Sam Altman stood beside President Donald Trump at the White House in January 2025. SoftBank chief Masayoshi Son and Oracle co-founder Larry Ellison flanked them. The occasion marked the formal launch of Stargate. This joint venture promised to pour as much as $500 billion into U.S. AI infrastructure over four years. The goal was clear. Secure American leadership in artificial intelligence. Create hundreds of thousands of jobs. Deliver the compute muscle needed for the next wave of breakthroughs.
One year later the picture looks more complicated. Stargate has made tangible progress. Yet it has also encountered delays, partner frictions and sobering questions about financing, power supply and execution. The project that began as a bold declaration now reveals the gritty realities of building at unprecedented scale.
Stargate started life as an internal code name. The Information first reported in March 2024 that Microsoft and OpenAI were planning a $100 billion data-center project featuring an AI supercomputer dubbed Stargate. That early concept called for launch around 2028 as the fifth phase in a multi-year roadmap. The 2025 announcement expanded the vision dramatically. It created Stargate LLC, a new entity with OpenAI and SoftBank each holding 40 percent stakes. Oracle and Abu Dhabi’s MGX took the remainder.
By September 2025 the venture had accelerated. OpenAI, Oracle and SoftBank announced five new U.S. AI data-center sites. The combined pipeline, including the flagship in Abilene, Texas, and work with CoreWeave, reached nearly 7 gigawatts of planned capacity. Investment topped $400 billion over the next three years. Officials declared the full $500 billion and 10-gigawatt target within reach ahead of the original schedule. “AI can only fulfill its promise if we build the compute to power it,” Altman said in the OpenAI announcement. “That compute is the key to ensuring everyone can benefit from AI and to unlocking future breakthroughs.”
The Abilene campus opened first. Located 180 miles west of Dallas, it began delivering NVIDIA GB200 racks in June 2025. One building was operational. Another neared completion. The site sits on Oracle Cloud Infrastructure. It already trains and serves frontier models. Oracle and OpenAI later agreed to develop up to 4.5 additional gigawatts. That pact alone exceeds $300 billion over five years. Masayoshi Son described the effort as harnessing SoftBank’s data-center design and energy expertise. “Together with OpenAI, Arm, and our Stargate partners, we are paving the way for a new era where AI advances humanity,” he stated.
Additional locations followed. Oracle added sites in Shackelford County, Texas, and Doña Ana County, New Mexico. A Midwest campus landed in Wisconsin through a partnership with Vantage Data Centers. SoftBank and OpenAI broke ground in Lordstown, Ohio, with operations slated for 2026. Another fast-build facility appeared in Milam County, Texas, backed by SB Energy. These campuses carry ambitious targets. Some aim to scale beyond 1 gigawatt each. A single gigawatt supplies electricity to roughly 750,000 American homes. The aggregate draw will test grid operators and local utilities in multiple states.
Yet momentum has not been uniform. Reports surfaced of internal stalls. The Information revealed in February 2026 that more than a year after launch the Stargate joint venture had not fully staffed up and was not developing any of OpenAI’s data centers. Three people familiar with the matter described OpenAI scrambling for computing power elsewhere. Negotiations with partners dragged. Location decisions sparked disagreements. SoftBank’s chief financial officer Yoshimitsu Goto cited protracted talks as a factor slowing progress.
Power constraints and supply-chain bottlenecks compound the pressure. Nearly half the AI data centers expected to come online in the United States in 2026 now face delays or cancellations, according to industry trackers. Trade tensions with China have tightened access to critical materials and electrical equipment. GPU shortages persist. Financing the chips themselves grows harder because their useful life remains uncertain in a rapidly changing field. OpenAI has signaled plans to tap debt markets for future builds, but it has not drawn on that option yet.
