OpenAI teams up with Oracle and SoftBank to build 5 new Stargate data centers (www.wired.com)

OpenAI announced five new U.S. data-center projects under its Stargate initiative, developed with Oracle (three sites: Shackelford County, TX; Doña Ana County, NM; an undisclosed Midwest site) and SoftBank’s SB Energy (Lordstown, OH; Milam County, TX). The additions push Stargate’s planned capacity to nearly 7 gigawatts of IT power—roughly equivalent to seven large nuclear reactors—and include a planned 600 MW expansion of the flagship Abilene site. Oracle’s Abilene build is already online for training and inference, will eventually support ~1.4 GW across eight halls (each ~100 MW), and is expected to house more than 400,000 GPUs; OpenAI says the five new sites plus Abilene expansion will generate >25,000 onsite construction jobs. OpenAI and partners reviewed ~300 proposals from ~30 states and surveyed ~700 potential sites. This is significant because it underlines a bet that raw compute scale and colocated GPU capacity are central to advancing large-model training and inference, and it frames infrastructure as a strategic tool in the U.S.–China AI competition. The move dovetails with a separate Nvidia pact that targets 10 GW and 4–5 million GPUs (Vera Rubin systems operational H2 2026), highlighting massive capital and energy commitments. Key implications for the AI/ML community include tightened access to large-scale specialized infrastructure, higher electricity and environmental footprints, potential supply-chain and cost risks, and an open question—raised by experts—about whether sheer scale will continue to guarantee model leadership.