OpenAI and Oracle Commit 4.5 GW Data Center Expansion to Stargate Network

OpenAI and Oracle today announced a binding agreement to add 4.5 gigawatts of power-dense data center capacity to the Stargate network, pushing the project past the 5 GW milestone and keeping it on track to house over two million AI chips. While SoftBank remains a name on Stargate’s January press release, the Japanese conglomerate is not contributing capital to this specific tranche, leaving Oracle as the primary financier and operator for now. The new build-out will join Stargate I in Abilene, where Oracle delivered the first NVIDIA GB200 racks last month and where OpenAI is already running frontier model training workloads. Construction crews from more than twenty states are on site, and OpenAI projects that the expanded footprint will generate over 100,000 US jobs, spanning electricians, mechanical technicians, and indirect manufacturing roles, within the next four years.

The announcement arrives less than a week after the Wall Street Journal characterized Stargate as a “stalled” venture, a claim OpenAI CEO Sam Altman swiftly rebutted on X, affirming that total investment will “significantly exceed” the originally pledged $500 billion and that GPU counts are on pace to break the two‑million‑chip barrier this year. Site selection for the 4.5 GW increment remains under review, with Texas, Michigan, Wisconsin, and Wyoming under consideration. Oracle has reportedly placed a $40 billion order for additional NVIDIA silicon, showing the scale of its commitment. Power availability remains the critical constraint, as 4.5 GW equals the average draw of 3.5 million homes, prompting parallel negotiations with utilities for on-site generation and dedicated transmission lines. Abilene’s H-shaped halls, each designed to hold 50,000 GB200s, serve as a live prototype, and lessons learned there, including liquid-cooling loops and modular substations, will be replicated across future campuses.