Nvidia Gives the Navy a Blackwell Ultra Supercomputer

Nvidia (NVDA ) founder and CEO Jensen Huang traveled to Monterey, California, on July 22, 2026, to commission a DGX GB300, the company’s most powerful enterprise AI system, at the Naval Postgraduate School. The machine did not arrive through a defense procurement contract. Nvidia donated it.

The commissioning, held during the school’s Converge @ NPS event with U.S. Pacific Command chief Adm. Samuel Paparo in attendance, brings a rack-scale Blackwell Ultra system online for the roughly 1,500 in-resident students and 600 faculty at the U.S. military’s flagship graduate school. NPS plans to run model training and inference on campus for weather prediction, cybersecurity, ocean modeling and disaster-response planning.

A gift, not a purchase

The economics are the story. The DGX GB300 reached NPS as a donation routed through the Naval Postgraduate School Foundation, a nonprofit that channels private technology gifts to the school, under a cooperative research agreement the two sides signed in December 2024. Dell Technologies (DELL ) and Sterling Computers supplied the surrounding infrastructure so the system can run across unclassified, controlled and classified environments, and the foundation committed up to $2 million to staff the effort with mentors and technicians. Nvidia first said in October 2025 that NPS would be the first command in the Departments of the Navy and War to field the system.

Randy Pugh, who led the school’s AI Task Force during the rollout, valued the system at about $15 million and said it was the second unit off Nvidia’s assembly line. He described the company’s motive plainly: Nvidia understood commercial AI deployments but had little exposure to military use cases, and wanted to learn by running frontier hardware in an operational defense setting for a year or two. The supercomputer anchors a wider Nvidia AI Technology Center on the Monterey campus that also includes a roughly $5 million Omniverse simulation instance and access to the company’s training programs for faculty.

That is the trade. Nvidia spends a rack to buy a foothold in defense AI and a live view of workloads it rarely sees, while NPS gets on-premises compute at a scale no defense school has had, with no procurement cycle to wait out.

One liquid-cooled rack, existing power and water

The donation is a single DGX GB300 NVL72, a liquid-cooled rack that Nvidia configures with 72 Blackwell Ultra GPUs and 36 Grace CPUs joined into one memory domain, built for the inference and reasoning workloads that increasingly dominate AI compute. It runs Nvidia’s Mission Control orchestration software, and the company bills the DGX GB300 as the highest-performing system it sells to enterprises. Nvidia also claims the generation delivers up to 50 times the AI-factory output of its older Hopper systems, a projection it has not tied to an independent benchmark.

The infrastructure detail is the one worth dwelling on. Blackwell Ultra parts are power-hungry, each drawing more than a kilowatt, and a fully populated GB300 rack is dense enough that many facilities cannot power or cool it without new construction. NPS fit this one in anyway, according to Navy Times, which reported the machine runs within the school’s existing power and water infrastructure and billed it as the Pentagon’s most powerful supercomputer. Vertiv supplied the racks, cooling and power hardware and handled the commissioning, while DDN and VAST Data provided the storage and data platform.

Keeping the compute on campus is the point. For an institution handling controlled and classified data, owning the rack outright sidesteps both the per-hour cost of renting scarce Blackwell capacity in the cloud and the security friction of sending sensitive workloads off site. NPS plans to train foundation models and run an in-house generative AI system, an “NPS GPT,” so sensitive data never leaves its network, and to build high-fidelity digital twins of maritime environments through a framework developed with the nonprofit MITRE on Nvidia’s Omniverse platform. Public research institutions have been moving the same way: Argonne National Laboratory opened a large-scale AI inference service for open science earlier in 2026.

Why seed a defense school

For Nvidia, the placement extends a pattern of putting Blackwell hardware where it wants demand to grow. Days earlier, the company wired much of corporate Japan into its physical-AI stack; the NPS system does the equivalent for U.S. defense, seeding a generation of officers who will specify and buy compute later in their careers. The school stood up a master’s degree in artificial intelligence in December 2025, with a first cohort now underway.

The Defense Department already runs its own supercomputers, but those centers, such as the Army Corps of Engineers facility in Vicksburg, Mississippi, are built for operational production runs. NPS is pitching its machine as a place to rebuild and improve the models themselves. The base at Naval Support Activity Monterey also hosts the Navy’s Fleet Numerical Meteorology and Oceanography Center, and on-campus Blackwell Ultra capacity lets researchers push weather models to far higher resolution than the roughly 50-kilometer grids used for general forecasting, a difference that matters for planning operations rather than a morning commute. Paparo told attendees that officers will increasingly command in AI-enabled settings where response times compress and advantage goes to those who can act faster than adversaries. The compute to support that is now in Monterey, and Nvidia, not the Navy’s budget, paid for it.