Stargate Project
Last reviewed
May 8, 2026
Sources
55 citations
Review status
Source-backed
Revision
v1 · 5,737 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 8, 2026
Sources
55 citations
Review status
Source-backed
Revision
v1 · 5,737 words
Add missing citations, update stale details, or suggest a clearer explanation.
The Stargate Project is an American AI infrastructure joint venture announced on January 21, 2025 at the White House by SoftBank, OpenAI, Oracle, and MGX, with President Donald Trump. The partners committed to deploying $100 billion immediately and stated an intention to invest up to $500 billion over four years to build new AI compute infrastructure for OpenAI inside the United States. SoftBank chairman Masayoshi Son took the role of chairman; SoftBank holds financial responsibility while OpenAI holds operational responsibility, with Oracle, MGX, Microsoft, Nvidia, and Arm named as initial technology partners. According to OpenAI's own announcement, the venture's first site was already under construction in Abilene, Texas, with additional campuses to be evaluated across the country [1][2][3].
Stargate was unveiled the day after Trump's second inauguration and one day after he revoked the Biden administration AI executive order 14110. It was the first major industrial announcement of the new administration and was framed both by the White House and by the partners as the largest private AI infrastructure commitment in history. Trump indicated that the federal government would use emergency declarations and accelerated permitting to speed up the buildout, a posture that was later codified in the July 2025 AI Action Plan and accompanying executive orders, including Executive Order 14318 on data center permitting [4][5][6].
The name had been used inside OpenAI and Microsoft a year earlier for an unrelated supercomputer concept reported by The Information in March 2024, a roughly $100 billion, 5 GW project that Microsoft would have built and OpenAI would have rented. The 2025 Stargate Project is a different vehicle with different financial sponsors, and its launch coincided with a renegotiation of the Microsoft-OpenAI cloud agreement that ended Microsoft's exclusive right to host OpenAI workloads, replacing it with a right of first refusal on incremental capacity [7][8][9].
Progress through 2025 and into 2026 has been uneven. The first two buildings of the flagship Abilene, Texas campus, developed by Crusoe Energy on Lancium's 875-acre Clean Campus, were energized in mid-to-late 2025 and announced as live on Oracle Cloud Infrastructure on September 30, 2025. Five additional U.S. sites were announced on September 23, 2025, taking planned capacity to roughly 7 GW, and an international Stargate UAE campus, an Argentina partnership, and short-lived Norway and United Kingdom projects followed. By early 2026 several of the international components had been scaled back or handed to other operators, and parts of the Abilene expansion were dropped, with Crusoe capacity reportedly redirected to Meta. Independent reporting raised persistent questions about whether the headline $500 billion figure was a binding commitment or a forward-looking aspiration [10][11][12].
The Stargate name first appeared publicly in March 2024, when The Information reported that Microsoft and OpenAI were drafting plans for a five-phase data center program culminating in a single supercomputer, internally codenamed Stargate, that could cost more than $100 billion and draw roughly five gigawatts of power. The reported goal was to have Stargate online around 2028 to train OpenAI's next generation of frontier models. Fortune, The Register, Tom's Hardware and others picked the story up; the name was reportedly inspired by the 1994 film of the same name [7][8][13].
Under that earlier conception, Microsoft would have been the sole financial sponsor and OpenAI would have been the tenant. Microsoft was already OpenAI's exclusive cloud provider under their 2019-2023 commercial agreements, and Azure had hosted every public ChatGPT and GPT-4 deployment. Capital expenditure at Microsoft was already running at historically high levels because of OpenAI workloads; in early January 2025, Microsoft formalized an $80 billion AI-enabled data center spend for fiscal year 2025, which CEO Satya Nadella described as the largest single-year corporate technology investment ever announced [14].
