Cerebras CEO Andrew Feldman to Address AI Scaling Limits at Disrupt 2026

Cerebras has spent a decade betting against conventional chip architecture, favoring a wafer-scale design that integrates processors directly onto silicon. This approach is currently being put to the test following the company’s $5.5 billion IPO in May and a significant multiyear agreement with OpenAI to deploy 750 megawatts of system capacity through 2028. With the August launch of the CS-4 system, the company is attempting to prove that specialized hardware can outpace the limitations of traditional, individual chip manufacturing.

Beyond the processor, Feldman’s focus rests on the physical reality of supporting AI: data centers, cooling, and massive electricity consumption. Cerebras reports over 600 megawatts of data center capacity either live or contracted for 2027, alongside a tenfold increase in manufacturing throughput during 2026. As the company expands into Europe, the core question remains whether these industrial efforts can scale fast enough to satisfy the relentless appetite of next-generation models. Feldman will discuss these challenges during the October 13-15 event at Moscone West in San Francisco.