
OpenAI’s first custom AI chip, built with Broadcom, is aimed at faster, lower-cost inference as the company seeks more control over infrastructure and less reliance on third-party hardware including Nvidia.
OpenAI and Broadcom have unveiled Jalapeño, OpenAI’s first custom AI chip, designed to speed up and reduce the cost of inference, the process of serving AI models to users. The Broadcom-made ASIC, which the companies call an "Intelligence Processor," was designed in nine months, with a physical sample due on Wednesday and initial deployment targeted by the end of 2026. The launch extends plans disclosed in October to develop OpenAI-designed chip racks and reflects OpenAI’s effort to build more of its AI infrastructure in-house while diversifying beyond suppliers such as Nvidia, Amazon Web Services, AMD and Cerebras. Reuters has also reported that OpenAI explored alternatives after dissatisfaction with some Nvidia hardware for certain inference tasks, though Sam Altman said OpenAI expects to remain a major Nvidia customer. Broadcom shares rose about 2% after the announcement.