OpenAI and Oracle Commit to 4.5 GW of New Stargate Data Centers

Oracle and OpenAI have signed an agreement to develop 4.5 gigawatts of additional Stargate data center capacity in the United States. The deal, announced by OpenAI, pushes total Stargate capacity under development past 5 gigawatts — enough, by the company's own accounting, to run more than 2 million chips.

The scale is the story. Four-and-a-half gigawatts roughly equals the output of several large power plants, and OpenAI says it will now likely exceed the $500 billion, 10-gigawatt infrastructure commitment it announced at the White House in January, thanks to what it calls "strong momentum" with partners including Oracle and SoftBank. That January pledge — four years, half a trillion dollars, U.S. soil — was already the largest private infrastructure commitment of the AI buildout era. This agreement moves OpenAI meaningfully closer to it, ahead of schedule.

What the deal covers

The announcement frames the Oracle agreement across three dimensions: capacity, jobs, and geopolitics.

On capacity, the 4.5 GW of new data centers join Stargate I in Abilene, Texas, bringing the program past 5 GW under development. OpenAI did not disclose in this announcement how the new capacity splits across sites, nor the dollar value of the Oracle agreement itself. What it did disclose is the trajectory: the company now expects to exceed its initial $500 billion commitment.

On jobs, OpenAI estimates that building, developing, and operating the additional 4.5 GW will create more than 100,000 jobs in the United States. That figure bundles three categories: direct full-time roles needed to run Stargate data centers, short-term construction positions, and indirect employment in manufacturing and local services. The company highlights the specialized electricians already at work in Abilene as an example of the construction pipeline. Across the full Stargate program, OpenAI says it expects "hundreds of thousands of jobs" over the coming years.

On economics and policy, OpenAI describes the investment as one that will "accelerate America's reindustrialization" and "help advance U.S. AI leadership." The company credits "important leadership from the White House, which has recognized the critical role AI infrastructure will play in driving innovation, economic growth, and national competitiveness." The framing positions Stargate not as a conventional cloud contract but as an industrial policy instrument — one in which private capital, chip supply, and federal priorities point in the same direction.

Abilene is no longer a render

The most concrete evidence that Stargate is real sits in Texas. Construction of Stargate I in Abilene, OpenAI says, is progressing, and parts of the facility are already operational. Oracle began delivering the first Nvidia GB200 racks last month. OpenAI has started running early training and inference workloads on that capacity, and says it is using the site to "push the limits" of its next-generation frontier research.

That detail matters for anyone tracking the frontier race. Training runs on GB200 hardware, on dedicated gigawatt-scale campuses rather than shared cloud capacity, translate directly into how fast OpenAI can iterate on its most advanced models. Compute is the binding constraint on frontier AI, and Stargate is OpenAI's structural answer to it.

The Abilene site has already created thousands of jobs, according to the company, with more expected as operations expand. The workforce includes specialized roles for electricians, equipment operators, and technicians who have come from more than 20 states — a small data point that captures how the AI buildout is pulling labor nationally, not just locally.

The partner map: Oracle, SoftBank, CoreWeave — and still Microsoft

The Oracle agreement is one pillar of a broader structure. Stargate, as OpenAI now defines it, is the company's "overarching AI infrastructure platform," encompassing data center partnerships with Oracle, SoftBank, and CoreWeave, plus international investments in U.S. infrastructure through its "OpenAI for Countries" program.

The SoftBank track is moving "with strong momentum," according to OpenAI. The two companies are conducting site assessments and, in OpenAI's words, "reimagining how data centers are designed to power advanced AI." Both the Oracle and SoftBank relationships are described as "critical to meeting OpenAI's continually expanding compute needs."

One line in the announcement will draw attention from anyone following the OpenAI–Microsoft relationship: "Microsoft will continue to provide cloud services for OpenAI, including through Stargate." After years in which Microsoft was effectively OpenAI's sole infrastructure provider, OpenAI has diversified across Oracle, SoftBank, and CoreWeave. This statement confirms that Microsoft remains in the stack — but as one supplier among several, embedded within a program it no longer defines.

OpenAI also notes that growing interest in Stargate over the past six months has "helped shape and expand" its vision for the effort. Stargate began as a specific project announcement in January; it is now a platform brand covering multiple partners, multiple sites, and an international investment channel.

Why the stakes are this large

The context is a compute arms race with defined numbers attached. OpenAI's January commitment — $500 billion and 10 gigawatts over four years, announced at the White House — set the benchmark for what a frontier lab believes it needs to build and train next-generation models. Rivals have made their own multi-hundred-billion-dollar commitments. Power, land, chips, and skilled labor are the scarce inputs, and the companies that lock them in earliest will set the pace of frontier model development.

Oracle's role in this dynamic is commercially significant. The company has repositioned itself as a major AI infrastructure provider, and a 4.5 GW agreement with OpenAI deepens that bet on AI-driven cloud demand. For Nvidia, the deal implies continued large-scale deployment of its data center hardware — GB200 racks are already shipping to Abilene.

The White House dimension is not decorative. OpenAI explicitly ties the program to national competitiveness, and the January announcement venue signaled administration backing for AI infrastructure as an economic and strategic priority. Stargate has become the flagship example of that alignment: private capital building sovereign-scale compute, with federal encouragement, on U.S. soil.

What comes next

OpenAI says it now expects to exceed the original $500 billion, 10-gigawatt commitment, with more capacity announcements likely as SoftBank site assessments progress and Abilene scales toward full operation. The near-term milestones to watch are concrete: additional GB200 deliveries, expanded training workloads in Texas, and the disclosure of where the new 4.5 GW will physically land. If those sites materialize on the timeline implied by the January commitment, Stargate will have moved from announcement to industrial reality inside a single year — and the bottleneck for frontier AI will shift, at least partially, from capital to construction.