Reduced Plans for Stargate Project: SoftBank and OpenAI Lower Expectations for $500 Billion AI Data Facility Construction
The Stargate Project, an ambitious initiative aimed at bolstering America's AI infrastructure, has taken a significant step forward with OpenAI and Oracle agreeing to develop an additional 4.5 gigawatts (GW) of AI data center capacity in the United States. This expansion brings the total planned Stargate infrastructure beyond 5 GW, capable of supporting over 2 million AI chips.
Originally conceived as a massive joint venture between SoftBank, OpenAI, and Oracle, the project was launched in January as a critical national investment to accelerate the U.S. lead in AI. However, by mid-2025, the project reportedly encountered significant setbacks and was scaled back from the planned 20 large data centers to a smaller initial build, mainly due to disagreements among partners on site locations and other key decisions.
Despite these challenges, the core mission of the Stargate Project remains unchanged: building world-class AI computing infrastructure in America. The data center will utilise Crusoe's technology and be powered by approximately 400,000 NVIDIA GB200 chips.
SoftBank, though initially a Stargate partner, is no longer financially involved in the current buildout portion. However, the company continues to pursue its broader AI strategy, with the "Project Crystal Land" being a notable example. This $1 trillion industrial complex aims to establish a high-tech manufacturing hub in Arizona, linked to both semiconductor industry partnerships and U.S. political support.
The first site of the Stargate Project will be an Ohio data center, operational by year-end. Other potential sites for the project include Texas, Michigan, Wisconsin, Wyoming, New Mexico, Georgia, and Pennsylvania. Notably, Oracle is building a 1.2 GW Stargate data center in Abilene, Texas.
As the Stargate Project moves forward, the partners must rebuild trust and secure operational agreements to move beyond Ohio. OpenAI continues to deepen ties with key technology partners including Microsoft, NVIDIA, Cisco, and UAE's G42. The company has also signed additional capacity contracts with CoreWeave, bringing its total data center capacity closer to 5 GW for 2025.
In a separate development, OpenAI independently struck a $30 billion annual infrastructure deal with Oracle to rent 4.5 GW of data center capacity. This deal, combined with the ongoing expansion of the Stargate Project, positions the U.S. well in the global AI race against China.
As the Stargate Project progresses, it remains a key element in advancing U.S. AI infrastructure and strengthening the country's position in the global AI race against China. However, the practical execution challenges and evolving partner roles, especially regarding SoftBank, are factors to consider moving forward.
[1] The Information [2] Bloomberg [3] TechCrunch
- The ongoing expansion of the Stargate Project, driven by OpenAI and Oracle, is set to bolster the AI infrastructure in the United States, with an additional 4.5 gigawatts (GW) of AI data center capacity planned.
- While SoftBank is no longer financially involved in the current buildout portion of the Stargate Project, it continues to pursue its broader AI strategy through projects like "Project Crystal Land," a $1 trillion industrial complex in Arizona.
- OpenAI has signed additional capacity contracts with CoreWeave, bringing its total data center capacity closer to the Stargate Project's planned 5 GW for 2025. Apart from this, OpenAI has independently struck a $30 billion annual infrastructure deal with Oracle to rent 4.5 GW of data center capacity.
- As the Stargate Project advances, it will leverage Crusoe's technology and NVIDIA GB200 chips to construct world-class AI computing infrastructure in America and compete with China on the global AI stage. Potential sites for the project include Texas, Michigan, Wisconsin, Wyoming, New Mexico, Georgia, and Pennsylvania.
