OpenAI unveils Jalapeño AI chip with Broadcom

Key takeaway

• OpenAI has announced Jalapeño, a custom-designed AI chip built with Broadcom that performs at the level of Nvidia's Blackwell and Google's Tensor processing units.

• The move allows OpenAI to reduce its reliance on Nvidia's GPU supply constraints and control its own hardware development; the chip is intended for AI inference workloads and is expected to deploy by the end of 2026.

3 Key Points

FAQ

What is Jalapeño designed to do?

Jalapeño is an ASIC chip designed specifically for AI inference—the process where an AI model processes a user's request to run an agent like Codex or generate a response from ChatGPT. It is not designed for AI training, which involves consuming vast amounts of data.

When will Jalapeño be available?

OpenAI expects to deploy Jalapeño by the end of 2026 as the first step in a multi-generation compute platform.

How does Jalapeño compare to competitors' chips?

According to Broadcom's CEO, Jalapeño matches the performance of Nvidia's Blackwell chips and Google's Tensor processing units. OpenAI says early testing shows it will deliver performance per watt substantially better than current state-of-the-art.