“Stargate” is the codename for a multi-phased, five-year infrastructure plan by Microsoft and OpenAI to build the world’s largest AI supercomputing clusters. Originally estimated at $100 billion, the project expanded in January 2026 into a broader global alliance including SoftBank and Oracle, targeting up to $500 billion in AI-ready data center investment. The centerpiece is a planned 5 GW supercomputer in the United States, likely consisting of multiple interconnected campuses.

The project is structured in five phases:

Phases 1 & 2: Smaller training clusters (operational).
Phase 3: Current mid-scale clusters utilizing NVIDIA H100 and Blackwell GPUs.
Phase 4: A massive $10 billion “supercluster” slated for 2026.
Phase 5 (Stargate): The 5 GW flagship facility, scheduled for 2028-2030, designed to house millions of next-generation accelerators (Vera Rubin and beyond).

To solve the “power bottleneck,” Stargate is increasingly pivoting toward nuclear energy, including SMRs (Small Modular Reactors) and agreements with Constellation and Helion. The initiative serves as the primary infrastructure engine for OpenAI’s pursuit of Artificial General Intelligence (AGI), with additional “Stargate” hubs now planned for the UAE and East Asia.

Stargate AI Supercomputer

Sources