Cerebras

Cerebras builds AI chips the size of a dinner plate — the largest chips made for AI — and uses them to run inference at speeds that leave typical GPU-based systems well behind, fast enough to make real-time voice AI and low-latency trading applications genuinely viable. Its cloud service runs open models like Llama at those speeds through a developer-accessible API, and the company filed for an IPO in 2024, a milestone that put it firmly on the radar as a serious Nvidia alternative rather than just a research curiosity.