Broadcom has become the latest chip maker to team up with OpenAI, with the pair partnering on custom AI accelerators.

The companies signed a deal that’ll see the ChatGPT developer design its own GPUs, which will be jointly created alongside Broadcom. In addition to the chip designing agreement, OpenAI will also tap Broadcom to deploy its networking solutions in its racks.

The jointly developed hardware aims to be in production in the second half of 2026, with a view to completion by the end of 2029.

“Partnering with Broadcom is a critical step in building the infrastructure needed to unlock AI’s potential and deliver real benefits for people and businesses,” OpenAI CEO Sam Altman noted. “Developing our own accelerators adds to the broader ecosystem of partners, all building the capacity required to push the frontier of AI to provide benefits to all humanity.”

OpenAI’s Broadcom deal is a major milestone for the startup, with it having mulled over designing its own hardware since at least 2023. Altman himself is among the OpenAI leaders backing the custom hardware project, having previously expressed dismay over access to hardware.

Having relied on chips from Nvidia for years, OpenAI recently joined forces with AMD in a deal that would see the hardware maker supply 6 gigawatts (GW) worth of semiconductors in exchange for up to 160 million shares, or around 10% of the company.

Now, OpenAI has enlisted Broadcom as it looks to snap up even more hardware, with this deal covering some 10 GW of accelerators.

“Broadcom’s collaboration with OpenAI signifies a pivotal moment in the pursuit of artificial general intelligence,” Broadcom CEO Hock Tan added. “OpenAI has been at the forefront of the AI revolution since the ChatGPT moment, and we are thrilled to co-develop and deploy 10 gigawatts of next-generation accelerators and network systems to pave the way for the future of AI.”

Reports previously suggested OpenAI was working with Broadcom and Taiwan Semiconductor Manufacturing Co. (TSMC) on custom chips, a move that makes some sense considering that Broadcom previously worked on Google’s custom TPUs. The startup has pinched a handful of Google engineers who helped develop its TPUs, including project lead Richard Ho, who now serves as OpenAI's head of hardware.

Prior reporting suggests OpenAI’s custom hardware will be made using TSMC’s 3-nanometer technology, with production slated for 2026. A Reuters report back in February cited sources saying the chips will feature a similar systolic array architecture with high-bandwidth memory to hardware from Nvidia.

OpenAI has teamed up with a hardware developer that’s been pushing the forefront of building out infrastructure for AI. Just last week, Broadcom unveiled Tomahawk 6 Davisson, its third-generation co-packaged optic (CPO) Ethernet switch. For comparison, Nvidia only unveiled plans to launch its own CPO in February this year.