From the NVIDIA edition of September 15, 2026
NPS documents GB300 peak power (~135 kW) and closed-loop liquid cooling inside existing capacity
Naval Postgraduate School AI Task Force · Defense higher education · United States (Monterey, California)
- Publisher
- NPS AI Task Force GB300 page
- Original publication
- 2026 (page current at retrieval; system delivered 2026)
- Source retrieved
- 2026-09-15
What happened
NPS publishes operator-facing facts for the donated DGX GB300: 72 Blackwell Ultra GPUs, 36 Grace CPUs, ~135 kW peak compute draw with expected daily operations at 50–80% of peak, and dual closed-loop liquid cooling that rejects heat via roof cooling water. The page states the platform operates within existing power and water capacities.
Why it matters
Gives SLED facility and procurement teams quantitative planning inputs for Blackwell rack-scale education/research deployments, and clarifies that within-existing-capacity claims still require local site verification.
Evidence and measured results
NPS GB300 page lists architecture (18 trays x 4 GPU/2 CPU), FP4 performance, 20 TB HBM, use-case list (weather, cyber, LLM fine-tuning, multi-agent tactics), and Power and Cooling section with 135 kW peak and 50–80% anticipated daily operations.
Limitations and uncertainty
Institutional specifications and anticipated utilization bands are not independent measurements. Performance figures (PFLOPS) are product specs, not NPS workload benchmarks.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-15; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Campus/agency rooms often lack documented headroom for ~100+ kW liquid-cooled AI racks.
- Stakeholders
- facilities
- CIO
- HPC lead
- EHS
- procurement
- Discovery questions
- What is measured power/water headroom at the proposed room?
- Who owns liquid-cooling maintenance?
- What duty cycle is planned for the first 90 days?
- Value hypothesis
- Published NPS figures provide a concrete planning envelope for Blackwell education deployments when verified locally.
- Potential engagement
- Joint facilities+HPC assessment against the NPS-published envelope before proposal commitment.
- Unsupported claims
- Any site can host GB300 without upgrades
- 135 kW covers full support infrastructure
Pre-sales engineering
Role takeaway
- Fit
- Fit when site can sustain ~135 kW compute peaks with closed-loop liquid cooling and documented water rejection path.
- Architecture and integration
- Map compute trays, NVLink switches, CDUs/heat exchangers, and facility chilled water into one design package.
- Prerequisites
- Accurate one-line diagrams
- Coolant chemistry/monitoring
- Seismic/anchoring as required
- Duty-cycle telemetry
- Constraints
- Peak vs average draw differs
- Support components add thermal load
- Existing rooms may need upgrades despite within-capacity claims elsewhere
- Security
- Isolate research datasets and experiment tenancy on shared system.
- Proof of value
- Measured peak draw within design
- Cooling loop stability under load
- Documented tray/switch inventory
Delivery
Role takeaway
- Work
- Facility conversion, loop commissioning, and operator training for liquid-cooled operations.
- Dependencies
- mechanical/electrical contractors
- OEM partners
- HPC ops
- EHS
- Ownership
- Facilities owns plant; HPC owns compute operations; joint incident response.
- Skills and adoption
- Operators need liquid-cooling and high-density power runbooks, not only CUDA skills.
- Governance checkpoints
- Pre-energization checklist
- Leak/response drill
- 30-day duty-cycle review
- Acceptance criteria
- System runs within published power band
- Cooling alarms exercised
- Shared access SOP live
- Risks
- Underestimated support power
- Cooling water constraints
- Operator skill gaps
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Budget liquid cooling CDUs/heat exchangers, dual-loop facility water, and rack support gear as first-class design objects alongside GPUs. Size power distribution for ~135 kW compute peaks plus support loads.
Governance
Who approves, reviews and stays accountable for outcomes?
Shared research capability with approved collaborations requires written access criteria spanning faculty, students, labs, and industry partners aligned to defense priorities.
Security and privacy
What data, permissions and controls need testing?
Use cases include synthetic data for denied/classified sparsity and adversarial resilience training—implying strong data classification and experiment isolation practices.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Shared capability expands who can train on frontier systems, but only if intake and training paths are funded.
Procurement
What should contracts, pricing and exit terms secure?
Require vendors to price power, cooling, and support racks with the compute BOM; treat 135 kW as a minimum planning input to verify.
Operating model
Which teams own the service once it runs?
Expected 50–80% duty cycle implies capacity planning and chargeback/fair-share policies for campus and partner users.
Publication history
- 2026-09-15NVIDIA · Issue 1010 resources
Stable resource ID: nps-dgx-gb300-power-cooling-specs-2026