Cerebras builds wafer-scale AI platform amid growth test

2 min read

Cerebras has centered its growth case on a wafer-scale approach that keeps compute and memory on a single wafer, claiming lower latency for large-model workloads. Its WSE-3 includes about 4 trillion transistors, 900,000 AI cores, 44 gigabytes of on-chip memory, 21 petabytes per second of memory bandwidth and 214 petabits per second of fabric bandwidth. The company sells CS-3 supercomputers plus networking, cluster software and cloud services, and reported core revenue up 92% to $191.3 million and cloud/services up 167% to $79.8 million in Q1 2026. That momentum faces tests from delivery obligations, margin pressure and competition from NVIDIA and AMD, and shares recently underperformed peers.

WSE-3 and cloud growth drive recurring revenue but scaling is uncertain