OpenAI launches its first AI inference chip, Jalapeño, with Broadcom

OpenAI announced the release of Jalapeño, its first custom AI inference ASIC, developed in partnership with Broadcom. The chip, designed to run OpenAI's core large‑language‑model workloads, aims to lessen the company's dependence on Nvidia GPUs and offers energy‑efficiency and performance comparable to leading AI processors.

The development cycle was completed in nine months, using AI‑driven design tools and expertise from hardware lead Richard Ho, who previously helped build Google's TPU. Ho said, "Based on early tests, the Jalapeño chip can approach the theoretical performance limits of the hardware when running our most critical workloads." The processor is currently undergoing internal testing and is expected to see broader deployment by the end of the year.