Cerebras CSO Says Fastest Tokens Will Win AI’s Next Phase as Agentic Workflows Demand Speed

Hock’s argument rests on how agentic systems work. A single question from a user into an agentic system does not just spawn one intelligent agent that goes off to do work or answer the question, he said. It spawns a family of daughter agents that do different work and coordinate together. “That’s where speed really makes the difference,” Hock said.

Cerebras has built its business on making processors much faster. Hock said the company did not build its chips big because big looks cool. “We built it big because AI wants big chips,” he said. AI needs a lot of compute, but it needs that compute close together and close to memory, according to Hock. The company’s Wafer-Scale Engine is more than 50 times bigger than any prior chip, and Cerebras claims 10 to 15 times faster inference than any other platform, Hock said.

The company raised $5.5 billion in its May 2026 initial public offering. It reported first-quarter 2026 results in June. In the two months since, foundation labs and enterprises have moved away from “token maxing” and toward getting the most value out of each token, Hock said. They are making decisions based on that value-maximization paradigm for models, datasets and infrastructure, he said. On the infrastructure side, Cerebras’ end users are finding the greatest value from the fastest tokens, according to Hock.

The interview was part of SiliconANGLE’s and theCUBE’s coverage of the AI Data Pipeline Forum. The segment was sponsored by ScaleFlux, and neither ScaleFlux nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.

Editor's Summary

Hock said speed is now a requirement for agentic AI and that Cerebras is positioning its wafer-scale chips as the fastest-token platform. The company, which raised $5.5 billion in a May 2026 IPO, says labs and enterprises are increasingly measuring value per token rather than maximizing token counts.