The $500 billion &#34;Stargate&#34; AI data center plan announced by Trump at the White House has been stalled for over a year since its launch, forcing OpenAI to abandon its ambition to build its own data centers and instead rely on partnerships with cloud service providers to meet its urgent computing power needs.
According to informed sources, the joint venture formed by OpenAI, Oracle, and SoftBank has yet to be staffed or develop any OpenAI data centers. Within weeks of the project's announcement, the three parties fell into disagreements over key issues such as leadership, division of responsibilities, and cooperation structure, resulting in the &#34;Stargate&#34; plan being unable to advance.
Faced with a shortage of computing power, OpenAI once planned to build and own data center campuses on a large scale to reduce reliance on cloud service providers. However, when lenders refused to provide billions of dollars in financing for this unproven business model that burns billions of dollars annually, the plan was ultimately shelved. It wasn't until months later, confronted with financial realities, that OpenAI turned back to its &#34;Stargate&#34; partners to seek to advance the project.
This setback has put pressure on OpenAI's financial situation. Last year, the company's gross margin was below expectations as it had to purchase expensive computing power at the last minute. OpenAI also raised its forecast for computing power spending before 2030 from $450 billion to $665 billion. Meanwhile, Google and Anthropic have grown into more threatening competitors over the past year.
<h2>Disagreements Among Three Parties Lead to Project Stagnation</h2>
When President Trump announced the &#34;Stargate&#34; project in January 2025, the three companies responsible for developing the project promised to move forward at lightning speed, almost immediately investing $100 billion to begin construction of 10 gigawatts of computing capacity.
However, according to three informed sources involved in the plan, within weeks of the project's announcement, &#34;Stargate&#34; fell into trouble, lacking leadership and coordination. AI model developer OpenAI, cloud service provider Oracle, and Japanese conglomerate SoftBank could not reach consensus on who was responsible for what and how the collaboration should be structured.
Some executives had proposed splitting &#34;Stargate&#34; into an independent entity that would build the facilities and lease them back to OpenAI, with some OpenAI employees potentially transferring to that entity. Executives also discussed whether &#34;Stargate&#34; could serve as a financing tool to raise capital for chips and infrastructure. However, these ideas ultimately did not materialize.
Last year, OpenAI failed to achieve its established goal of securing a commitment for 10 gigawatts of capacity through SoftBank and Oracle over the next three years. During the finalization of the &#34;Stargate&#34; deal, OpenAI turned to other suppliers such as Amazon Web Services and Google Cloud to sign computing power agreements to meet the required computing capacity.
<h2>Self-Build Plan Encounters Financial Reality</h2>
After the White House announced &#34;Stargate,&#34; OpenAI's senior leadership began exploring how to build data centers on its own. Company employees traveled across the country to find potential sites that could support 800 megawatts to 1.2 gigawatts campuses, prioritizing locations that could provide ample power in 2026 and 2027
But when OpenAI assessed the site and calculated costs, executives realized that the company would have to pay a high price to secure the necessary financing. If a tenant with better credit, such as Oracle, signed the lease and then subleased the capacity to OpenAI, the costs would be significantly reduced—this is a model OpenAI has used in other data center deals.
It wasn't until spring that OpenAI's &#34;Stargate&#34; team held meetings with Oracle's executive team, including Clay Magouyrk, who was then the head of Oracle's cloud business and is now the company's co-CEO. The two sides finalized a new vision: Oracle would directly enter into a major data center deal with OpenAI for 4.5 gigawatts of power.
According to attendees, this broader framework allowed the two companies to spread the risk across multiple campuses rather than signing projects one by one. Within about two months, OpenAI paused its plans to build its own data centers. The campuses OpenAI initially considered, such as the Vantage Data Centers project in Wisconsin, were assigned to Oracle for development.
<h2>A Compromise of Control Rather Than Ownership</h2>
