OpenAI Unveils Jalapeno LLM Inferencing Accelerator Built in Collaboration with Broadcom

OpenAI unveiled Jalapeno, its in-house LLM inference accelerator chip developed with Broadcom, achieving tape-out in nine months. The chip is designed to accelerate inferencing for OpenAI's LLM stack powering ChatGPT, Codex, and future agentic AI products, with initial deployment targeted for late 2026.

OpenAI today unveiled Japaneo, its in-house, purpose-built LLM inference accelerator chip, developed in collaboration with Broadcom. The chip has fixed-function and programmable compute hardware to accelerate inferencing of OpenAI LLM stack that powers ChatGPT, Codex, the OpenAI API, and the company's future agentic AI products. It is conceptually similar to Google's TPU, except designed for the OpenAI stack; and while TPU is used for both training and inferencing, Jalapeno appears to be limited to inferencing. The training is probably still done with GPUs. The company said that in partnership with Broadcom, the chip was designed and reached manufacturing tape-out in just nine months, making it the fastest ASIC development cycle among advanced semiconductors. Jalapeno is part of a compute platform that will span multiple generations of these chips. The platform is being designed for initial deployment toward the end of 2026. There are no technical details about the chip itself, except that it's a contemporary multi-chip module with an interposer, a large and centrally-located logic tile, flanked by eight HBM3E stacks.