NVIDIA, partners set out blueprint to build US AI infrastructure
At GTC Washington, D.C., NVIDIA detailed new national‑lab supercomputers, a Virginia research centre and a gigawatt‑scale “AI factory” blueprint developed with US partners across energy, cloud and industry, positioning the company at the core of America’s AI build‑out.
NVIDIA has announced a far‑reaching plan with US government labs, cloud providers and industrial groups to build out America’s AI infrastructure, outlining new national‑lab supercomputers, a research centre in Virginia and a “gigawatt‑scale” blueprint intended to power what it has called the next industrial revolution.
The US Department of Energy’s Argonne and Los Alamos National Laboratories unveiled seven NVIDIA‑accelerated systems under the initiative.
At Argonne, a Solstice supercomputer will feature 100,000 Blackwell GPUs, with a second system, Equinox, planned with 10,000 Blackwell GPUs and availability expected in 2026; together they have been stated to deliver 2,200 exaflops of AI performance over NVIDIA networking.
Los Alamos selected the Vera Rubin platform with Quantum‑X800 InfiniBand for its next‑generation Mission and Vision systems, which HPE is contracted to deliver; mission has been targeted for late‑2027 operations.
Gigawatt‑scale “AI factory” template
NVIDIA has said it is building an AI Factory Research Center at a Digital Realty campus in Virginia to host the first Vera Rubin infrastructure and to prove out Omniverse DSX — a blueprint that combines digital‑twin design, modular build‑outs and autonomous control of power and cooling for multi‑generation, gigawatt‑scale data centres.
Engineering partners including Bechtel and Jacobs, equipment suppliers such as Eaton, GE Vernova, Mitsubishi Electric, Schneider Electric, Siemens, Siemens Energy, Tesla, Trane and Vertiv, and software providers including Cadence, PTC, Siemens and ETAP have been named as contributors to the centre and blueprint.
System makers Cisco, Dell Technologies, HPE and Supermicro have been named as collaborators on secure, full‑stack AI infrastructure.
Cisco introduced Nexus N9100 switches built on NVIDIA Spectrum‑X silicon, while Oracle launched its OCI Zettascale10 offering powered by NVIDIA systems.
Cloud and AI builders highlighted by NVIDIA have included Akamai (Inference Cloud), CoreWeave (a new federal unit), Google Cloud (Blackwell‑based A4X Max VMs), Microsoft (GB300 NVL72 deployments), Together AI (new sites in Maryland and Tennessee) and xAI (the Colossus 2 data centre plan in Memphis).
Healthcare groups Lilly and Mayo Clinic have also been cited as deploying NVIDIA‑based “AI factories” for drug discovery and clinical research.
What is an “AI factory”?
In NVIDIA’s view, an AI factory is a tightly integrated compute facility — from GPUs and networking to power, cooling and software — designed to train and serve AI models at scale.
The Omniverse DSX blueprint uses digital twins and agentic software to design, operate and continually optimise those facilities for performance, energy efficiency and resilience.
The announcement arrived amid a broader race to expand AI compute and the energy capacity to run it. Separate multi‑billion‑dollar programmes have been advanced this year, including the OpenAI‑Oracle‑SoftBank “Stargate” build‑out targeting 10GW of capacity across new US sites, and new private‑sector energy partnerships to power future data centres.