Cerebras Systems Inc. CSO: fastest tokens win in agentic AI

Cerebras Systems Inc. is making its case that the AI infrastructure race is shifting from raw token volume to the value each token delivers. Its chief strategy officer, Andy Hock, said the change happened over the past two months, as foundation labs and enterprises moved away from "token maxing." In his account, those buyers are now making decisions on models, datasets and infrastructure under a value-maximization paradigm, and on the infrastructure side they are finding the greatest value in the fastest tokens.

The company's pitch rests on how agentic systems actually work. Hock described an agentic query not as one intelligent agent answering a question, but as a family of daughter agents doing different work and coordinating together. That architecture, he argued, is where speed makes the difference. Cerebras points to hardware built around that idea: its Wafer-Scale Engine is more than 50 times bigger than any prior chip, and the company claims 10 to 15 times faster inference than any other platform.

Cerebras reported its first-quarter 2026 results in June, and it raised $5.5 billion in its May 2026 initial public offering. Hock made the comments in an interview with theCUBE Research's John Furrier at the AI Data Pipeline Forum, part of SiliconANGLE's and theCUBE's coverage of the event.