Cerebras

Cerebras builds the Wafer-Scale Engine, a processor cut from a single wafer rather than diced into chips, which it says is 58 times larger than a GPU, and sells it both as the CS-3 system and as the Cerebras Inference Cloud, an OpenAI-compatible API serving open models such as Llama, Qwen and GLM. The pitch is inference speed: it claims more than 2,000 tokens a second and complex reasoning returned in under a second. Customers it names include OpenAI, AWS, CrowdStrike, AlphaSense, GSK, Mayo Clinic and Notion. Based in Sunnyvale, California, and led by co-founder and chief executive Andrew Feldman.