One year after the White House briefing, the launch has still not been initiated, and OpenAI's $500 billion "Interstellar Gate" plan has stalled.

The $500 billion “Stargate” AI data center plan announced prominently by Trump at the White House has stalled more than a year after its launch. OpenAI has been forced to abandon its ambition of building its own data centers and instead relies on partnerships with cloud service providers to meet its urgent computing needs.

Sources familiar with the matter reveal that the joint venture formed by OpenAI, Oracle, and SoftBank has yet to staff up or develop any OpenAI-specific data centers. Within weeks of the project’s announcement, the three parties fell into disagreements over leadership, responsibilities, and collaboration structure, causing the “Stargate” plan to remain stalled.

Faced with a shortage of computing power, OpenAI initially planned to massively build and own data center campuses to reduce dependence on cloud providers. However, when lenders refused to provide hundreds of millions of dollars in financing for a company with an unproven business model burning billions annually, that plan was ultimately shelved. It wasn’t until months later, confronting financial realities, that OpenAI turned back to its “Stargate” partners to push the project forward.

This setback has put financial pressure on OpenAI. Last year, the company’s gross margin fell below expectations as it had to purchase expensive compute capacity at the last minute. OpenAI also raised its forecast for compute spending before 2030 from $450 billion to $665 billion. Meanwhile, Google and Anthropic have grown into more formidable competitors over the past year.

Disputes Among Partners Stall the Project

When President Trump announced the “Stargate” project in January 2025, the three companies responsible for its development pledged to move at lightning speed, nearly immediately committing $100 billion to build 10 gigawatts of computing capacity.

However, according to three insiders involved in the plan, within weeks of the announcement, “Stargate” ran into trouble due to lack of leadership and coordination. OpenAI, Oracle, and SoftBank struggled to agree on responsibilities and the collaboration framework, especially over who would lead and how to structure the partnership.

Some executives proposed splitting “Stargate” into an independent entity that would build the facilities and lease them back to OpenAI, with some OpenAI staff potentially transferring to that entity. Others discussed whether “Stargate” could serve as a financing tool to raise capital for chips and infrastructure. None of these ideas materialized.

Last year, OpenAI failed to meet its goal of securing commitments for 10 gigawatts of capacity from SoftBank and Oracle over the next three years. During the finalization of the “Stargate” deal, OpenAI instead signed compute agreements with other providers like Amazon Web Services and Google Cloud to meet its capacity needs.

Self-Build Plans Hit Financial Realities

After the White House announced “Stargate,” OpenAI’s senior leadership began exploring how to build its own data centers. Employees traveled nationwide to identify potential sites capable of supporting campuses from 800 megawatts to 1.2 gigawatts, prioritizing locations with ample power supply in 2026 and 2027.

But as OpenAI evaluated sites and calculated costs, executives realized the company would need to pay hefty premiums to secure the necessary financing. If a more creditworthy tenant like Oracle signed a lease and then subleased capacity to OpenAI, costs could be significantly reduced—this is a model OpenAI has used in other data center deals.

It wasn’t until spring that OpenAI’s “Stargate” team held meetings with Oracle executives, including then-Oracle Cloud head and current co-CEO Clay Magouyrk. The two sides agreed on a new vision: Oracle would directly strike a major deal with OpenAI for 4.5 gigawatts of data center capacity.

Sources say this broader architecture allows both companies to spread risk across multiple campuses rather than signing individual projects. Within about two months, OpenAI paused its self-build plans. The first campuses considered, such as Vantage Data Centers in Wisconsin, were assigned to Oracle for development.

Compromise on Control Rather Than Ownership

In late July 2024, Oracle and OpenAI announced a deal to develop 4.5 gigawatts of data center capacity across multiple U.S. sites. According to two insiders, both sides agreed to share some of the financial risks associated with this large-scale project. This means that if there are delays or cost overruns, both will share the burden; if costs come in under budget, both will benefit from the savings.

Following Oracle’s announcement, OpenAI staff began preparing to revive its self-build ambitions. A particularly important project was a 1-gigawatt campus located between Austin and Waco, Texas.

In early 2025, OpenAI had drafted plans for this project, but negotiations with Oracle led to delays and suspension of construction. The company even lost its general contractor due to the delays. Ultimately, OpenAI did not proceed alone but instead partnered with another “Stargate” partner, SoftBank, and publicly announced plans to develop the site jointly with SoftBank Energy.

However, two people involved in the discussions say conflicts arose over control. OpenAI wanted the Texas campus to be its first self-built data center, while SoftBank preferred to develop and own the project.

Between September and October, the “Stargate” team made multiple trips to Japan to meet with SoftBank’s Masayoshi Son. In one lengthy meeting in a large conference room beneath Son’s main office, negotiations lasted several hours. OpenAI staff repeatedly went to nearby 7-Eleven stores to buy sodas and milk teas, filling the table with empty bottles at the end.

Ultimately, both sides reached a compromise: OpenAI would sign a long-term lease and control the facility design, while SoftBank Energy would develop and own the project. Internally, this marked a shift—OpenAI’s immediate goal of building its own data center was no longer a near-term priority.

The two companies broke ground in October. While this wasn’t OpenAI’s first self-built data center, employees felt relief having completed a nearly year-long negotiation. OpenAI hopes this project will serve as a template for expanding capacity. The agreement grants the company design control over aspects like cluster architecture, cooling systems, rack configurations, and power infrastructure, without requiring the startup to directly fund or own the project.

Leadership Restructures Infrastructure Team

OpenAI CFO Sarah Friar signaled this shift last month at the Davos World Economic Forum. In an interview with The Information editor Jessica Lessin, she said OpenAI is “leveraging our (cloud) partners because that’s the way to keep our balance sheet lean.”

“We’re not fully self-building yet, because frankly, we have good partners,” she said. “But where will we be in three years?”

Two OpenAI employees say the company still plans to build its own data centers, but this is no longer a near-term priority. Instead, the company is leveraging the “Stargate” initiative to establish a series of cloud and infrastructure agreements, some of which grant it special rights over data center design.

The evolution of “Stargate” and OpenAI’s realization that it cannot build and own its own infrastructure for now coincided with a leadership change in its infrastructure organization. In November last year, OpenAI hired former Intel CTO and AI executive Sachin Katti to lead the infrastructure team, reporting to CEO and co-founder Greg Brockman.

Risk Warning and Disclaimer

Market risks are inherent; invest cautiously. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should evaluate whether any opinions, viewpoints, or conclusions herein are suitable for their circumstances. Invest at your own risk.