[Revised February 21, 2026]

The Stargate Project: An Overview

In January 2025 OpenAI unveiled “The Stargate Project,” a massive AI infrastructure venture to build world-scale data centers. The new entity (funded by OpenAI, SoftBank, Oracle, and Abu Dhabi’s MGX) will invest an initial $100 billion, ramping up to $500 billion over the next four years to deploy AI compute in the U.S. (^[1]) (^[2]). SoftBank (CEO Masayoshi Son as chairman) handles the financing, while OpenAI manages operations (^[3]) (^[4]). In December 2025, SoftBank completed a $41 billion investment in OpenAI, securing roughly an 11% stake — one of the largest private funding rounds ever (^[5]). This is framed as a public–private effort to “secure American leadership in AI,” creating jobs and strategic capability (^[1]) (^[2]). Key technology partners include Microsoft, NVIDIA, Oracle, Arm, and others (^[6]) (^[7]). (OpenAI will continue using Microsoft Azure alongside Stargate’s on-premises infrastructure (^[8]).) The project grew out of a White House event on January 21, 2025 (^[9]) (^[10]), and aims to meet explosive AI compute demand with a 10-gigawatt (GW) target (roughly 10 GW is enough to power ~7.5 million homes).

U.S. Data Center Rollout

Abilene, Texas: Stargate’s Flagship Site

The first Stargate campus is in Abilene, Texas, a consortium-built facility on a vast site. Construction began almost immediately after the announcement. Oracle is overseeing Abilene, which is being developed on leased land with Crusoe Energy reportedly involved. Published reports say the Abilene buildout could reach nearly 1 GW of power by mid-2026, costing on the order of $3–4 billion (^[11]). At a White House briefing Oracle’s Larry Ellison noted that “10 [half-million-square-foot] buildings” are under construction here (^[12]) (^[11]). By late 2025 the Abilene campus already employed 6,400+ construction workers (^[13]). Sam Altman says portions of Abilene are "up and running", with the first two buildings operational since September 2025 and NVIDIA GB200 AI racks already delivered for early model training (^[14]). Oracle will ultimately deploy over 450,000 GB200 GPUs at the Abilene campus under a 15-year lease agreement, with the remaining six buildings expected to be completed by mid-2026 (^[15]). The full Abilene campus will use 1.2 GW of power — enough for roughly one million homes (^[16]).

Expanding Beyond Abilene

Stargate quickly expanded to multiple new U.S. campuses. By mid-2025 OpenAI announced a partnership with Oracle to add 4.5 GW more of U.S. data center capacity (^[17]). Together with Abilene’s 1 GW, this puts over 5 GW under development, well on track toward the 10 GW goal. In September 2025, OpenAI, Oracle and SoftBank announced five more U.S. sites. Three are Oracle-led campuses (in Shackelford County, Texas; Doña Ana County, New Mexico; and a site in the U.S. Midwest) and two are SoftBank-led (Lordstown, Ohio, and Milam County, Texas) (^[18]) (^[19]). These five new campuses – together with Abilene and CoreWeave partnerships – push total planned capacity to about 7 GW (^[20]) (^[21]), on course to hit the full 10 GW, $500B commitment by end-2025, ahead of schedule (^[20]) (^[22]).

• Oracle sites (total ~5.5 GW) – The three Oracle-built campuses (initially announced July 2025) plus a 600 MW expansion at Abilene are expected to deliver about 5.5 GW combined (^[23]) (^[24]). (The Shackelford County site, for example, is a ~1.4 GW hyperscale campus being developed by Vantage and Oracle (^[25]).) Oracle is also contracted to supply OpenAI with cloud services (~$300B over 5 years) (^[26]) and is delivering NVIDIA GPU racks to Abilene (^[27]) (^[14]).
• SoftBank sites (total ~1.5 GW) – Two new campuses are via SoftBank: one in Lordstown, Ohio, and one in Milam County, Texas (^[28]). Together they can scale to ~1.5 GW in 18 months (^[28]) (^[29]).The Lordstown facility uses SoftBank’s advanced data center design and should be online soon, while the Milam site is developed with SB Energy (SoftBank’s power subsidiary) providing accelerated power infrastructure (^[30]) (^[29]).

In October 2025, OpenAI announced another major campus: Stargate Michigan, a 1 GW, $7 billion facility in Saline Township developed by Related Digital and Oracle. Dubbed "The Barn" (for a historic red barn preserved at the entrance), it will consist of three 550,000-sq-ft buildings on 250 acres, with construction beginning in early 2026 and over 2,500 union construction jobs (^[31]) (^[32]). Michigan Governor Whitmer called it "the largest economic project in the state's history."

