System-level modular infrastructure reference design for NVIDIA Vera Rubin NVL72

Data hall planning basis

The NVIDIA-based reference design supports 13.44 MW of AI load across 56 GPU racks at a nominal 200 kW and 64 support racks at 35 kW. At a planning PUE of approximately 1.27, total site load is estimated at approximately 17.1 MW, including 2.8 MW for mechanical loads.

AI Data Hall layout

Each AI Module aligns to a dedicated infrastructure zone for power, cooling, controls, and service access.

• Power block — Modular, prefabricated low-voltage power distribution sized around the AI module load profile
• Cooling block — Liquid- and air-cooling capacity coordinated to support mixed rack densities 
• Compute block — A white-space IT building block that combines a GPU module and a support module into one coordinated unit
• Controls zone — Module-level monitoring, controls, and alarm management
• Service boundary — Defined maintenance access, CDU service zones, busway access, and rack replacement aisles

Coordinated standard modular deployment blocks

The reference design organizes data hall infrastructure into standard deployment blocks so that compute, support, power, cooling, controls, and services scale together as coordinated units rather than as separately engineered systems.

• Accelerated deployment  —  Factory-integrated power, cooling and controls reduce onsite construction and commissioning delays, enabling faster time to capacity.
• More effective power utilization  —  Modular capacity blocks reduce stranded power while improving scalability and infrastructure efficiency. 
• Repeatable, predictable execution  —  Factory-built and tested solutions improve quality, deployment consistency, and schedule certainty. 
• Flexible cooling for evolving AI requirements  —  Integrated liquid and air cooling with coordinated controls support mixed rack densities and changing compute requirements. 
• Scalable path to 800 VDC  —  The future-ready architecture supports increasing AI power demands while providing a migration path to next-generation power distribution. 
• Resilient, AI-ready infrastructure  —  Modular power distribution, coordinated controls and NVIDIA DSX™ compatible designs balance reliability, efficiency, and operational flexibility.