Nvidia to Build Seven AI Supercomputers for U.S. Government

At the GTC developer conference in Washington, D.C., Jensen Huang announced that Nvidia will build seven new AI supercomputers for the U.S. Department of Energy (DOE). These systems, developed in partnership with Oracle, aim to accelerate scientific discovery and support the maintenance of the U.S. nuclear weapons arsenal and research into nuclear fusion. The largest project, the Solstice system at Argonne National Laboratory, will feature 100,000 Blackwell GPUs. A smaller system, Equinox, will also be deployed at Argonne, with delivery expected in 2026.

Simultaneously, Los Alamos National Laboratory selected Hewlett Packard Enterprise and Nvidia to develop two additional systems, Mission and Vision. Mission will operate in a classified environment to support national nuclear security modeling, while Vision will handle unclassified research. These systems will utilize the HPE Cray GX5000 architecture and the newly unveiled Vera Rubin platform, which includes Vera CPUs and Rubin GPUs. Huang stated the Rubin platform is expected to enter production by late 2026.

Huang credited President Donald Trump's policies for promoting domestic manufacturing, announcing that Blackwell AI GPUs are now being produced in Arizona. He committed to an all-American assembly line for the DOE projects and reported $500 billion in bookings for Blackwell and Rubin chips over the next five quarters. Beyond government contracts, Nvidia announced a $1 billion investment in Nokia for 6G efficiency, a logistics partnership with Palantir, and a Robotaxi collaboration with Uber via the Hyperion platform.