In January 2026, OpenAI and SoftBank each invested $500 million in SB Energy (SoftBank's power subsidiary), totaling a $1 billion commitment to accelerate data center development. SB Energy was selected to build and operate the 1.2 GW Milam County, Texas campus, with initial facilities expected to enter service in 2026 (^[33]) (^[34]).

In all, these U.S. Stargate centers are each designed for enormous scale (multi-hundred-megawatt buildings). For perspective, Larry Ellison said each building is about 500,000 square feet, and ten such buildings are already being built (^[11]). Adding the Michigan campus, the five September 2025 sites (Shackelford, Doña Ana, Lordstown, Milam, and a Midwest location), and Abilene brings planned compute to well over 8 GW — putting the project firmly on track for its 10 GW goal. OpenAI states this will create tens of thousands of jobs (25,000 onsite in the new sites alone) (^[35]) (^[36]), though independent sources caution that long-term employment is much lower (Bloomberg reported the Abilene site might yield only ~57 permanent jobs (^[37])).

International Expansion (“OpenAI for Countries”)

Alongside U.S. buildout, OpenAI has launched Stargate projects abroad under its “OpenAI for Countries” initiative (^[38]) (^[39]). The goal is to partner with friendly governments and firms to build local AI data centers that mirror Stargate’s capabilities.

• UAE (May 2025): In late May 2025 OpenAI announced Stargate UAE, a 1 GW data center venture in the United Arab Emirates (^[40]). Partners include AI firm G42, Oracle, Nvidia, Cisco, and SoftBank, with UAE investors also funding the project. Initially 200 MW will be deployed by 2026 (eventually up to 1 GW) (^[41]). OpenAI notes Stargate UAE will serve “up to half the world’s population” within ~2000 mi radius (^[40]). The deal benefited from U.S. export-rule changes to the UAE and was publicly lauded by officials (OpenAI thanked President Trump for support in the Middle East) (^[42]).

• Norway (July 2025): In July 2025 OpenAI announced Stargate Norway, its first European AI data center initiative (^[43]). The site is in Narvik, leveraging abundant hydropower. Initial capacity is 230 MW (planned 100,000 NVIDIA GPUs by end-2026), with room to expand an extra 290 MW (^[39]). This facility will run entirely on renewables with state-of-the-art liquid cooling, and even supply waste heat to local industry (^[44]). The joint venture project (with local partners Nscale and Aker) aims to feed compute into Norway’s and Northern Europe’s AI ecosystem (^[44]) (^[45]).

• United Kingdom (Sept 2025): In September 2025 OpenAI unveiled Stargate UK, a partnership with NVIDIA and UK firm Nscale (^[46]). It will expand Nscale’s planned sites (e.g. Cobalt Park) to provide “sovereign” UK compute. OpenAI will initially offtake ~8,000 GPUs in early 2026, scaling to 31,000 GPUs over time (^[47]). The aim is onshore specialist AI workloads (public services, finance, etc.) so that “world-leading AI models can run on local computing power” in the UK (^[46]) (^[47]). Sam Altman emphasized this builds on a July 2025 U.S.–UK collaboration pact and is part of broader efforts to share AI across partners (^[48]) (^[49]).

• Argentina (Oct 2025): OpenAI and Sur Energy signed a Letter of Intent for Stargate Argentina, a large-scale data center in Patagonia capable of reaching 500 MW, representing an estimated investment of up to $25 billion. The first 100 MW phase is targeted for 2027 (^[50]) (^[51]).

• South Korea (Oct 2025): OpenAI struck deals with Samsung Electronics and SK Hynix for high-bandwidth memory (HBM) chip supply, while SK Telecom signed an MoU to develop, build, and operate AI data centers in South Korea as part of the Stargate ecosystem. Initial plans call for two Korean facilities starting at 20 MW (^[52]) (^[53]).

• Other Countries: OpenAI has said it is exploring up to 10 international projects under the "OpenAI for Countries" initiative (^[38]). Discussions or memorandums have been reported with Estonia and the EU's "AI Gigafactories" initiative, and OpenAI executives have been touring Asia to scout potential sites in India, Japan, Singapore, and Australia (^[54]) (^[55]).

