NVIDIA has revealed its plan to collaborate with partners in the construction of seven AI supercomputers for the U.S. government. Each of these supercomputers will boast exascale (EFlops) - level computing capabilities. These cutting - edge machines are set to be utilized by the Argonne National Laboratory and the Los Alamos National Laboratory, both of which fall under the jurisdiction of the U.S. Department of Energy.

Among the seven supercomputers, the initial two will be constructed by Oracle specifically for the Argonne National Laboratory. The first one, dubbed 'Equinox,' will be outfitted with 10,000 Blackwell GPUs. It is projected to come online in 2026. The second, named 'Solstice,' is an even more powerful beast. It will make use of over 100,000 Blackwell GPUs and will have a power consumption of 200 megawatts. When combined, these two supercomputers will offer a total FP4 computing power of 2200 EFlops. To put it in simpler terms, they can perform 22 quintillion floating - point operations per second. This immense computing power allows them to handle not only traditional scientific computing tasks but also support AI large - model simulations with up to 3 trillion parameters.

As for the remaining five supercomputers, 'Tara,' 'Minerva,' and 'Janus' are all built on the Blackwell architecture and will also serve the Argonne National Laboratory. On the other hand, 'Mission' and 'Vision' will adopt the next - generation Rubin architecture and will be dedicated to serving the Los Alamos National Laboratory.