The 2025 Stargate Project replaced the original Microsoft-only plan with a multi-investor vehicle, and OpenAI used the announcement as the moment to renegotiate its cloud terms. Microsoft published a corresponding statement on January 21, 2025 confirming that it was now entitled to a right of first refusal on incremental OpenAI compute, but no longer held an exclusive lock on it. OpenAI, in turn, would be able to procure compute from other providers if Microsoft passed. The Financial Times and others reported that the new infrastructure built under Stargate would be "exclusive to OpenAI," a statement that paraphrased the OpenAI announcement [9][15].
Over the following year that decoupling went further. In April 2026 Microsoft and OpenAI announced a new agreement that ended exclusivity on OpenAI intellectual property as well, capped revenue share, and let OpenAI run its products on any cloud, while preserving Microsoft's IP license through 2032 and revenue share through 2030. Stargate was, in effect, the infrastructure side of a broader unwinding of the Microsoft-OpenAI exclusivity that had defined the prior phase of frontier AI [16][17].
The contrast between the 2024 conception and the 2025 launch is worth holding in view. The 2024 Stargate would have been a single company's bet, with Microsoft as the financier and OpenAI as the tenant, deferred to roughly 2028. The 2025 Stargate Project is a multi-investor vehicle with no single corporate balance sheet behind it, sized to break ground immediately, and structured so that Microsoft is a technology partner rather than the financial sponsor. SoftBank's involvement was the most consequential change, because it brought both equity capital and a chairman, Masayoshi Son, who had spent the prior twelve months publicly courting Trump and pitching trillion-dollar U.S. investment plans, including a separate $1 trillion Arizona idea floated to TSMC. SoftBank's role in Stargate was therefore an extension of Son's broader U.S.-investment posture, not a one-off [3][24].
The Stargate Project was unveiled in the Roosevelt Room of the White House on the afternoon of January 21, 2025, Trump's second day in office. Trump appeared with Sam Altman of OpenAI, Larry Ellison of Oracle, and Masayoshi Son of SoftBank. He described the venture as "the largest AI infrastructure project in history" and said the federal government would use emergency declarations to expedite power and permitting. Altman, Son and Ellison each took the lectern; Son said the project would create "the future of America" and that he would "not have done this if Trump had not won." Ellison claimed that the data centers under construction in Abilene were "actually being built right now" and described AI applications including 48-hour mRNA cancer vaccine design [1][3][4].
OpenAI's own press post the same day used careful, lawyered wording. The exact phrasing was that the venture "intends to invest $500 billion over the next four years" and "will begin deploying $100 billion immediately," with SoftBank and OpenAI as "lead partners," Oracle and MGX as additional equity funders, and Microsoft, Nvidia, Oracle and Arm as initial technology partners. The post stated that SoftBank "will have financial responsibility" and OpenAI "will have operational responsibility," and that Masayoshi Son "will be Chairman." That "intends to invest" formulation, rather than "has committed," became the focus of subsequent skepticism [1].
The announcement timing mattered. It came one day after Trump revoked Biden's Executive Order 14110 on the safe and secure development of AI, and on the same day that he signed Executive Order 14179 directing his administration to produce an AI strategy within 180 days. Stargate was therefore positioned as the centerpiece of the new administration's industrial-policy posture toward AI, and as a counter-narrative to Chinese model releases such as DeepSeek-V3, which had drawn attention earlier that month [4][5][6].
The Stargate Project is a private joint venture incorporated as Stargate LLC. The four named equity funders at launch are SoftBank Group, OpenAI, Oracle, and MGX, the Abu Dhabi sovereign-wealth-backed AI investment firm. The technology partner list, which appeared in OpenAI's announcement, was distinct: Arm, Microsoft, Nvidia, Oracle and OpenAI. Oracle therefore had two roles, as both an equity investor and a technology partner, and OpenAI was both the operating entity and a tenant.
