OpenAI and Broadcom have developed Jalapeño, a custom AI chip designed specifically for large language model inference workloads.

The custom silicon targets improved performance and efficiency for AI systems running inference at scale. The chip represents OpenAI's first major hardware partnership as the company seeks to reduce dependence on Nvidia's GPU dominance.

Broadcom brings decades of semiconductor design expertise to the collaboration. The company has previously worked with Google on custom AI chips and maintains relationships across major cloud providers.

Jalapeño focuses on inference rather than training workloads. This approach mirrors strategies from other AI labs seeking specialized hardware for deployment rather than model development.

The partnership signals OpenAI's growing infrastructure ambitions beyond software. The company has been exploring various hardware initiatives as it scales ChatGPT and API services globally.

Neither company disclosed technical specifications, pricing, or availability timelines for the new chip. Industry observers expect more details at upcoming conferences.

The move follows similar custom chip initiatives from Google, Amazon, and Meta. Each major AI provider has sought to optimize hardware for their specific workloads and reduce reliance on third-party suppliers.

OpenAI's hardware push comes as the company faces increasing compute demands from enterprise customers and API usage growth. Custom inference chips could help manage costs while improving response times for end users.