OpenAI taps Foxconn to co-design AI hardware for U.S. factories

OpenAI and manufacturing giant Foxconn are linking up to design and eventually build the next generation of physical hardware for AI data centers on American soil. The two companies announced a collaboration on Friday that will have OpenAI sharing its infrastructure roadmap with Foxconn, which in turn will engineer multiple generations of AI data center racks and key components—cabling, networking, cooling, and power systems—at its U.S. facilities.

The deal is notably light on commitments. There are no purchase orders or financial obligations attached yet. Instead, OpenAI gets early access to evaluate the co-designed systems and holds an option to buy them later. It’s a bet on shaping the supply chain before writing the check, something that makes sense when you consider the sheer scale and speed at which model training demands are evolving.

Foxconn Chairman Young Liu framed the partnership in suitably grand terms, calling OpenAI “a pioneer at the forefront of the AI digital age” and positioning his company—the world’s largest manufacturer of AI data servers—as the natural partner to deliver “trusted, scalable infrastructure.” Sam Altman, meanwhile, is pitching the effort as a down payment on reindustrialization. “The infrastructure behind advanced AI is a generational opportunity to reindustrialize America,” Altman said, adding that the work should ensure “the core technologies of the AI era are built here.”

There’s a clear political and strategic logic at play. By broadening sourcing to include more domestic suppliers and expanding localized testing and assembly, OpenAI and Foxconn aim to build what they call a more resilient American supply chain. Whether this actually accelerates deployment or simply adds another layer of design complexity without committed volume is the open question. For now, it’s a handshake agreement with a lot of ambition and no signed checks.