| Partner | Role | Stated responsibility |
|---|---|---|
| SoftBank | Equity, lead financial partner | "Financial responsibility"; chairman |
| OpenAI | Equity, lead operational partner | "Operational responsibility"; sole tenant |
| Oracle | Equity and technology partner | Builds and operates data centers; provides Oracle Cloud Infrastructure |
| MGX | Equity | Abu Dhabi-backed AI investment vehicle |
| Microsoft | Technology partner | Right of first refusal on incremental OpenAI compute |
| Nvidia | Technology partner | GB200 / next-generation GPU supplier |
| Arm | Technology partner | CPU IP for Nvidia Grace and partner silicon |
Reporting by The Information in early 2025 indicated that the named equity commitments included roughly $19 billion each from SoftBank and OpenAI, $7 billion each from Oracle and MGX, with the remainder of the $100 billion initial deployment expected to come from limited partners and debt financing. SoftBank chairman Masayoshi Son took the chairman seat; OpenAI COO Brad Lightcap, who became the day-to-day operations lead at OpenAI in March 2025 when Sam Altman shifted toward research and product, was reported by Yahoo Finance and others to be the senior OpenAI executive responsible for Stargate as part of his expanded global growth and infrastructure portfolio [18][19].
OpenAI added Chris Lehane, a former Clinton White House aide who joined OpenAI in 2024, as Chief Global Affairs Officer; Lehane became the public face of Stargate's policy and political work. In late 2025 OpenAI hired former United Kingdom Chancellor of the Exchequer George Osborne to lead the company's global Stargate data center expansion, a role focused on overseas siting and government relations rather than operations [19][20].
Because the announcement and the OpenAI press post used the verb "intends" rather than "has committed," reading the structure carefully matters. At launch, only the initial $100 billion was characterized as funding the partners would deploy in the near term; the path to $500 billion was described as a four-year intention, contingent on additional limited partners and debt. SoftBank's first $10 billion tranche, reported in April 2025, was funded via a syndicated loan from Mizuho Bank and other Japanese lenders rather than from SoftBank's balance sheet alone. JPMorgan Chase agreed in May 2025 to lend $2.3 billion to OpenAI and partners specifically for the Abilene buildout. Bloomberg reported in early August 2025 that the wider $500 billion remained unraised, with market conditions and trade policy cited as headwinds. SoftBank's parallel $40 billion direct investment in OpenAI was a separate transaction, although a portion was earmarked for Stargate use [21][22][23][24].
The Stargate financing stack, as it has assembled in public, mixes equity from the four named partners, project-level debt, and syndicated loans collateralized by partner balance sheets and partner equity stakes.
| Date | Event | Reported amount | Source |
|---|---|---|---|
| Jan 21, 2025 | Announcement: $100 billion immediate, $500 billion four-year intention | $100 billion / $500 billion | OpenAI announcement [1] |
| Apr 2025 | SoftBank to borrow first $10 billion tranche from Mizuho and other lenders | $10 billion | Reuters / Nikkei [21] |
| May 22, 2025 | JPMorgan Chase agrees to lend for Abilene buildout | $2.3 billion | Bloomberg [22] |
| Aug 7, 2025 | Bloomberg reports broader $500 billion not raised; project not yet at scale | n/a | Bloomberg [23] |
| Sep 23, 2025 | OpenAI states the five-site expansion takes program to "nearly 7 GW of planned capacity and over $400 billion in investment over the next three years" | $400 billion-plus planned | OpenAI announcement [10] |
| Oct-Dec 2025 | Lancium closes $600 million debt for Clean Campus development beginning at Abilene | $600 million | PR Newswire / Lancium [25] |
A recurring theme in the financing reporting is that the equity commitments named at launch are smaller than the headline figure, with the gap to be closed by limited partners, project finance, vendor financing, and offtake agreements. The Information and Bloomberg reporting throughout 2025 made clear that $500 billion is the four-year target rather than committed capital, and that getting from announced commitments to disbursed capital depends on individual site economics. SoftBank in particular has used its OpenAI equity stake as collateral for additional borrowing, including a reported $10 billion margin loan against OpenAI shares at SOFR plus 425 basis points [26].
