OpenAI and Broadcom Debut Jalapeño AI Chip to Power Next-Gen LLMs

OpenAI and Broadcom have unveiled Jalapeño, a custom “Intelligence Processor” designed to run large language models more efficiently. Built in just nine months with Broadcom and Celestica, Jalapeño is already running workloads including GPT‑5.3‑Codex‑Spark. Early tests show better performance per watt than current accelerators, with architecture optimized to reduce data movement and balance compute, memory, and networking. Positioned for gigawatt‑scale data centers from 2026, Jalapeño aims to deliver faster ChatGPT responses, cheaper APIs, and more reliable access.