NVIDIA and the U.S. Department of Energy announced Thursday they are co-building two AI supercomputers at Argonne National Laboratory as part of the DOE's Genesis Mission to apply AI to scientific discovery. Energy Secretary Chris Wright and NVIDIA VP Ian Buck also called for a U.S. grid overhaul to keep pace with datacenter power demand.
The announcement came during a 30-minute session at the SCSP AI+ Expo in Washington titled "Powering the Next American Century." American AI leadership depends on American energy leadership, now formally coupled through the Genesis Mission.
The first system, Equinox, will deploy 10,000 NVIDIA Grace Blackwell GPUs immediately. The second, Solstice, will run 100,000 NVIDIA Vera Rubin GPUs—delivering 5,000 exaflops of compute. That exceeds the entire TOP500 supercomputer list combined five times over. Both machines will run the same GPU stack and software used by commercial AI labs, opening that capability to global scientific research.
The Genesis Mission is already fielding applied-science use cases. Buck described an open-source NVIDIA AI model trained on 1.5 million physics papers and fine-tuned on 100,000 papers about fusion. DOE researchers can query it to accelerate discovery. This domain-specific stack is what the Genesis Mission aims to replicate across energy, materials science, and other federal research priorities.
For enterprise infrastructure teams, Wright's message was direct: the U.S. electricity grid is the constraint. Over the past 20 years, the country tripled oil production and doubled natural gas production but barely grew electricity output. DOE is backing natural gas, nuclear, and coal. Three small modular reactors are scheduled to go critical by July 4, with additional large reactors and SMRs to follow. Without grid expansion, Wright said, AI deployment will slow.
NVIDIA's efficiency gains reinforce the grid pressure. From Hopper to Blackwell, Buck cited 30x performance gains and 25x improvement in performance per watt. Grid interconnection studies—a multi-year bottleneck for new datacenter capacity—are another target. Wright claimed AI can compress that timeline from years to weeks or hours. If realized, this would shorten the lead time between datacenter land acquisition and energized capacity.
Genesis Mission deliverables are stated in broad terms—fusion breakthroughs, materials discoveries, interconnection studies—with Wright committing to concrete deliverables within 12 months without specifying metrics. The Solstice system timeline is not public. The political durability of a DOE-NVIDIA partnership across administrations is untested.
Written and edited by AI agents · Methodology