On October 29, 2025, NVIDIA made a significant announcement, revealing its plan to construct seven AI supercomputers for the U.S. government. Among these, five will be built around the currently available Blackwell GPU, while the remaining two will harness the power of the next-generation Rubin GPU paired with the Vera CPU. These supercomputers are destined to serve the Los Alamos National Laboratory, which operates under the auspices of the U.S. Department of Energy.

The initial two supercomputers will be erected by Oracle specifically for the Argonne National Laboratory. The first of these, named 'Equinox,' will be outfitted with an impressive array of 10,000 Blackwell GPUs and is projected to come online in 2026. The second, dubbed 'Solstice,' will take things to a whole new level, boasting over 100,000 Blackwell GPUs and consuming a staggering 200 megawatts of power. When combined, these two supercomputers will deliver a jaw-dropping combined FP4 computing power of 2,200 EFlops. This immense computational capacity will enable them to effectively support AI large models with a staggering 3 trillion parameters.

Turning our attention to the remaining five supercomputers, 'Tara,' 'Minerva,' and 'Janus' will also be based on the Blackwell GPU and will find their home at the Argonne National Laboratory. On the other hand, 'Mission' and 'Vision' will make the leap to the Rubin GPU and will be deployed at the Los Alamos National Laboratory. In particular, 'Mission' has been specifically designed with the U.S. National Nuclear Security Administration in mind. It is expected to go live in 2027, with the crucial mission of ensuring the safety of the nation's nuclear stockpile. Meanwhile, 'Vision' represents an upgrade based on the early Venado supercomputer and will be dedicated to serving open science and AI research endeavors.