NVIDIA will use 100,000 Blackwell GPUs to build „largest supercomputer” for U.S. government

NVIDIA is involved in another project focused on the development of artificial intelligence. This time, it’s about building several supercomputers, the project being carried out in collaboration with the United States government. The scale of the project is expected to be massive, as NVIDIA plans to use over 100,000 GPUs based on the Blackwell architecture, with the majority allocated to just two computers.

NVIDIA builds supercomputers with the US government

The announcement, published by NVIDIA on Tuesday, detailed a collaboration with Oracle and the United States Department of Energy aimed at building the world’s largest AI supercomputer. The project will include seven supercomputers in total, with five scheduled to go online in the first half of 2026.

The system, named Equinox, will be built using a total of 10,000 NVIDIA Blackwell GPUs. Significantly more resources will be allocated to building the remaining two supercomputers. Together, the two supercomputers, named Solstice, will harness the power of 100,000 NVIDIA Blackwell GPUs. According to the company, their mission is to “drive the technological leadership of the United States in security, science, and energy.”

NVIDIA CEO Jensen Huang commented on the ambitious project:

This time, the project is not expected to impact residents of smaller American towns. In recent months, more companies have been building data centers, a trend that is increasingly facing opposition from local residents. This time, both Equinox and Solstice will be constructed at the same site: Argonne National Laboratory in Illinois.

• Nvidia
• tech
• AI / Artificial Intelligence