OpenAI unveils 'Jalapeño' AI chip with Broadcom to cut GPU dependency
First in-house inference chip promises better performance and lower energy costs.
OpenAI has announced 'Jalapeño,' its first in-house AI chip, developed in partnership with Broadcom. Unlike training-focused GPUs, Jalapeño is designed specifically for inference — the process of running trained models to make predictions — where speed and power efficiency are critical. By optimizing for this workload, OpenAI expects to significantly lower operational costs and latency in its data centers, which currently depend heavily on expensive NVIDIA processors.
This move marks a strategic push toward vertical integration for OpenAI. By bringing chip design in-house, the company reduces its dependence on third-party suppliers and gains tighter control over hardware-software co-optimization. Jalapeño is expected to power key inference-driven products like ChatGPT and API services, potentially improving response times and energy use. The partnership with Broadcom leverages their expertise in custom silicon, accelerating development and production.
- Developed in partnership with Broadcom, OpenAI's first proprietary chip is named 'Jalapeño'
- The chip is optimized for inference tasks, not training, targeting better speed and energy efficiency
- Aims to reduce OpenAI's reliance on NVIDIA GPUs and give greater control over its AI infrastructure
Why It Matters
OpenAI gains hardware independence, potentially lowering costs and improving inference efficiency for its AI services.