The United States is accelerating efforts to build the physical infrastructure for the new AI economy. NVIDIA has announced, alongside the national laboratories, cloud providers, server manufacturers, and major corporations, a coordinated deployment of AI supercomputing, AI factories, and an industrial blueprint aimed at standardizing how these facilities are established and operated at a multi-generational and gigawatt scale. The company describes this as an “Apollo” moment for the industry: whoever controls the infrastructure and operation of AI will set the pace for scientific and economic innovation over the next decade.
Seven new systems at Argonne and Los Alamos
The Department of Energy (DOE) will incorporate seven NVIDIA-powered systems across the Argonne (Illinois) and Los Alamos (New Mexico) laboratories to drive applications in security, science, and energy.
- Solstice (Argonne): will be the largest AI supercomputer in the DOE, equipped with 100,000 NVIDIA Blackwell GPUs, focused on training and deploying frontier models and reasoning agents for open science.
- Equinox (Argonne): will add 10,000 Blackwell GPUs and become available in 2026.
Both systems will be interconnected via NVIDIA networks and, together, will deliver 2,200 exaflops of AI performance.
Additionally, Argonne unveils Tara, Minerva, and Janus, three NVIDIA-based systems that will expand on-demand AI computing access for researchers nationwide, reducing discovery times and boosting productivity.
At Los Alamos, the lab has selected the NVIDIA Vera Rubin platform along with the Quantum-X800 InfiniBand network for its Mission and Vision systems — built by HPE. Mission (ATS5), scheduled for late 2027, will run classified workloads from the Advanced Simulation and Computing program; Vision continues the trajectory of the supercomputer Venado in unclassified research. The Vera Rubin + Quantum-X800 duo aims for complex simulations in materials, climate, and quantum, featuring extreme bandwidth and ultralow latencies.
A research center and a plan for gigawatt-scale AI factories
NVIDIA will build the AI Factory Research Center in Virginia (operating within a Digital Realty campus), hosting the first Vera Rubin infrastructure and serving as a testbed for digital twins and large-scale simulation. On this foundation, the company introduces Omniverse DSX, a blueprint for building and operating multi-generation AI factories at gigawatt scale, supported by NVIDIA Omniverse libraries.
This proposal integrates virtual and physical systems to optimize performance, energy, and sustainability through AI agents that continuously learn from the facility’s lifecycle — from design to operation.
Industrial ecosystem of the blueprint
- Design and construction: Bechtel and Jacobs will develop digital twins for validated designs of architecture, power, mechanical, and electrical systems in complex environments.
- Energy, power, and cooling: Eaton, GE Vernova, Hitachi, Mitsubishi Electric, Schneider Electric, Siemens, Siemens Energy, Tesla, Trane Technologies, and Vertiv are modeling grid-factory interactions at gigawatt scale, with liquid cooling and power conversion optimized for Grace Blackwell and Vera Rubin.
- Software and operational agents: Cadence, Emerald AI, Phaidra, PTC, Schneider Electric ETAP, Siemens, and Switch provide digital twin and autonomous optimization solutions for power, cooling, and workloads, transforming the Omniverse DSX blueprint into a self-learning system that enhances network flexibility, resilience, and energy efficiency.
Industry alignment: servers, networks, and public sector standards
Cisco, Dell Technologies, HPE, and Supermicro will integrate NVIDIA GPUs and AI software into complete stack systems, including designs based on the newly announced AI Factory for Government, aimed at government agencies and regulated sectors. Notably, Cisco is launching the Nexus N9100 series with NVIDIA Spectrum-X Ethernet silicon, compatible with their existing management framework for AI factories with high-speed fabrics; switches will be available before year-end. Cisco will also offer an AI Factory compatible with the NVIDIA Cloud Partner program.
Cloud and “model builders”: more capacity closer to users
The deployment extends into the cloud and the large model builders in the U.S.:
- Akamai launches Inference Cloud, a distributed platform bringing inference from core to edge in 20 initial locations (including five in the U.S.), accelerated by RTX PRO Servers.
- CoreWeave creates CoreWeave Federal, with secure and compliant AI infrastructure for government, based on NVIDIA GPUs and validated designs — with expectations of FedRAMP and other authorizations.
- Global AI orders 128 racks of GB300 NVL72 (over 9,000 GPUs) for the largest NVL72 deployment in New York.
- Google Cloud offers new A4X Max VMs with GB300 NVL72 and G4 VMs with RTX PRO 6000 Blackwell, plus brings Blackwell on-premises and air-gapped solutions via Google Distributed Cloud.
- Lambda is constructing a >100 MW AI factory in Kansas City, Missouri with over 10,000 GB300 NVL72 GPUs in the initial phase.
- Microsoft utilizes RTX PRO 6000 Blackwell within Azure and has deployed a large-scale cluster with GB300 NVL72 for OpenAI; it will add Azure Local with RTX in the coming months.
- Oracle introduces OCI Zettascale10, its largest-ever cloud-based AI supercomputer, supported by NVIDIA infrastructure.
- Together AI, featuring 5C, operates an AI factory in Maryland (GPU B200) and will open another in Memphis, Tennessee with GB200 and GB300, with expansions planned in the short term and new sites in 2026.
- xAI is developing Colossus 2 in Memphis, with >500,000 NVIDIA GPUs for training and inference of next-generation models.
Pharmaceutical and health industries: vertical AI factories
- Lilly is building the most powerful AI factory in the pharmaceutical sector with DGX SuperPOD and DGX B300, Spectrum-X network, and Mission Control for biomedical foundational models that accelerate drug discovery and design.
- Mayo Clinic, with 20 million digitized pathology preparations and one of the largest patient databases, operates an AI factory with DGX SuperPOD (DGX B200) and Mission Control for research, digital pathology, and personalized medicine.
What this movement means
- Real-world scale for science and industry. The computing park (Solstice, Equinox, and complementary systems) and the standardization of AI factories bring frontier computing into operational territory, with timelines and costs aligned for science, business, and the public sector.
- A replicable blueprint. With Omniverse DSX, digital twins and autonomous operation via agents project a model that minimizes risks in design, commissioning, and operation of gigawatt-scale AI facilities.
- Sovereignty and cross-sector impact. From cybersecurity and defense to healthcare, energy, and advanced manufacturing, the approach aims for local capacity, compliance in FedRAMP/high-assurance environments, and direct transfer to the productive fabric.
- Energy efficiency as a pillar. The focus on liquid cooling, power modeling, and closed-loop optimization anticipates increasing densities with a controlled energy footprint.
Key figures at a glance
- 100,000 GPU Blackwell at Solstice and 10,000 at Equinox (Argonne), totaling 2,200 exaflops of combined AI performance.
- Seven new systems between Argonne and Los Alamos; Mission (ATS5) operational by 2027.
- >100 MW in Lambda’s new AI factory; over 500,000 GPU in Colossus 2 (Memphis).
- Vertical factories in pharma (Lilly) and health (Mayo Clinic) with DGX SuperPOD and Mission Control.
What to expect next
The success of this plan will depend on three key levers:
- Supply chain (HBM memory, networks, energy, cooling).
- Governance and security (from AI Factory for Government to multi-cloud operations and on-premise).
- Talent and productivity (training to operate agents, digital twins, and MLOps in high-density facilities).
If the roadmap is followed, the United States will transition from flagship projects to a network of AI factories and supercomputing capabilities that, connected through twins and agents, will continually elevate the country’s scientific and industrial productivity.
via: nvidianews.nvidia