The inaugural Stargate site is in Abilene, Texas, on Lancium's Clean Campus, a roughly 875-acre property that Lancium had been developing since 2020. Crusoe Energy is the general contractor and has built the data center buildings; Oracle holds the lease on the campus and supplies the operating cloud layer through Oracle Cloud Infrastructure (OCI); the Nvidia GB200 NVL72 racks are the first deployed compute. Construction on the first two buildings began in June 2024, before Stargate was publicly announced, and proceeded continuously through 2025 [27][28][29].
The site is sized for 1.2 GW of total capacity once the eight planned buildings are complete. Each of the first two buildings is roughly 980,000 square feet and supports about 200 MW of IT load. The campus connects into the Electric Reliability Council of Texas (ERCOT) grid through a 1.2 GW interconnect that received formal ERCOT approval after a multi-year review with the relevant transmission service providers. Lancium's energy strategy combines wind and solar from the regional grid, on-campus solar generation, large-scale battery storage, and roughly 360 MW of on-site natural gas turbines for firm backup power. Crusoe stated the buildings use a combined liquid and air cooling design optimized for high energy-density AI hardware [27][30][31].
Oracle began delivering the first Nvidia GB200 racks to Abilene in June 2025. Crusoe announced on September 30, 2025 that the first phase of the campus was live on OCI and "continues to progress rapidly," running early training and inference workloads [29]. Lancium closed a $600 million debt financing in late 2025 to advance Clean Campus development, beginning at Abilene [25]. Construction is scheduled to continue at Abilene through early 2027 according to local briefings to the Abilene city government [32].
| Item | Abilene, Texas detail |
|---|---|
| Developer | Crusoe Energy (general contractor) |
| Land owner / energy partner | Lancium (Clean Campus, since 2020) |
| Operator / cloud layer | Oracle / Oracle Cloud Infrastructure |
| Tenant | OpenAI (sole) |
| Total planned capacity | 1.2 GW across 8 buildings |
| Building 1-2 IT load | 200 MW each, 980,000 sq ft each |
| Compute | Nvidia GB200 NVL72 racks; first deliveries June 2025 |
| Grid | ERCOT 1.2 GW interconnect, approved |
| On-site generation | Solar, large-scale battery storage, ~360 MW natural gas turbines for firm power |
| First phase live | September 30, 2025 |
| Construction window | June 2024 to early 2027 |
The Abilene site has also been the subject of a dispute over future expansion. In early 2026, DCD and other trade outlets reported that Oracle and OpenAI had walked away from a planned further expansion of the flagship Abilene campus and that Meta was in talks, with Nvidia's involvement, to take over a portion of Crusoe's available capacity at the site. The reporting did not affect the existing Stargate buildings already energized, but it did signal that Stargate's footprint at Abilene was capped earlier than originally implied [11].
On September 23, 2025, OpenAI, Oracle and SoftBank announced five additional U.S. Stargate sites, which the partners said took the project to roughly 7 GW of planned capacity and more than $400 billion of investment over the next three years. OpenAI characterized the announcement as putting the program "on a clear path to securing the full $500 billion, 10-gigawatt commitment we announced in January by the end of 2025, ahead of schedule" [10][33].
