OpenAI, Broadcom unveil Jalapeño chip for AI models

Key takeaway

• OpenAI and Broadcom have unveiled Jalapeño, a custom chip designed specifically for language model inference, marking OpenAI's first venture into custom hardware design.

• The chip was designed from scratch in nine months and will be deployed at gigawatt scale by late 2026, reflecting OpenAI's strategy to control its full technology stack from chip to product for faster, more reliable, and lower-cost operations.

3 Key Points

FAQ

When will Jalapeño chips be available?

The first deployment is planned for late 2026 at gigawatt scale, together with Microsoft and other partners.

How does Jalapeño differ from existing chips?

Jalapeño was designed from scratch specifically for modern language model inference, rather than being a modified general-purpose chip. Early tests showed performance per watt that OpenAI claims is "substantially better" than current state-of-the-art hardware, though these results have not been independently verified.

What are the roles of OpenAI, Broadcom, and Celestica in this project?

OpenAI handles the chip design, Broadcom contributes silicon manufacturing and networking technology including its Tomahawk networking chips, and Celestica handles boards, racks, and system integration.