OpenAI Unveils First AI Inference Chip Jalapeño With Broadcom and TSMC

OpenAI has long relied on outside suppliers for AI computing resources. As the use of large language models expands, inference costs and chip availability have become critical. Its first in-house accelerator, Jalapeño, was co-designed with Broadcom and manufactured by TSMC for model inference workloads. The chip could help OpenAI improve infrastructure efficiency and reduce its reliance on general-purpose GPUs.

OpenAI recently formally unveiled Jalapeño about nine months after development began and expects to deploy it in its own computing infrastructure by the end of 2026. The chip is optimized for large language model inference, with Broadcom and TSMC among the partners involved. At the time of the announcement, OpenAI had not disclosed the process node, deployment volume, or investment and procurement amounts.