Palantir CEO Alex Karp publicly criticized the recent trend of "tokenmaxing" in Silicon Valley, arguing that simply increasing the usage of AI does not equate to creating real business value. In an interview during Palantir AIP Con 10, he stated that the market has shifted from discussing "whether AI truly exists" to "AI is indeed effective, but in many scenarios it doesn't function as expected."

The controversy points to high-consumption use

A token is a basic unit of measurement used by large language models to process text, and AI service providers typically charge based on token consumption. In the past few weeks, some practitioners in Silicon Valley have begun to rethink the "tokenmaxxing" culture, which involves expanding the scale of AI usage almost without restrictions in order to keep up with the development speed of AI agents.

Karp argues that more tokens often mean more low-quality outputs, not higher-value results. Palantir's CTO, Shyam Sankar, expressed a similar view during last month's earnings call, stating that the company emphasizes a "no slop zone" and opposes treating cheap model calls as value itself.

Palantir emphasizes system rather than model heap size.

Sankar stated at the time that cheaper AI does not automatically lead to higher returns; businesses still need systems like Palantir AIP to connect model capabilities with real-world business environments and avoid financial losses from erroneous outputs.

In a recent interview, Karp further stated that the real challenge is not generating generic content using a model, but rather embedding AI into continuously running business processes. For example, a large model can handle writing a report on China's GDP growth quite well; however, when it comes to complex tasks such as oil and gas extraction, supply chain adjustments, military manufacturing, or automobile production, AI itself cannot replace specific processes.

Complex business processes still need to be continuously executed.

He believes that these types of issues often involve costs, compliance, ethics, and implementation details simultaneously, requiring precise and continuous operational processes. Larger models can enhance these processes, but cannot directly replace them.

Karp also mentioned that the industry is gradually realizing that AI's capabilities have been validated, but if companies want to truly turn it into commercial results, the key is not to infinitely increase the number of model calls, but to whether they are clear about what business problems they need to solve and how to integrate the model into an executable system.