A joint project with Oracle and SoftBank hasn’t moved forward, and OpenAI is seeking computing resources from partners.
OpenAI has encountered problems creating its own infrastructure for training AI models. According to The Information, Reuters, and Bloomberg, its flagship Stargate project has been effectively shelved, and plans to build a network of data centers have stalled.
The project was announced in early 2025 at the White House, attended by OpenAI CEO Sam Altman, SoftBank CEO Masayoshi Son, and Oracle founder Larry Ellison, with Donald Trump in attendance. It was billed as the largest AI infrastructure project in history, with up to $500 billion in investment and 10 GW of capacity.
More than a year later, construction has still not begun. Stargate lacks a full-fledged management team, no sites have been approved, and the partners have failed to agree on the division of responsibilities and funding. The project has reached a managerial impasse.
OpenAI’s attempt to build data centers on its own also failed. The company sought debt financing, but lenders refused to support a business with significant losses and an unsustainable profit model.
As a result, OpenAI has placed its bet on partnerships. In 2025, the company reached an agreement with Oracle to jointly develop approximately 4.5 GW of capacity with shared risks. At the same time, additional contracts were signed with Amazon Web Services and Google Cloud. To reduce its dependence on a single supplier, the company began collaborating not only with Nvidia but also with AMD and the startup Cerebras.
Despite these measures, by the end of 2025, OpenAI had secured only about 7.5 GW instead of the planned 10. The forecast for computing costs through 2030 increased from $450 billion to $665 billion.
CFO Sarah Friar stated that the company is focusing on a partnership model to avoid overburdening its balance sheet with capital expenditures. Owning its own data centers remains a long-term goal.
Competition is intensifying: according to The Wall Street Journal, Google DeepMind and Anthropic are actively developing their own infrastructure. To strengthen their control over this area, OpenAI has hired former Intel executive Sachin Kutty.
Stargate’s problems have also impacted its partners. Oracle, which signed a contract worth approximately $300 billion in 2025, faces mounting debt and the risk of a credit rating downgrade. SoftBank is discussing large-scale borrowing to cover future expenses. Nvidia’s position has added further uncertainty: its previously discussed participation in OpenAI financing worth approximately $100 billion has never materialized.
Against this backdrop, Elon Musk commented on the situation with the phrase “Hardware is hard,” emphasizing the difficulty of building AI infrastructure.
Ultimately, OpenAI’s strategy boils down to “control without ownership”: the company aims to secure priority access to capacity and participate in data center design, shifting the primary risks to partners. The plan to create its own global infrastructure remains on paper for now and depends on the company’s success in overcoming the management and financial crisis surrounding Stargate.
