AINEWS
Housed at the Naval Postgraduate School instead of a weapons lab, the NVIDIA built DGX GB300 will train models for logistics, cyber defense, and higher resolution weather and ocean forecasting.
The Pentagon's most powerful AI supercomputer went live at the Naval Postgraduate School on July 22, 2026, an NVIDIA-built DGX GB300 system that the Navy says is the Defense Department's most powerful.
The system pairs 36 NVIDIA Grace CPUs with 72 NVIDIA Blackwell Ultra GPUs, the components that turn a server rack into an AI training platform. It was activated at a ribbon-cutting attended by NVIDIA founder and CEO Jensen Huang and Adm. Samuel Paparo, commander of U.S. Pacific Command. The hardware sits inside the school's existing power and water infrastructure on the NPS.edu network, delivered under a cooperative research and development agreement (CRADA) between NPS and NVIDIA that began in December 2024.
The deployment looks like a press-release install: ribbon-cutting, vendor CEO, the Pentagon's biggest commercial AI box. The mechanism is more interesting. The Naval Postgraduate School is a graduate institution, not a weapons lab, and the students and faculty who share time on the machine are the load-bearing part of the deal.
Capt. Michael Owen, NPS vice provost for warfare studies and the school's AI Task Force lead, framed the early use cases as defense logistics, cybersecurity threat detection models, and higher-resolution meteorology and oceanography modeling for the Fleet Numerical Meteorology and Oceanography Center (FNMOC), which sits on the same base. FNMOC's forecasting mission is described by Navy officials as unique within the Defense Department, and it is co-located at Naval Support Activity Monterey with the Naval Research Lab.
That co-location is what turns a single AI box into a capability story. A weapons lab buys a machine, plugs it into a program of record, and trains models on classified data. A graduate school buys the same machine and gives it to students who graduate into acquisition, intelligence, and operational roles across the Navy and the wider joint force, plus a co-located forecasting center whose models also serve civilian agencies. The students learn on the same hardware their commercial-sector counterparts use. The school's faculty get a training platform that, until recently, was available only to the largest commercial AI labs.
The agreement also gives NVIDIA a permanent DoD installation, a year-round proving ground where each generation of AI hardware can be tested against real military and oceanographic workloads before wider procurement. The CRADA that started in December 2024 is now an operational system, and the NPS Foundation is raising private support to expand the school's AI capacity.
The "most powerful" framing is a Navy claim, not an independent benchmark against other Defense Department HPC sites, including the Air Force Research Laboratory DSRC or the Army Engineer Research and Development Center. The Navy release and NVIDIA's blog say it, and the trade press repeated it. A direct comparison to AFRL's DSRC or the Army's supercomputers will require an independent benchmark.
What the next year looks like, in other words, depends on which models get trained. If the early work stays in logistics, cyber threat detection, and FNMOC's ocean and weather models, the machine has earned its title. If most of the compute goes to demos and conference papers, the CRADA is still real, but the capability claim is a press release.