OpenAI and Broadcom unveil LLM-optimized inference chip

OpenAI and Broadcom have jointly developed Jalapeño, a custom AI inference chip optimized for large language models (LLMs). Designed and produced in nine months, the chip promises significantly improved performance per watt and will support OpenAI's models including GPT-5.3-Codex-Spark. The chip is set for deployment at gigawatt scale with data center partners, enhancing OpenAI's full-stack AI platform.