Cerebras

Cerebras Systems designs the Wafer-Scale Engine, a processor the size of an entire silicon wafer and the largest chip ever built, aimed at both AI training and ultra-fast inference. Its systems avoid the multi-chip communication overhead of GPU clusters, and its inference cloud posts some of the fastest tokens-per-second figures in the industry for large open-weight models.

Founded in 2015 by Andrew Feldman and colleagues (who had previously sold SeaMicro to AMD), Cerebras bet that a single enormous chip could outperform networks of smaller ones for AI. Its Wafer-Scale Engine packs hundreds of thousands of cores and enormous on-chip memory bandwidth onto one wafer-sized die.

The CS-series systems and Cerebras's inference cloud target workloads where speed matters most, delivering very high throughput on large models and attracting customers in research, healthcare, and sovereign-AI programs. Deep partnerships in the Middle East, particularly with the UAE's G42, have been central to its growth and compute deployments.

Cerebras is one of the highest-profile AI-hardware challengers to NVIDIA, competing with Groq and others on the argument that specialized architectures win on inference economics. Its wafer-scale approach remains the most physically distinctive bet in AI silicon.