OpenAI unveils first custom chip with Broadcom

OpenAI is moving deeper into the AI infrastructure stack. Its first custom chip shows the company is not only building models, but also trying to control more of the hardware economics behind them.

What happened

OpenAI unveiled Jalapeño, its first custom-built inference processor, designed and manufactured with Broadcom.

The chip is still being tested, but OpenAI says early results show better performance-per-watt than current alternatives. It is designed for inference rather than model pre-training.

Why it matters

This is a major AI infrastructure signal.

Inference is where models meet real customer usage, and cost efficiency matters at scale. A custom chip could help OpenAI reduce dependency on external suppliers and optimise hardware around its own workloads.

The bigger picture

Frontier AI companies are increasingly becoming full-stack infrastructure companies. The race is no longer only about model quality; it is also about chips, memory, networking, energy and deployment economics.