| Site | State / county | Lead | Notes |
|---|---|---|---|
| Abilene Clean Campus | Taylor County, Texas | Crusoe / Lancium / Oracle | First site; 1.2 GW; first phase live Sep 30, 2025 |
| Shackelford County | Texas | Oracle | Co-located with Vantage Data Centers' Frontier project; ~1,200-acre campus |
| Doña Ana County | New Mexico | Oracle | Adjacent to Project Jupiter campus by Stack and BorderPlex Digital Assets |
| Milam County | Texas | SoftBank / SB Energy | SB Energy delivers powered shell for fast-build site |
| Lordstown | Ohio | SoftBank | Advanced data-center design; SoftBank had broken ground by announcement |
| Undisclosed Midwestern site | Midwestern U.S. | Mixed | Location not publicly disclosed at announcement |
| Stargate UAE | Abu Dhabi, UAE | G42 / Oracle / Nvidia / SoftBank / Cisco / OpenAI | 1 GW announced; first 200 MW phase due Q3 2026 |
| Stargate Argentina | Patagonia, Argentina | Sur Energy / OpenAI | Up to $25 billion; up to 500 MW; first 100 MW phase targeted 2027 |
| Stargate Norway (cancelled) | Narvik, Norway | Nscale / OpenAI | Originally 230 MW with growth to 520 MW; transferred to Microsoft tenancy in early 2026 |
| Stargate UK (cancelled) | United Kingdom | Nscale / OpenAI | Halted citing electricity costs |
Three of the five September U.S. sites are Oracle-led; two are SoftBank-led, including Lordstown, Ohio, where SoftBank had already broken ground when the announcement was made, and Milam County, Texas, where SoftBank's energy subsidiary SB Energy is providing powered infrastructure. The Shackelford County site is contiguous with Vantage's $25 billion Frontier project. The Doña Ana County, New Mexico site sits adjacent to the planned $165 billion Project Jupiter being developed by Stack Infrastructure and BorderPlex Digital Assets [10][34].
In May 2025 OpenAI announced Stargate UAE, the first international Stargate site, in partnership with G42, Oracle, Nvidia, SoftBank, and Cisco. Stargate UAE is part of a broader 5 GW UAE-U.S. AI Campus and will debut with a 1 GW Stargate cluster, of which the first 200 MW phase is scheduled to come online in the third quarter of 2026. The UAE site is to be powered by a mix of nuclear, solar, and natural gas [35][36].
In July 2025 OpenAI announced Stargate Norway, an initial 230 MW site in Narvik, with planned growth to 520 MW, leveraging hydropower and built by Nscale. In October 2025, OpenAI and Sur Energy signed a letter of intent to build Stargate Argentina in Patagonia, with up to 500 MW of capacity and an estimated investment of up to $25 billion under Argentina's RIGI investment regime, with first 100 MW targeted for 2027. A United Kingdom project was also explored briefly [37][38][39].
By early 2026, however, both the United Kingdom and Norway projects had been pulled, with reporting attributing the U.K. cancellation to electricity costs and the Norway facility being handed to Microsoft, which agreed to take it over with a multi-year rental of roughly 30,000 Nvidia Vera Rubin chips through Nscale. The international Stargate footprint contracted accordingly during the first quarter of 2026 [12][40].
Stargate is the most visible private piece of an industrial-policy push that took shape in the first six months of the second Trump administration.
On January 20, 2025, Trump revoked Biden's Executive Order 14110 on safe AI development. On January 23, 2025 he signed Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence," which directed senior White House officials to produce an AI strategy within 180 days. That deliverable, America's AI Action Plan, was released on July 23, 2025 with three pillars: accelerating innovation, building infrastructure, and leading in international diplomacy and security [4][5][6].
Released on the same day as the AI Action Plan were three accompanying executive orders: EO 14318 ("Accelerating Federal Permitting of Data Center Infrastructure"), EO 14319 (preventing "woke AI" in federal procurement), and EO 14320 (the American AI Exports Program). EO 14318 in particular was structured around "Qualifying Projects," defined to include data centers with more than $500 million in capital and electric load additions greater than 100 MW, exactly the size class of Stargate sites. The order created NEPA categorical exclusions, federal financial-support mechanisms (loans, loan guarantees, grants, tax incentives, offtake agreements), and a process to identify federal sites suitable for data center construction. The Department of Energy announced four federal sites under the program: Idaho National Laboratory, Oak Ridge Reservation, Paducah Gaseous Diffusion Plant, and Savannah River Site [41][42][43].
The AI Action Plan itself referenced the January 21 Stargate announcement as evidence that the United States was already attracting private commitments at scale. EO 14318 created the federal toolkit that Stargate sites and other large data center developments would lean on for permitting and energy.
