OpenAI details its first custom AI chip
OpenAI custom AI chip Jalapeño boosts inference efficiency for ChatGPT and agents See how this Broadcom-built ASIC could cut Nvidia dependence and scale compute fast
OpenAI has revealed new details about Jalapeño, its first custom AI chip developed with Broadcom. The company says the chip moved from design to factoryready status in nine months, with OpenAI models helping optimize the process.
Jalapeño is an ASIC built for inference, the phase where a trained model serves users, rather than for training new models. OpenAI says testing showed performance per watt that is substantially better than current stateoftheart chips.
The chip is part of OpenAI’s broader plan to expand its custom compute stack and reduce dependence on Nvidia for serving products like ChatGPT, Codex, and future agents. OpenAI says it aims to power 10 gigawatts of compute with custom chips by 2029, while still using Nvidia for model training.