At the end of July 2024, Oracle and OpenAI announced a deal to develop 4.5 gigawatts of data center capacity at multiple locations in the United States. According to two insiders, the two sides agreed to share some of the economic risks of this large project in this rare arrangement. This means that if there are delays or cost overruns, both parties will share the costs; if the project costs less than expected, both will also share the savings.
After Oracle's announcement, OpenAI employees began preparing to revive their ambitions for building their own data centers. One project was particularly important in their planning: a 1-gigawatt campus located between Milam County, Texas, Austin, and Waco.
At the beginning of 2025, OpenAI had developed plans for the project, but negotiations with Oracle led to the project being shelved, construction halted, and OpenAI even lost its general contractor due to the delays. Ultimately, the company did not undertake the project itself but instead established a partnership with another &#34;Stargate&#34; partner, SoftBank, and publicly stated that it would jointly develop the site with SoftBank's subsidiary, SoftBank Energy.
However, according to two individuals involved in the discussions, there was a conflict over control issues. 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 &#34;Stargate&#34; team made multiple trips to Japan to resolve issues with SoftBank's Masayoshi Son. During one meeting, the team met in a large conference room one floor below Son's main office, with negotiations lasting for hours. OpenAI employees repeatedly went to a nearby 7-Eleven to buy cola and milk tea, and by the end of the meeting, the table was filled with empty bottles.
Ultimately, the two sides reached a compromise: OpenAI signed a long-term lease and controlled the facility's design, while SoftBank Energy was responsible for developing and owning the project. Within OpenAI, this moment marked that its immediate ambition to build its own data centers was no longer a near-term goal.
The two companies broke ground in October. Although this is not OpenAI's first self-built data center, employees breathed a sigh of relief as they completed a deal that had been in preparation for nearly a year. OpenAI hopes that this project can serve as a template for expanding more capacity. The agreement grants the company design control without requiring the startup to directly fund or own the project, with control over decisions including cluster architecture, cooling systems, rack configurations, and power infrastructure.
<h2>New Leadership Restructures Infrastructure Team</h2>
OpenAI's Chief Financial Officer Sarah Friar's remarks at last month's World Economic Forum in Davos reflected this shift. In an interview with The Information's editor-in-chief Jessica Lessin, she stated that OpenAI is &#34;leveraging our (cloud) partners because it's a way to keep the balance sheet light.&#34;
&#34;We're not fully self-built today, because frankly, we have great partners,&#34; she said, &#34;but you know, where will we be three years from now?&#34;
According to two OpenAI employees, the company still intends to build its own data centers, but this is no longer a near-term priority. Instead, the company is building a series of cloud and infrastructure agreements under the banner of &#34;Stargate,&#34; some of which grant it special permissions for data center design.
The evolution of &#34;Stargate&#34; and OpenAI's realization that it cannot temporarily build and own its own infrastructure coincided with high-level changes in its infrastructure organization. In November of last year, OpenAI hired former Intel Chief Technology and AI Officer Sachin Katti to lead the infrastructure team, reporting to OpenAI President and Co-founder Greg Brockman

ORCL

According to informed sources, the joint venture formed by OpenAI, Oracle, and SoftBank has yet to staff any personnel or develop any OpenAI data centers. Within weeks of the project's announcement, the three parties fell into disagreements over key issues such as leadership, division of responsibilities, and cooperation structure, causing the "Stargate" plan to be unable to advance

- The "Stargate" AI data center project, announced by Trump with a $500 billion budget, has been stalled for over a year, forcing OpenAI to rely on cloud partnerships for computing needs.  
- Disagreements among OpenAI, Oracle, and SoftBank regarding leadership and responsibilities hindered progress, and financial limitations ultimately sank OpenAI's plans for self-built data centers.  
- OpenAI's CFO stated the company is now prioritizing cloud partnerships over immediate self-building ambitions, modifying its strategy amid organizational changes.

A year after the White House press conference, OpenAI's $500 billion "Stargate" plan remains stalled