OpenAI, Broadcom unveil Jalapeño: OpenAI's in-house chip to speed AI inference

OpenAI and Broadcom announced Jalapeño, their first custom AI chip designed for inference to accelerate chatbot queries and compete with Nvidia’s Blackwell and Google’s TPUs. Built with Broadcom’s design input and Celestica assembling servers, the chip is manufactured by TSMC and is intended for deployment by year-end as part of a multi-generation chip plan; OpenAI has already tested it with GPT‑5.3‑Codex‑Spark, highlighting broader industry moves toward in-house AI silicon amid high memory-demand pressures that affect margins.