The sharpest public criticism came within hours of the announcement, from Elon Musk, CEO of xAI and a former co-founder of OpenAI. On January 22, 2025, Musk wrote on X that "They don't actually have the money," and added that "SoftBank has well under $10B secured. I have that on good authority." Sam Altman replied publicly: "wrong, as you surely know. want to come visit the first site already under way? this is great for the country. i realize what is great for the country isn't always what's optimal for your companies, but in your new role i hope you'll mostly put [America] first." The exchange was extensively covered by Axios, CNBC, and CBS News, and became one of the early flashpoints between Musk's role at the new Department of Government Efficiency and Trump's preferred AI partners [44][45][46].
In July 2025, Musk doubled down, saying the partners "simply don't" have the money for the $500 billion target. Arm CEO Rene Haas countered that the financial backing was "quite solid." The Musk-Altman feud over Stargate ultimately tracked the broader Musk-Altman rivalry that has shaped the post-2023 AI industry [47].
More measured skepticism came from financial press coverage. The Information and Bloomberg both noted that the announced equity at launch was significantly smaller than $100 billion, let alone $500 billion, and that headline numbers depended on debt financing, limited partners, and offtake agreements that had not yet closed. Bloomberg reported in August 2025 that no funds had been raised toward the larger $500 billion goal and that market and trade conditions were complicating fundraising [23][26].
Analysts also flagged that $500 billion over four years would represent a sustained capex pace comparable to the entire annual hyperscaler total, and that even Microsoft's record-setting $80 billion fiscal year 2025 spend was a single-company commitment for one year against existing infrastructure. Stargate's financing strategy has therefore relied heavily on parent-balance-sheet borrowing, vendor financing through Nvidia, AMD, and Broadcom (where AMD agreed to supply up to 6 GW of Instinct GPUs and Broadcom signed a 10 GW custom-hardware agreement), and chip-stake structures including a potential 10% OpenAI stake in AMD if milestones are achieved [48].
Environmental and infrastructure groups raised energy and water concerns, particularly in Texas. ERCOT projects that grid demand will roughly double from 85 GW in 2024 to as much as 218 GW by 2031, with data centers as the largest driver. The Texas Tribune and Texas Public Radio reported through 2025 that data centers were on track to consume 46 billion gallons of water in Texas in 2025, with one projection rising to 399 billion gallons (about 7% of state water use) by 2030 [49][50][51]. Lancium's Abilene site uses a closed-loop cooling design, which mitigates water draw, but other Texas Stargate sites have not committed to the same design.
The on-site natural gas turbines (about 360 MW) at Abilene, disclosed in public filings, drew criticism from clean-energy advocates who pointed out that the campus's marketed identity as a "clean campus" was contingent on grid wind and solar that ERCOT is racing to add. Local utility commissions and transmission operators across Texas, New Mexico, and Ohio have been working to fast-track interconnect approvals [52][53].
Local reception in Abilene has been broadly positive, with city officials emphasizing job creation and tax base. Across Texas more broadly, including the Milam County and Shackelford County sites, the picture has been more mixed: some county judges and water districts have raised resource concerns. In Lordstown, Ohio, the SoftBank site was framed as a partial answer to the post-General Motors industrial decline of the area. Reporting in 2026 by the Buenos Aires Herald and the Argentine business press indicated significant public debate around Stargate Argentina's RIGI tax treatment in Patagonia [38][54].
A quieter strand of skepticism has come from inside the AI research community itself. The two-year frontier model release cycle that Stargate is sized for assumes that the dominant return on more compute is bigger pretraining runs of large language models. If post-training, reasoning, and synthetic-data techniques close more of the capability gap than raw scale, as some Anthropic and DeepMind researchers have argued, the marginal value of a 10 GW dedicated buildout to a single tenant is harder to justify. OpenAI's own 2025 roadmap, which included GPT-5 and reasoning-oriented successors, was sized for the kind of compute that Stargate is meant to provide; the bet is therefore directional, with AI safety and competition concerns layered on top [55].