Overall, Stargate’s global network is designed to deliver “democratic AI” infrastructure: data-sovereign clouds in allied nations that follow shared principles, rather than authoritarian alternatives (^[38]) (^[44]).

Technology and Capacity

The Stargate centers will be stuffed with cutting-edge AI hardware. OpenAI and partners are locking in massive GPU supply. NVIDIA signed on heavily: as of Sept 2025 OpenAI announced receiving a $100 billion investment from NVIDIA to purchase its AI processors for these data centers (^[56]) (^[57]). Likewise, AMD agreed in Oct 2025 to provide GPUs (announced at 6 GW of capacity for future deployment). Beyond third-party GPUs, OpenAI is developing its own custom AI chip — codenamed "Titan" — in collaboration with Broadcom, to be fabricated on TSMC's 3nm process. Mass production is targeted for the second half of 2026, with the chips optimized for inference workloads. OpenAI's in-house chip team, led by Richard Ho, has doubled to around 40 engineers. The company plans to deploy these custom accelerators across 10 GW of Stargate capacity, reducing its long-term reliance on NVIDIA (^[58]) (^[59]). Additionally, NVIDIA's next-generation Vera Rubin architecture is slated to power the first gigawatt of Stargate infrastructure coming online in the second half of 2026, succeeding the current Blackwell GB200 racks (^[60]).

Each Stargate campus is envisioned as a hyperscale AI factory. For example, Oracle’s Abilene campus will house millions of GPUs (^[57]). The combined sites will require immense power delivery. SoftBank’s SB Energy and local utilities are bringing on accelerated power (e.g. new substation in Milam TX) to meet the gigawatt-scale loads. Renewable sources (solar/wind/hydro) will supply much of the energy; Norway and UAE sites explicitly run on green power (^[44]) (^[61]). Cooling will use techniques like closed-loop liquid cooling to keep GPUs at optimal temperatures (^[44]).

Partnerships extend beyond chips. For instance, Arm (UK chip designer owned by SoftBank) will provide CPU designs in NVIDIA’s next-gen chips (^[62]). CoreWeave (a GPU-focused cloud provider) has a multi-billion-dollar deal with OpenAI to supply additional cloud-based GPU capacity to complement the physical datacenters. Microsoft remains a cloud partner as well, with Azure usage continuing alongside Stargate (^[8]) (^[63]).

Economic Impact and Infrastructure

OpenAI and its partners tout “hundreds of thousands of jobs” and massive economic benefit from the build-out (^[1]) (^[35]). They estimate roughly 100,000–200,000 construction and operations jobs will be created by the U.S. phase (^[64]) (^[36]). (For instance, OpenAI says the five new U.S. sites alone will generate ~25,000 onsite jobs (^[36]).) Work includes electricians, technicians, manufacturing and service roles across all 50 states. Significant investments in power plants, transmission lines, and water systems are also anticipated to support these centers.

However, some experts caution that long-term job numbers may be much lower. Large automated data centers often need only dozens of permanent staff once built. Bloomberg reported that Abilene’s first facility may only employ ~57 ongoing workers, despite promises of thousands (^[37]). Critics also note the enormous energy and water demands of AI facilities (^[65]). Data centers consume ~2% of U.S. electricity, and cooling can use hundreds of thousands to millions of gallons of water per day (^[65]). Local communities will need upgrades to power grids and possibly new generation capacity. Regulators and utilities are closely involved to manage this strain.

Nonetheless, many analysts argue the U.S. needs such “AI megafactories” to stay competitive. Past projects (like nationwide broadband and highway construction) took decades and similar sums, and yielded vast productivity gains. OpenAI’s Llewellyn Thankful (Infrastructure lead) and others compare this to the Interstate Highway or electric grid investments of previous generations, underscoring it as "the compute needed to fuel American innovation" (^[17]) (^[66]).

Criticism and Challenges

Despite broad support from tech and government leaders, Stargate has faced skeptics and hurdles. Most notably, Elon Musk (who has his own AI company xAI and is advising the U.S. administration) publicly lambasted the plan as “fake”, claiming SoftBank hadn’t secured the money to fund it (^[67]) (^[10]). Musk tweeted that “they don’t actually have the money” and that SoftBank had “well under $10B secured” (^[67]) (^[10]). OpenAI’s Altman quickly refuted this on social media, acknowledging that “what is great for the country isn’t always what’s optimal for your companies” and urging Musk to “put 🇺🇸 first” (^[68]). The feud highlighted tensions (Musk and Altman have been public adversaries) but did not derail the project.

