The Stargate AI data center project, announced by President Trump in January 2025 with a budget of $500 billion, was initially promised by the three responsible companies to advance at lightning speed—with an immediate investment of $100 billion to build new data centers with a total computing capacity of 10 gigawatts. Over a year later, according to three individuals involved in the stalled initiative, the Stargate joint venture has yet to assemble a team or construct any data centers for OpenAI. These sources indicated that the project stalled within weeks of its announcement due to a lack of leadership and coordination mechanisms. The three parties behind Stargate—AI model developer OpenAI, cloud service provider Oracle, and Japanese conglomerate SoftBank—were deeply divided over roles, responsibilities, and the collaborative framework.
Consequently, OpenAI, which is in urgent need of computing power, was forced to seek solutions independently. The company at one point prepared to undertake most of the construction work itself, planning to directly lease or even own large data center campuses in certain projects. This strategy aimed to reduce reliance on other cloud providers and strengthen control over computing resources. However, this plan ultimately stalled as lending institutions were unwilling to provide billions of dollars in project financing to a company with an unproven business model that burns through tens of billions of dollars annually.
It was only months later, after OpenAI and its CEO Sam Altman confronted this financial reality, that they returned to advance the project with their Stargate partners. Even then, the collaboration primarily took the form of bilateral agreements between OpenAI and SoftBank, and between OpenAI and Oracle, rather than a coordinated three-party effort.
Despite the rocky start, data centers under the Stargate banner have finally begun to materialize. OpenAI has moved away from aggressively building and operating its own facilities through debt-fueled expansion. Instead, it now heavily relies on partnerships with cloud providers, adopting a model that enhances control over computing power without requiring significant capital investment.
This strategic shift was signaled last month at the World Economic Forum in Davos by OpenAI Chief Financial Officer Sarah Friar. In an interview with The Information's Editor-in-Chief Jessica Lessin, Friar stated that OpenAI is "leveraging cloud partners to maintain a lighter balance sheet." She added, "We don't currently have plans for full-scale self-building; frankly, we have excellent partners. But what the situation will be in three years, no one can say for sure."
According to two OpenAI employees, the company still harbors ambitions to build its own data centers, but this is no longer a near-term priority. Instead, it is signing a series of cloud service and infrastructure agreements under the "Stargate" banner, some of which grant OpenAI special privileges in data center design. Spokespeople for OpenAI, SoftBank, and Oracle all declined to comment.
It is difficult to determine if the delays in Stargate's progress have slowed down OpenAI's model development. However, over the past year, competitors like Google and Anthropic have become more significant threats. Last year, OpenAI failed to meet its goal of securing a commitment for 10 gigawatts of computing power over the next three years through SoftBank and Oracle. As Stargate negotiations dragged on, OpenAI also signed agreements with other suppliers like Amazon Web Services and Google Cloud to acquire much-needed computing capacity. The company additionally announced plans to use non-Nvidia chips, forming agreements with Advanced Micro Devices (AMD) and chip startup Cerebras to diversify its computing power supply.
OpenAI's computing power challenges have also impacted its financial performance. Last year, due to temporary high-cost procurement of computing resources, its gross margin fell below expectations. While the company had previously projected spending $450 billion on computing power by 2030, it has recently revised this forecast upward to $665 billion.
Following the White House's announcement of the Stargate project, OpenAI's leadership began planning a path to building its own data centers. Employees scouted locations across the United States for sites capable of supporting single-campus capacities of 800 megawatts to 1.2 gigawatts, prioritizing areas that could provide large-scale power by 2026-2027. However, the operational structure of the Stargate joint venture and the specific roles of Oracle and SoftBank remained unclear.
According to individuals involved in the planning, some executives proposed spinning off Stargate as an independent entity responsible for building facilities and then leasing them back to OpenAI, with some employees transferring to Stargate. Management also discussed whether Stargate could serve as a financing vehicle for raising capital for chips and infrastructure.
