OpenAI and Broadcom team up on Jalapeño AI chip, aiming for year-end launch

OpenAI and Broadcom have jointly developed an AI inference ASIC chip called Jalapeño, built on a 3nm process and manufactured by TSMC. The chip is designed to optimize large language model inference and is expected to be deployed by the end of 2024. It features 8 HBM stacks and a dual-die main SoC. The two companies are already planning a second-generation AI ASIC project, which may use TSMC's A16 node with backside power delivery for better density and performance. This move aims to reduce OpenAI's reliance on Nvidia GPUs and improve AI inference efficiency.