Another issue is cost escalation. In mid-2025 trade tariffs began hiking the price of imported servers, chips, and cooling equipment. Bloomberg reporting (cited by TechCrunch) warned that new U.S. tariffs could add 5–15% to data center build costs (^[69]). This, combined with tech market volatility and AI “bubbles,” made some investors wary. By May 2025, reports emerged that SoftBank had not finalized majority funding and was still courting banks and asset managers (^[69]). OpenAI has stated it will cover the initial $100B through equity and debt. SoftBank's completion of its $41 billion OpenAI investment in December 2025 — and the joint $1 billion SB Energy commitment in January 2026 — have substantially addressed earlier funding skepticism (^[5]) (^[33]). However, the full $500 billion commitment still requires ongoing fundraising from banks and asset managers over the coming years (^[70]).

While critics initially noted a "slow start" in 2025, the project has gained significant momentum: two Abilene buildings went operational in September 2025, and multiple new campuses are now under active construction (^[57]). Some observers still wonder whether the full $500B will be deployed or if projections will grow further. Altman himself has acknowledged the difficulty of building U.S. infrastructure, and has called for easing bureaucratic hurdles (^[71]).

Finally, national-security and economic concerns are debated. An analyst noted that the underlying training data and control of AI models matter: if government-sponsored servers run U.S.-centric AI, that is seen as a benefit; but if foreign governments could leverage this infrastructure, questions arise (^[10]). OpenAI stresses sovereign compute (especially in “OpenAI for Countries” projects) and has engaged allied governments directly to assuage these worries (^[46]) (^[72]).

Status as of February 2026 and Outlook

As of early 2026, Stargate has moved from vision to reality at remarkable scale. The Abilene flagship campus has two buildings operational since September 2025, with the remaining six expected to complete by mid-2026 — ultimately housing over 450,000 NVIDIA GB200 GPUs (^[15]). Six additional U.S. campuses are in various stages of development (Shackelford County TX, Doña Ana County NM, Lordstown OH, Milam County TX, a Midwest site, and Saline Township MI), with land secured and permits progressing (^[73]) (^[31]). SoftBank has completed its $41 billion investment in OpenAI, and the two companies jointly invested $1 billion in SB Energy to accelerate power infrastructure for the buildout (^[5]) (^[33]). OpenAI claims over $400 billion of the $500B commitment is already in play (^[20]).

Internationally, the picture has expanded significantly. Four overseas Stargate projects have been announced — UAE, Norway, UK, and Argentina — with South Korea joining as a hardware supply and data center partner. OpenAI continues to explore sites in India, Japan, Singapore, Australia, and through EU partnerships (^[74]) (^[55]).

On the technology front, two major milestones are expected in the second half of 2026: the rollout of NVIDIA's next-generation Vera Rubin GPU architecture across Stargate sites, and mass production of OpenAI's first custom "Titan" AI chip — fabricated by TSMC on its 3nm process and designed in partnership with Broadcom (^[59]).

Going forward, OpenAI will continue selecting further U.S. site locations from hundreds of proposals (^[75]), and its global "10-country" rollout is gaining momentum. Supply-chain deals (with NVIDIA, AMD, Samsung, SK Hynix, and others) continue to expand. Analysts will watch project financing, regulatory approvals, and power delivery — any delays in energy permits or grid upgrades could slow progress.

In summary, as of February 2026, Stargate is a sprawling, multi-continental initiative to build roughly 10 GW of AI data center capacity underwritten by an unprecedented $500 billion investment. It is led by OpenAI with SoftBank and Oracle, leveraging a wide circle of technology partners from chipmakers to energy developers. The technical details (sites, power, GPUs, custom silicon) are being fleshed out in real time, as are the economics (jobs, local benefits vs resource costs). With two Abilene buildings already running and multiple new campuses under construction, Stargate is on track to be one of the largest infrastructure projects ever undertaken in the tech sector (^[1]) (^[20]), positioning its backers — and arguably the U.S. — at the forefront of next-generation AI.

Sources: Official OpenAI announcements and leading news reports (OpenAI blog, TechCrunch, CNBC, DCD, The Register, Broadcom, etc.) provide the above details (^[76]) (^[73]) (^[77]) (^[31]) (^[34]) (^[78]) (^[79]) (^[80]) (^[81]) (^[59]). These cover project scope, funding, partners, locations, technology, and critical commentary.