There is also a vendor-concentration concern. Stargate's first site is a tightly coupled stack: Crusoe-built buildings, Lancium-controlled land and energy, Oracle-leased operations, Nvidia GB200 hardware, and OpenAI as sole tenant. If any single vendor in that stack stumbles, including Nvidia given the AMD and Broadcom counter-commitments later in 2025, the others share in the disruption. The five-site September announcement, where Oracle leads three sites and SoftBank leads two, partially diversified that exposure but did not change the basic single-tenant logic.
Stargate's $500 billion target should be read against the simultaneous capex of the other large U.S. AI infrastructure builders, all of whom were ramping data center spend through 2025 and 2026.
| Program / company | Stated AI / data-center commitment | Period | Notes |
|---|---|---|---|
| Stargate Project (SoftBank/OpenAI/Oracle/MGX) | $100 billion immediate; up to $500 billion intended | 2025-2029 | Joint venture; not a single corporate balance sheet |
| Microsoft | $80 billion | Fiscal year 2025 | Largest single-year corporate technology spend ever announced |
| Microsoft | Indicated doubling of AI infrastructure | 2025-2027 | Per Next Platform (2026) on infrastructure ramp |
| Meta Platforms | ~$72 billion in 2025; "notably larger" in 2026 | 2025-2026 | Capex includes AI servers, networking, real estate |
| Alphabet (Google) | ~$75 billion | 2025 | Above prior analyst estimates |
| Amazon (AWS) | $100-105 billion | 2025 | Largest of the four hyperscaler 2025 capex budgets |
| Combined Microsoft + Google + Meta + Amazon | ~$725 billion (estimate) | 2026 | Per Statista / Tom's Hardware on hyperscaler capex projections |
The distinguishing feature of Stargate is that, unlike Microsoft's, Meta's, Google's, or Amazon's spend, it is a tenant-financed venture that builds for one customer, OpenAI, rather than for a multi-tenant cloud business. That asymmetry, OpenAI as the only paying customer for the resulting capacity, is part of why the financing is structured as a separate vehicle and why partner equity is paired with project debt and vendor financing rather than coming straight off Microsoft- or Oracle-style hyperscaler balance sheets [48][55].
As of May 2026, Stargate is partially built, partially under construction, and partially recontracted, with the gap between the headline $500 billion figure and disbursed capital still substantial. The Abilene flagship is operational at the first two buildings, with subsequent buildings under construction; the five September 2025 U.S. sites are at varying stages from groundbreaking (Lordstown) to early development; Stargate UAE is on track for a 200 MW first phase in the third quarter of 2026; Stargate Argentina remains a letter of intent; Stargate Norway and Stargate UK have been pulled, with Norway transferred to Microsoft. The OpenAI-Microsoft commercial reset of April 2026 ended Microsoft's exclusivity on OpenAI IP, which both freed Stargate compute to be shared more flexibly and rebalanced the broader Microsoft-OpenAI relationship.
No public reporting through this date confirms that the headline $500 billion has been raised. Bloomberg, The Information, and DCD have reported on continued financing rounds, vendor agreements, and partner exits without convergence on a single committed figure. The clearest concrete number is OpenAI's September 23, 2025 statement that planned capacity stood at "nearly 7 GW" with more than $400 billion of "planned" investment over three years, language that, like the original announcement, carefully kept the verb in the planning rather than committed tense [10][12][23].
What Stargate has clearly done, even where individual sites have changed hands, is anchor a wave of multi-gigawatt AI-only data centers in the United States and tie that buildout to the federal regulatory architecture that Trump's AI Action Plan and EO 14318 created. Whether the venture ultimately delivers $500 billion or considerably less, it has already shifted both OpenAI's compute relationship away from sole reliance on Microsoft Azure and the U.S. policy environment for AI infrastructure permitting.