OpenAI Unveils Jalapeño AI Inference Chip With Broadcom and TSMC

OpenAI has unveiled Jalapeño, its first custom AI accelerator, co-designed with Broadcom and manufactured by Taiwan Semiconductor Manufacturing Co. The chip is optimized for inference, the computing-intensive process of running trained large language models to generate responses. The project could reduce OpenAI’s reliance on Nvidia hardware while giving the ChatGPT maker greater control over the cost, performance and energy efficiency of its infrastructure.

Initial benchmarks released by OpenAI showed Jalapeño delivering higher inference throughput and efficiency than Nvidia’s GB200 and GB300 systems in some tests. The results have not yet been independently validated, and OpenAI has not disclosed production volumes or total development costs. The company plans to deploy the chip by the end of 2026, marking its first move from buying general-purpose AI accelerators to operating silicon designed around its own workloads.