Cerebras: Wafer-Scale AI Inference at 3,000 Tokens Per Second

Cerebras built the largest chip ever made — a silicon wafer the size of a plate — and the results are extraordinary. gpt-oss-120B at 3,000 t/s. Llama 4 Scout at 2,600 t/s. Backed by a $20B OpenAI compute deal. Completed the largest tech IPO of 2026 on May 14 ($185/share, 70% first-day surge, $70B market cap). The catch: a catalog of ~5 models and no fine-tuning on the public API.