OpenAI unveils Jalapeno inference chip with Broadcom

OpenAI unveiled Jalapeno, its first custom inference processor, in partnership with Broadcom, according to a Mobile World Live report published on 24 June 2026. OpenAI said Jalapeno is the first chip in a multigenerational computing platform that it is developing jointly with Broadcom for AI inference workloads.

OpenAI designed Jalapeno from scratch based on its model roadmap, cores, server systems, and product requirements, while Broadcom and Celestica handled chip manufacturing, board and rack integration, networking, and production scaling. OpenAI said the architecture is designed to reduce data movement and balance processing, memory, and network resources, and that Broadcom’s Tomahawk networking chip and integration technology form the base for large-scale production.

OpenAI said it is still finalizing performance measurements, but early tests indicate Jalapeno delivers higher performance per watt than current-generation AI accelerators. Bloomberg reported the accelerator is achieving cost savings of about 50% compared with typical AI graphics processing units. OpenAI also said the chip moved from initial design to fabrication in nine months, and that initial deployment is planned for late 2026.