OpenAI and Broadcom Launch Jalapeño AI Chip with Improved Performance per Watt
OpenAI and Broadcom introduced Jalapeño, their first custom AI inference processor designed to power large-scale language model workloads. Early tests indicate the first-generation chip delivers substantially better performance per watt compared with current state-of-the-art accelerators.
1. Chip Development Partnership
OpenAI partnered with Broadcom to design and manufacture Jalapeño, its first custom AI inference accelerator, eight months after their initial agreement on custom chip development.
2. Performance Testing Results
Early sample testing shows the first-generation Jalapeño delivers substantially better performance per watt than current state-of-the-art AI accelerators, although full performance metrics are still under evaluation.
3. Strategic Infrastructure Implications
The custom processor underscores a shift toward vertical integration of AI hardware and software, aiming to reduce reliance on third-party chips while optimizing cost and efficiency for large language model inference.
4. Production Timeline and Outlook
OpenAI plans to ramp up production for deployment in its data centers over the next year, targeting large-scale workloads and potential future chip iterations to further enhance performance.
