⚡ OpenAI & Broadcom Reveal Jalapeño — Their Custom AI Chip Built for LLM Inference
OpenAI and Broadcom unveiled Jalapeño on June 24, OpenAI's first-ever custom inference chip designed from scratch for running large language models. Remarkably, it went from initial design to tape-out in just nine months — with OpenAI's own AI models accelerating parts of the design process. Early tests show substantially better performance per watt than current GPUs, with deployment targeted by end of 2026. Could custom silicon finally challenge NVIDIA's dominance in AI compute?