AI
OpenAI's Jalapeño chip claims 1.5-1.9x Nvidia efficiency on inference
OpenAI's Jalapeño inference ASIC, co-designed with Broadcom and fabricated by TSMC, was presented at Hot Chips 2026 as delivering 1.5 to 1.9 times higher throughput per kilowatt and 1.7 to 3.6 times lower latency than Nvidia's GB200 and GB300 on selected inference workloads. The chip went from architectural concept to running OpenAI's Codex model in nine months, with AI models writing more than half the core in the XLS hardware language. Independent verification has not yet been published, and Nvidia's $1 trillion combined Blackwell-and-Vera-Rubin forecast for 2026-27 is unchanged.