Stargate is an artificial intelligence computing infrastructure initiative announced on January 21, 2025, initially focused on building facilities for OpenAI in the United States. Its founding announcement identified SoftBank, OpenAI, Oracle, and MGX as initial equity funders and set an intended investment of US$500 billion over four years, beginning with US$100 billion. Subsequently, OpenAI used Stargate more broadly for its network of data-center and computing partnerships, including international projects. The announced investment and capacity targets describe commitments and development plans, rather than expenditure already completed or facilities already operating. (openai.com)
Establishment and organization
Stargate was publicly introduced at the White House on January 21, 2025. The founding announcement described it as a new company, with SoftBank responsible for finance and OpenAI responsible for operations. Masayoshi Son was designated chairman. Arm, Microsoft, Nvidia, Oracle, and OpenAI were named as the initial technology partners. Construction was already underway in Texas, while additional locations were being evaluated. (group.softbank)
The project's scope expanded beyond that initial corporate description. By July 2025, OpenAI described Stargate as its overarching AI infrastructure platform, encompassing partnerships with Oracle, SoftBank, and CoreWeave, together with international investment associated with OpenAI for Countries. Microsoft continued to provide cloud services to OpenAI, including through Stargate. Consequently, a project described as part of Stargate need not have the same developers, financing arrangements, or ownership structure as another Stargate campus. (openai.com)
Technical purpose
Stargate supplies computing capacity for both training AI models and inference, the operation of trained models to produce outputs. Its flagship campus in Abilene, Texas, operates on Oracle Cloud Infrastructure and uses Nvidia GB200 systems, incorporating graphics processing units designed for large-scale AI workloads. OpenAI reported that early training and inference workloads were running there in 2025. (group.softbank)
The initiative involves more than purchasing processors. Its site-selection process sought land, access to electricity, utility expansion potential, water, construction access, and local-government engagement. OpenAI's request for proposals allowed submissions covering combinations of land, power, building design, and equipment fit-out. This made electricity availability and the ability to deliver an operational campus central elements of the development process. (cdn.openai.com)
Development in the United States
Abilene
Abilene became the first operating Stargate campus and the project's flagship site. Oracle began delivering Nvidia GB200 racks in June 2025, and by September the partners reported that the campus was running training and inference workloads while construction continued. (group.softbank)
The operating campus should be distinguished from proposed adjacent developments. In March 2026, Microsoft took over a separate expansion involving two additional buildings and an on-site power plant after OpenAI stepped back from that expansion. This did not constitute cancellation of the existing OpenAI–Oracle Stargate campus. (apnews.com)
Additional campuses
In July 2025, OpenAI and Oracle announced an agreement to develop an additional 4.5 gigawatts of Stargate data-center capacity. In September, OpenAI, Oracle, and SoftBank announced five further locations: Shackelford County, Texas; Doña Ana County, New Mexico; an initially unidentified Midwestern location; Lordstown, Ohio; and Milam County, Texas. Together with Abilene and projects involving CoreWeave, the partners reported nearly 7 gigawatts of planned capacity and more than US$400 billion in prospective investment over three years. These were development figures, not a statement that all the capacity was operational. (openai.com)
On January 9, 2026, OpenAI and SoftBank announced that each would invest US$500 million in SB Energy. OpenAI also signed a lease for a 1.2-gigawatt data center in Milam County, selecting SB Energy to build and operate the site. The arrangement linked campus development with associated energy infrastructure. (openai.com)
On April 29, 2026, OpenAI stated that it had surpassed its objective of securing 10 gigawatts of United States AI infrastructure, originally targeted for 2029. That wording concerns capacity secured through its development and partnership arrangements; it should not be read as confirmation that 10 gigawatts of completed facilities were simultaneously in service. (openai.com)
International extension
OpenAI introduced OpenAI for Countries on May 7, 2025, as an initiative within Stargate. Its proposed partnerships combined development of in-country data-center capacity with investment in the wider Stargate effort. The initial stated goal was to pursue ten projects with countries or regions. (openai.com)
On May 22, 2025, OpenAI announced Stargate UAE, describing it as Stargate's first international deployment. The agreement involved G42, Oracle, Nvidia, Cisco, and SoftBank and proposed a 1-gigawatt cluster in Abu Dhabi, in the United Arab Emirates. At announcement, the partners expected an initial 200-megawatt phase to enter service in 2026. The agreement also included UAE investment in United States Stargate infrastructure; the announcement itself was not evidence that the proposed cluster had been completed. (openai.com)
Energy, cooling, and community commitments
Electricity supply, grid connections, cooling, and local infrastructure are major implementation requirements. In January 2026, OpenAI introduced a Stargate Community framework, committing to locally tailored plans and to covering energy-related costs so that Stargate operations would not increase other customers' electricity prices. It outlined different arrangements for individual sites, including new generation and storage in Texas and project-funded energy infrastructure in Wisconsin and Michigan. These statements describe commitments and proposed mechanisms, rather than independently established long-term outcomes. (openai.com)
The framework also prioritized closed-loop or low-water cooling. In its April 2026 account of Abilene, OpenAI stated that cooling water circulates through sealed pipes rather than being routinely consumed in evaporative cooling towers. Such descriptions are specific to the cooling design and should not be treated as a complete accounting of every direct and indirect water requirement associated with the campus. (openai.com)
Interpreting investment and capacity figures
Stargate announcements distinguish several stages that are not interchangeable: proposed investment, financing commitments, contracted capacity, construction, equipment installation, and operational computing capacity. The original US$500 billion figure was an intended four-year investment, while later announcements reported planned campuses and capacity secured through partnerships. Neither figure alone establishes actual expenditure or the amount of computing available at a particular date. (openai.com)
Employment projections also depend on their definitions. OpenAI's estimate associated with the additional 4.5-gigawatt Oracle agreement included permanent operating positions, temporary construction work, and indirect employment in manufacturing and local services. It was therefore not a count of permanent data-center employees. Delivery schedules likewise depend on land, permits, transmission infrastructure, workforce availability, community support, and partner readiness—conditions that OpenAI identified as requirements for further expansion. (openai.com)