- Nvidia announced plans to build seven new AI supercomputers in collaboration with Oracle and the DoE
- These supercomputers will be used to build trillion-parameter models for AI, simulation and scientific computing
- Nvidia is also working with HPE on Mission and Vision systems for the DoE’s Los Alamos National Lab
NVIDIA GTC, WASHINGTON DC – The U.S. is about to get a major AI boost courtesy of Nvidia, AMD and a number of other brand name tech companies. But they’re not working on another Stargate project or even a hyperscaler deployment. They’re working with and for the U.S. government and specifically the Department of Energy (DoE).
Nvidia announced plans to build seven new AI supercomputers in collaboration with Oracle and the DoE. These include the Solstice system, with 100,000 Blackwell GPUs, and the Equinox system, with 10,000 of the same, at the Argonne National Laboratory. The latter is expected to come online in the first half of next year.
Nvidia said these supercomputers will be used to build trillion-parameter models for AI, simulation and scientific computing.
“This system will seamlessly connect to forefront DOE experimental facilities such as our Advanced Photon Source, allowing scientists to address some of the nation’s most pressing challenges through scientific discovery,” Paul K. Kearns, director of Argonne National Laboratory, stated.
Nvidia is also working with HPE on Mission and Vision systems for the DoE’s Los Alamos National Lab. These will be powered by Nvidia’s Vera Rubin platform and its Infiniband networking fabric. This system will be used to run classified applications and is expected to be operational in 2027.
The collaboration was such a big deal that DoE Secretary Chris Wright even made an appearance with Nvidia CEO Jensen Huang at a media roundtable. He described the deal with Nvidia as a "partnership made in heaven."
But Nvidia isn’t alone in pursuing projects with the DoE.
AMD announced this week it will build two high performance systems at the DoE’s Oak Ridge National Lab: The Lux AI and Discovery supercomputers. The former will be built alongside HPE and Oracle using AMD’s Instinct GPUs, EPYC CPUs and Pensando networking fabric and is expected to come online next year. The latter, which will be delivered in 2028 and come online for users in 2029, will also use AMD’s GPUs and CPUs.
Together, AMD said these projects represent a $1 billion investment of private and public funding.
But why the DoE?
So why are companies like Nvidia and AMD pursuing projects with the DoE when they’ve already got other large-scale projects on their plates?
“For a lot of companies, the DoE ends up being a proving ground because they are the ones who scale very large in a non-commercial environment,” VAST Data’s Field CTO for HPC Kyle Lamb told Fierce.
That is, they offer companies on the bleeding edge of AI and other technology a chance to work out kinks and prove their high-performance compute systems work in a real-world environment before they take these solutions to enterprises.
VAST Data has itself worked with the DoE and Lamb was formerly an engineer in the Los Alamos National Laboratory.
“The DoE ends up being a proving ground to show that you can actually get into those environments, that you can scale to the scale that is required, and it gives a confidence level whenever you’re talking to customers outside the DoE,” Lamb said. “The DoE Labs have plenty of people who are specifically tasked with diving into the weeds, the details of how to make these things work.”
Earlier this year, the DoE announced it would make seven AI testbeds at its national lab facilities available to researchers. The DoE has also assumed a vital role on the power side of the AI equation, opening federal sites for nuclear development and data center development.