OpenAI Joins Broadcom to Build First Custom AI Chips, Targets 10GW by 2029

OpenAI has struck a deal with Broadcom to design and build its first in-house AI processors.  Under the agreement, OpenAI will lead chip design while Broadcom handles development and production, with deployment beginning in late 2026. 

This project aims to scale up to 10 gigawatts of custom AI hardware—power equivalent to the needs of over 8 million U.S. homes.  OpenAI plans to complete full deployment by 2029. 

The move signals a push to reduce reliance on Nvidia’s chips. Still, industry analysts say it likely won’t dethrone Nvidia’s lead in AI accelerators.  The Broadcom deal complements OpenAI’s recent 6 GW chip deal with AMD and Nvidia’s $100 billion investment pledge. 

Broadcom’s stock jumped over 10% after the announcement, making the chip firm one of the prominent beneficiaries of the AI infrastructure race.  Under the deal, Broadcom’s own networking gear will support the new systems, offering an alternative to Nvidia’s InfiniBand networking solution. 

OpenAI CEO Sam Altman called the collaboration “a critical step” toward building the infrastructure needed for AI’s next phase.  While the financial terms were not disclosed, the scale of the build suggests multi-billion dollar investments ahead.