The project gives NPS the foundation to run advanced AI workloads closer to the research, training, and mission needs they support.
From autonomous systems and battlefield planning to cyber defense and logistics, AI is reshaping how the Department of Defense trains leaders, conducts research, and solves mission-critical problems. The question is no longer whether to deploy AI, but whether it can be deployed inside the facilities already in place.
That is the challenge many enterprise, federal, and university, customers face. They have buildings, sensitive data, and urgent mission or business needs. What they do not always have is a purpose-built AI facility.
The Naval Postgraduate School (NPS) in Monterey, California, is well positioned to show what is possible. As the Department of the Navy’s graduate university, NPS educates service members and government civilians in disciplines critical to national security. Its graduates return to leadership roles, and its research can move quickly from classroom and lab into operations.
NPS had the mission need and institutional role to advance AI, but its existing computer room was built for an earlier era of IT, not for the power density, liquid cooling, and thermal load of current-generation AI platforms. It is a common constraint for organizations that need AI close to their data, people, workflows, and missions.
A platform for national security research
To expand AI capabilities, NPS and the NPS Foundation engaged industry partners that could support advanced computing, infrastructure, and deployment. The effort led to a Cooperative Research and Development Agreement with NVIDIA, designating NPS as an NVIDIA AI Technology Center. Through the agreement, NVIDIA committed advanced AI computing technology to the NPS Foundation.
NVIDIA provided the NVIDIA DGX GB300, which houses 72 NVIDIA Blackwell Ultra GPUs operating as a unified computing environment capable of more than 1,000 petaflops of AI performance, based on NVIDIA specifications. The system was selected to support NPS workloads ranging from model training and simulation to real-time inference and experimentation.
The computing platform was only part of the challenge. NPS also had to determine whether a liquid-cooled, high-density AI system could operate reliably in a room never designed for it.
Infrastructure engineered for AI compute
Vertiv supported the NPS deployment with a three-rack Vertiv™ SmartIT module, a pre-integrated AI infrastructure building block that brings power, cooling, rack infrastructure, monitoring, and deployment services into a defined solution boundary. The approach helped the team adapt Vertiv’s NVIDIA DGX GB300 reference design to the site’s physical constraints while reducing custom integration across separate infrastructure systems.
The configuration uses the Vertiv™ Liebert® APM2 480V UPS for continuous conditioned power, with two high-density PDUs managing rack-level distribution.
Cooling combines direct-to-chip liquid cooling for GPUs and CPUs with Vertiv™ Liebert® DCD47 rear door cooling units for residual heat from networking and storage components. Three Vertiv™ Liebert® XDU coolant distribution units condition and circulate coolant between the facility chilled water supply and rack-level cooling systems.
Two Vertiv™ Open Rack MGX chassis house the compute and supporting equipment. Built for the NVIDIA DGX GB300 architecture, each chassis includes a 1,400A DC power busbar and supports up to eight scalable DC Power Shelves. Seismic floor anchoring kits were installed to meet California equipment installation standards.
This was not a simple product deployment. It was the conversion of an existing room into a high-density, liquid-cooled AI environment using a defined three-rack infrastructure module. Three requirements shaped the work:
Physical readiness. The room had to be assessed and upgraded for power density, thermal load, and liquid cooling infrastructure beyond its original design.
System integration. Power, cooling, racks, seismic anchoring, fluid management, installation sequencing, and commissioning had to be engineered as one system. Vertiv worked with NPS staff, NVIDIA engineers, and the general contractor to align the deployment across each phase.
Deployment confidence. The project shows what must be in place physically to operate Blackwell-class AI on-premises: a reference architecture, site assessment, engineered power and cooling package, liquid cooling integration, commissioning process, and deployment support.
A repeatable model
The NPS deployment demonstrates that existing facilities can be converted into AI-ready environments when infrastructure is treated as part of the system design, not as an afterthought. For customers, that can reduce friction in bringing AI closer to data, people, and mission workflows. For NVIDIA and Vertiv, it reinforces the role of engineered infrastructure in making advanced AI deployable, serviceable, and reliable.
Immediate impact
The AI cluster is enabling NPS faculty, students, and research teams to take on challenges beyond the reach of legacy systems. Work is underway in AI-enhanced maritime command and control, autonomous systems, aviation maintenance, space-based sensing, and energy resilience, supported by greater compute scale and faster experimentation cycles.
The NPS deployment shows that advanced AI capability does not always require a new campus or multi-year construction project. With the right infrastructure design, it can be brought into the environments organizations already operate.