OpenAI partners with Broadcom on custom AI hardware for 10GW of new compute capacity

OpenAI has entered a multiyear partnership with Broadcom to develop custom chips and networking hardware that will power 10 gigawatts of new AI data center capacity, according to a statement released Monday by both companies, outlining a plan that begins hardware deployment in the second half of 2026 and completes by 2029.

The agreement means OpenAI will design its own processors while Broadcom handles development and manufacturing. The chips will embed OpenAI’s own model-training experience “directly into the hardware,” aiming to improve performance and efficiency across its infrastructure.

The new systems will be built into server racks and rolled out gradually through OpenAI’s own facilities and those run by its cloud-computing partners.

“By designing its own chips and systems, OpenAI can embed what it’s learned from developing frontier models and products directly into the hardware, unlocking new levels of capability and intelligence,” Broadcom said in the press release.

Right after OpenAI and Broadcom announced the partnership, the latter’s market value instantly surged, as its stock rallied by more than 12%, per data from Google Finance.

The company, which produces components used in iPhones, networking gear, and data infrastructure, has become one of the biggest winners from the ongoing AI spending surge. Broadcom (AVGO) stock is now up 40% in 2025, beating the 29% gain of the Philadelphia Semiconductor Index.

OpenAI says it aims for flexibility with Broadcom’s networking tech

CEO Hock Tan, who first hinted at the partnership during an earnings call last month, confirmed the collaboration through Monday’s announcement. He did not specify the contract’s value but said the deal would help accelerate Broadcom’s work on custom AI hardware.

Analysts had speculated that OpenAI was the mysterious $10 billion customer Tan mentioned during the September earnings call, but that turned out to be incorrect.

Charlie Kawwas, president of Broadcom’s semiconductor division, appeared on CNBC’s “Squawk on the Street” alongside OpenAI President Greg Brockman to discuss the details. Kawwas clarified that OpenAI wasn’t the company behind the $10 billion order, joking that he’d “love to take a $10 billion [purchase order] from my good friend Greg,” but hadn’t received it yet.

While Broadcom will provide the chips and networking systems, it won’t be the one building or operating data centers. Instead, it will install custom server racks built with OpenAI’s designs inside facilities owned either by OpenAI or by its cloud partners.

The full deployment of these 10 gigawatts of compute power, roughly the equivalent of several large-scale hyperscale campuses, is expected to finish before 2030. Both companies see this as part of a long-term infrastructure strategy.

The order increased Broadcom’s forecast for AI revenue next year, when shipments will begin, Tan said during the call.

OpenAI and Broadcom have been working together for the last 18 months, and they will begin deploying racks of custom-designed chips starting late next year, the companies said Monday.

Claim your free seat in an exclusive crypto trading community - limited to 1,000 members.