OpenAI and Broadcom unveil custom Jalapeño AI chip

OpenAI has revealed a custom inference chip called Jalapeño, developed in partnership with Broadcom, which previously helped Google build its TPU accelerator line. Unlike Nvidia's Rubin GPUs, Jalapeño is designed solely for inference workloads, with early testing showing significantly higher performance per watt than current state-of-the-art chips. The chip's architecture focuses on reducing data movement, a key inference bottleneck, and will be paired with Broadcom networking technology in custom server racks built with Toronto-based Celestia Inc. OpenAI plans to bring its first Jalapeño servers online by year's end, describing the chip as the first step in a multi-generation compute platform ahead of its anticipated public offering.