OpenAI and Broadcom unveil Jalapeño, a custom chip designed for LLM inference at scale
Tags AI · Infrastructure
OpenAI and Broadcom announced Jalapeño, a custom silicon processor purpose-built for large language model inference in data centers, developed from design to production in nine months. OpenAI leveraged its own models to accelerate chip design and optimization. This marks OpenAI's first custom chip and the beginning of a multi-generation roadmap. CEO Sam Altman appeared alongside Broadcom CEO Hock Tan for the announcement, signaling vertical integration by a major AI lab into silicon — a trend previously led by Google's TPUs and Amazon's Trainium.
Technical significance
This accelerates the trend of AI labs designing custom silicon optimized for their own workload patterns, reducing dependency on general-purpose GPU fleets. The nine-second-month design-to-production timeline, reportedly aided by OpenAI models in the chip design loop, suggests AI-assisted EDA tools are maturing rapidly. Second-order effect: Broadcom gains a marquee AI customer, while NVIDIA faces long-term competitive pressure from vertically integrated inference silicon.