OpenAI, Oracle, and SoftBank announced five new U.S. data center sites for Stargate, their shared AI infrastructure project. The expansion would push Stargate’s planned capacity toward nearly 7 gigawatts and bring total investment to about $400 billion over the next three years.
The build aims to support ChatGPT’s hundreds of millions of weekly users and to train upcoming AI models, but critics question whether the funding structure can be sustained amid rising costs and uncertain demand.
Three of the new sites will be developed through the OpenAI-Oracle partnership: Shackelford County in Texas; Doña Ana County in New Mexico; and an unnamed Midwest location. Alongside a 600-megawatt expansion near Abilene, Texas, these sites would collectively exceed 5.5 gigawatts of capacity, enabling up to 5.5 gigawatts of on-site electrical draw at full tilt.
SoftBank will contribute two sites: Lordstown, Ohio—already under construction and slated to be online next year—and Milam County, Texas, developed with SB Energy. Together these two projects could scale to about 1.5 gigawatts within 18 months.
The previously announced flagship Stargate campus in Abilene remains central; Oracle has started delivering Nvidia hardware there, and OpenAI is already training and running inference on the site. The broader plan includes a stated goal of reaching $500 billion in investment and 10 gigawatts of capacity by the end of 2025, a target that is increasingly viewed with skepticism by some analysts.
Industry observers are debating the ‘circular investment’ model, where infrastructure vendors finance AI developers and recirculate profits through recurring compute contracts. If demand slows, the oversized compute footprint could become a painful write-down.
