Key facts
- OpenAI and Broadcom have partnered to develop and deploy custom AI chips.
- The new chip, named Jalapeño, is designed specifically for large language model inference.
- This collaboration aims to reduce OpenAI's reliance on Nvidia for AI hardware.
- Deployment of the custom chip systems is slated to begin in the second half of 2026.
- The partnership involves 10 gigawatts of custom chips, with full deployment by the end of 2029.
OpenAI has unveiled Jalapeño, its first custom artificial intelligence chip, developed in partnership with Broadcom. This move marks a significant step in OpenAI's strategy to control more of the hardware powering its AI models, including ChatGPT and future products. The Jalapeño chip is specifically designed for large language model inference, aiming to provide more computing power with greater efficiency and lower energy consumption compared to current leading AI chips.
OpenAI CEO Sam Altman stated that the Jalapeño chip is integral to the company's long-term, full-stack infrastructure strategy, which seeks to make compute more abundant and accessible. The company claims that by designing more of its stack internally, it can achieve greater efficiency and expand access to advanced AI.
The announcement confirms previous reports that OpenAI was developing its own silicon to reduce its dependence on Nvidia, a move also pursued by other major tech companies like Google and Amazon. Early versions of the Jalapeño chip are already undergoing testing in OpenAI's labs. Broadcom President and CEO Hock Tan highlighted the collaboration as a commitment to scaling the physical infrastructure for the next decade of AI, with deployment of gigawatt-scale data centers planned with Microsoft and other partners starting in 2026.