Summary
Bloomberg Technology sits down with Cerebras CEO Andrew Feldman following the company’s first post-IPO quarterly earnings report. Despite record revenue of $191 million — up 92% year over year, with cloud revenue growing 167% — shares fell 17.5% on the day as investors reacted to near-term margin contraction driven by the company renting back hardware from customers to keep pace with demand.
Feldman explains Cerebras’s distinctive position in the AI hardware market: its wafer-scale architecture avoids reliance on HBM memory (made by Micron, SK Hynix, and Samsung, all currently supply-constrained), CoWoS packaging from TSMC, and the three-nanometer node — three of the most acute bottlenecks facing GPU-based competitors. Operating at the five-nanometer node, Cerebras claims the fastest inference throughput in the industry by an order of magnitude. As a concrete example, Feldman notes that Cerebras signed its OpenAI contract on December 24 and had the customer in full production by February 1.
The interview also covers the broader data center capacity crunch, which Feldman describes as the single largest constraint for the entire AI industry — hyperscalers, neo-clouds, and specialized inference providers alike. Cerebras has responded with a 120-megawatt data center partnership with Bell Canada slated for 2027, alongside active pursuits in the US, Europe, and the Middle East. The discussion closes with Feldman’s reaction to OpenAI’s new Jalapeno intelligence processor and the broader trend of frontier labs developing custom ASICs.
📺 Source: Bloomberg Technology · Published June 24, 2026
🏷️ Format: Interview
