OpenAI has revealed its first custom AI processor, called Jalapeño, developed in partnership with Broadcom. Multiple outlets report that the chip is an ASIC built specifically to support AI inference—the phase where existing models generate outputs—rather than training. The companies position Jalapeño as an efficiency-focused hardware upgrade for the systems that run models behind OpenAI’s products, including large language model workloads such as those used in ChatGPT and Codex.
Coverage agrees that OpenAI and Broadcom aim to reduce the cost of running AI. Bloomberg reports the partners say the chip can cut costs by about 50%, while other outlets describe similar claims about performance and cost improvements. The Verge and TechCrunch characterize the design as tailored to OpenAI’s inference needs.
Broadcom is also described as the manufacturing partner for the silicon. Bloomberg adds that OpenAI is testing samples of the chip. Times of India reports expected initial deployments later in the year, with scaling through 2027, while other articles emphasize the broader strategy of building more of the “full stack” around OpenAI’s models and services.
