July 24, 2025 /SemiMedia/ — OpenAI said Tuesday it will deploy more than 2 million AI-optimized chips across new U.S. data centers developed in partnership with Oracle, as part of a large-scale expansion to support future artificial intelligence training and inference workloads.

The effort includes adding 4.5 gigawatts of new capacity, pushing OpenAI’s total infrastructure to over 5GW. Candidate sites include Texas, Michigan, Wisconsin, and Wyoming, though final locations have not been disclosed.

The first racks containing NVIDIA’s GB200 chips have already been delivered to OpenAI’s Abilene, Texas site, with portions of the system already operational. The facility forms part of the broader “Stargate” initiative between OpenAI and Oracle.

“This site is our beta test for deploying advanced compute infrastructure at scale,” said Chris Lehane, OpenAI’s global policy vice president. He noted the Abilene center is already handling select training workloads.

The new data center buildout is expected to generate more than 100,000 construction and operations jobs across the U.S. Although SoftBank is not funding this particular phase, OpenAI indicated the Stargate project remains on track. A scaled-down pilot site in Ohio is targeted for completion by the end of 2025.