Microsoft’s role adds another layer. The company maintains its long-standing partnership with OpenAI. It continues to provide Azure capacity and is building its own advanced AI data centers, including a $3.3 billion campus in Wisconsin expected online in early 2026. In one Texas location Microsoft stepped in to expand a data-center project after OpenAI stepped back. The two organizations now sit as neighbors at a massive complex in Abilene. A new agreement grants Microsoft right of first refusal on future OpenAI cloud computing needs while permitting OpenAI to seek capacity from others if Microsoft declines. The arrangement reflects both cooperation and growing independence.
Skeptics have voiced louder doubts. Some analysts question the circular nature of the funding. OpenAI generates revenue but its capital appetite far outstrips current earnings. A company with roughly $20 billion in annual revenue cannot easily shoulder trillions in cumulative infrastructure commitments. Robert Bryce outlined four reasons for skepticism in January 2025, highlighting the staggering sums and the reliance on massive tech balance sheets. Microsoft CEO Satya Nadella, when asked about funding, replied that he was good for his $80 billion portion. Such statements underscore the concentration of financial firepower among a handful of players.
Local pushback has emerged too. Grassroots opposition to data centers has grown in certain communities over water use, noise and land consumption. OpenAI has responded with commitments to be good neighbors. In Abilene the water draw for the Stargate site equals half what the city uses in a single day, according to local officials. Similar efficiency designs apply across other campuses. In Wisconsin partners pledged at least $175 million in infrastructure upgrades and water restoration. OpenAI also plans to open its first Stargate community academy in Abilene in spring 2026 and to engage with labor unions.
Construction activity remains impressive despite the headwinds. Oracle reported more than 6,000 daily construction jobs at Abilene at peak and 1,700 long-term positions. The venture overall aims to create more than 25,000 onsite roles and tens of thousands more across the supply chain. These numbers matter in towns hungry for economic revival. But delivery depends on overcoming persistent obstacles.
By mid-2026 the project had shifted shape in meaningful ways. Early visions of a single monolithic supercomputer gave way to a distributed network of massive but geographically dispersed campuses. OpenAI broadened its supplier list. It deepened ties with Oracle while preserving and sometimes reducing direct dependence on Microsoft for certain builds. The joint venture itself appears less central to day-to-day execution than originally conceived. Some sites have advanced under bilateral deals rather than unified Stargate governance.
And the timeline has stretched in places. Initial promises of immediate $100 billion deployment met slower partner alignment. A planned U.K. facility was paused over power costs and regulatory hurdles. An Abu Dhabi-linked effort faced external security risks. These setbacks do not cancel the overall thrust. They do illustrate that even the best-resourced technology effort collides with physical limits. Electricity generation cannot scale overnight. Transformer lead times run long. Skilled labor stays tight.
OpenAI executives maintain optimism. The company has already put substantial capacity to work. Frontier models are training on Stargate hardware today. Inference workloads run at scale. Altman and his team argue that the only path to meaningful AI progress runs through this kind of investment. Without the compute, the promise remains theoretical. With it, new applications in science, medicine and industry become feasible.
Still, the market is beginning to demand proof. Revenue growth must justify the capital outlay. Technological leaps must materialize on schedule. Investors and partners watch closely as 2026 unfolds. Some forecasts already predict a scaling back or stretching of AI data-center plans this year as economic reality bites. The question is no longer whether the industry will build enormous facilities. It is whether the build can happen fast enough, cheaply enough and sustainably enough to deliver returns before the hype cycle cools.
Stargate stands as the clearest test case. Its successes will be measured in gigawatts delivered, models trained and economic activity generated. Its shortfalls will be tallied in delayed sites, renegotiated contracts and unmet expectations. For now the project continues. New racks arrive in Texas. Ground breaks in Ohio. Planning advances in Wisconsin and New Mexico. The stakes could hardly be higher. The outcome will influence not only OpenAI’s competitive position but the broader balance of technological power between the United States and its rivals.
So far the venture has moved faster than many critics expected on paper. It has also revealed frictions few anticipated in practice. The coming months will determine whether Stargate becomes the foundation for a new era of American AI dominance or a cautionary tale about the difficulty of turning ambition into infrastructure at planetary scale.
WebProNews is an iEntry Publication