Cerebras

Cerebras is the AI hardware company that builds wafer-scale WSE chips instead of standard GPUs. The WSE-3 generation packs millions of cores and massive memory bandwidth on a single chip, producing eye-catching inference TPS numbers. Together with Groq it is one of the main actors driving the TPS Wars, serving open-weight models like Llama 3, Llama 4, and Qwen 3 at very high throughput. Against classical GPU-based inference it is a niche but growing alternative.