises in this country is I'm going to chillax and waste my time with tokens."
As AI costs rise and each model iteration becomes pricier, enterprises are shifting their focus from consumption-based token models to return on investment. This situation is leading certain companies to choose open weight models, which can handle comparable jobs for much less cost. Meanwhile, Chinese models are rapidly advancing, sparking worries that this competitor might soon match the leading U.S. AI labs.
Karp warned CNBC that the pace of China's AI model development should not be underestimated. Given these trends, numerous companies are moving away from broad AI models and instead developing and training their own specialized, cost-effective systems.
According to Karp, open weight models could address the frustrations that corporate leaders have with AI labs. "What aligns me with Nvidia, and I think is what the technical customers want, which is control over their compute, their models, their data stack and their alpha," Karp said. "They want to know they own the means of production. It's not being transferred to someone else."
Get your free investing masterclass bonus when you join Market Briefs, our free daily newsletter
The same investors pouring money into OpenAI and Anthropic are backing a model Karp calls "completely wrong."
Enterprises are increasingly evaluating the total cost of owning AI systems. While frontier models from labs like OpenAI and Anthropic offer cutting-edge capabilities, the recurring per-token charges can accumulate rapidly, especially for high-volume usage. This has led many corporate leaders to explore open-weight alternatives that allow them to run AI on their own infrastructure, thereby avoiding ongoing fees and gaining greater control over data security and customization.
The debate over AI pricing comes as companies pour billions into the sector. This shift reflects a deeper tension: while venture capital bets on massive general-purpose models, many corporations prefer to own their own AI infrastructure rather than pay per token indefinitely.
The growing preference for open-weight and custom models signals a potential slowdown in revenue growth for proprietary AI labs, as enterprises seek to reduce dependency on per-token pricing. For investors, this could mean that the biggest winners in AI may not be the frontier labs themselves, but the companies that help businesses build and control their own AI systems.
Subscribe to Market Briefs, our free daily newsletter, and claim your bonus investing masterclass