These ideas never materialized. After site selection and financial analysis, OpenAI realized the cost of self-financing was prohibitively high. It determined that costs would be significantly lower if a company with stronger credit, like Oracle, signed the lease and subleased to OpenAI—a model it employs in other data center partnerships.
It wasn't until that spring that the OpenAI Stargate team met with Oracle executives, including then-cloud chief and current co-CEO Clay Magouyrk. The parties agreed on a new plan: Oracle would enter into a major data center agreement directly with OpenAI, with a total scale of 4.5 gigawatts. According to a participant, this overarching structure distributed risk across multiple campuses instead of signing individual project agreements.
Roughly two months later, OpenAI paused its plan to build its own data centers. Previously secured sites, such as the Vantage data center project in Wisconsin, were incorporated into the new agreement and assigned to Oracle. Finally, at the end of July, Oracle and OpenAI announced a collaboration to build 4.5 gigawatts of data center capacity across multiple locations in the United States.
The two companies reached an unusual arrangement, according to two informed sources: they agreed to share the economic risk of the large-scale project. This means costs for delays or overruns would be split, and savings from coming in under budget would also be shared—details not previously disclosed publicly.
After the Oracle partnership was announced, OpenAI employees resumed preparations for building their own data centers, restarting negotiations with equipment suppliers to accelerate new projects. One key site was a 1-gigawatt campus in Milam County, Texas, located between Austin and Waco. During a visit in early 2025, OpenAI employee Keith Heid jokingly told colleagues he could determine clay content by tasting the soil, proceeding to kneel down, pick up some dirt, and put it in his mouth—an incident that became an internal anecdote illustrating the team's extreme dedication to site selection. (Heid left OpenAI several weeks ago.)
OpenAI had drafted plans for the Texas project earlier that year, but negotiations with Oracle caused the project to halt, construction to stop, and even led to the loss of the general contractor. Despite this, OpenAI still opted not to develop the site itself. Instead, it partnered with the other Stargate partner, SoftBank, and publicly announced it would co-develop the campus with SoftBank's subsidiary, SoftBank Energy.
However, according to two people involved in the talks, a conflict over control arose: OpenAI wanted the Texas campus to be its first self-built data center, while SoftBank wanted to develop and own the project. Between September and October, the Stargate team made several trips to Japan to negotiate with SoftBank founder Masayoshi Son. One meeting, held in a large conference room below Son's main office, lasted for hours, with OpenAI employees frequently running to a nearby 7-Eleven for Coke and milk tea, leaving the table littered with empty bottles by the end.
A compromise was eventually reached: OpenAI signed a long-term lease and gained control over the facility's design, while SoftBank Energy was responsible for development and ownership. Internally at OpenAI, this moment signaled that the ambition of building its own data centers was no longer a near-term goal. Construction began in October. Although it wasn't OpenAI's first truly self-built data center, employees were relieved that the deal, nearly a year in the making, was finally proceeding.
OpenAI hopes this project will serve as a template for future computing capacity expansion: the agreement grants design control—including decisions on cluster architecture, cooling systems, cabinet configuration, and power infrastructure—without requiring full capital investment or project ownership.
The evolution of the Stargate project and the pragmatic acceptance of "not building and operating infrastructure for now" coincided with leadership changes in OpenAI's infrastructure department. In November, the company hired former Intel CTO and AI lead Sachin Katti to head the infrastructure team, reporting to President and co-founder Greg Brockman. A key task for Katti is helping OpenAI develop data center-related intellectual property to ensure infrastructure agreements precisely meet the company's needs. He is also responsible for the overall computing power roadmap, including decisions on which chips OpenAI will use. Several executives who previously reported to Brockman—including Data Center Lead Chris Malon and VP of Industrial Compute Peter Hoschler—now report to Katti. According to informed sources, Katti is also reorganizing other teams within the computing power department.