OpenAI unveils its first custom inference chip with Broadcom

TechCrunch reported that OpenAI has unveiled Jalapeno, its first custom inference processor, built in collaboration with Broadcom. The chip is still being tested, but OpenAI says early results show better performance per watt than current alternatives. The processor is aimed at inference workloads such as real-time coding models and other user-facing AI tasks, while more compute-heavy training work is still likely to rely on Nvidia hardware. The move pushes OpenAI deeper into the full-stack infrastructure race: models, products, data centers, and now the silicon underneath them.

Key details: Published June 24, 2026 at 07:54 PDT, Jalapeno is OpenAI's first custom inference processor, The chip was developed with Broadcom, OpenAI says early testing shows better performance per watt than current alternatives.

Why it matters: Inference cost is one of the biggest constraints on AI products; OpenAI designing its own chip changes the economics and supplier leverage behind ChatGPT, Codex, and future agents.