OpenAI and Broadcom announce chip designed for LLM inference at scale

OpenAI and Broadcom unveil "Jalapeño," a custom chip built for LLM inference

OpenAI and Broadcom unveiled "Jalapeño," a custom inference chip built from scratch in nine months, scaling to gigawatt production by late 2026. Vertical stack integration cuts inference costs and latency; Microsoft's 40% commitment signals industry consolidation around proprietary accelerators over commodity GPUs.