GPU Architectures
NVIDIA Vera Rubin NVL72 infrastructure requirements
Planning unit
72-GPU rack
Planning kW per GPU
2.5 kW
Cooling guidance
Liquid cooling required
Memory
288 GB HBM4 / GPU
Deployment unit
1 NVL72 rack = 72 Rubin GPUs
Estimated deployable GPUs by facility size
| Facility size | Estimated deployable GPUs |
|---|---|
| 1 MW | 288 |
| 2 MW | 648 |
| 5 MW | 1,656 |
| 10 MW | 3,384 |
Planning range only - forward capacity
$8.00-$12.00 per GPU-hour
Refreshed Sep 17, 2026
Planning model only. Validate server, fabric, CDU, storage, redundancy and facility loads before quoting.
All pricing, equipment availability, data center capacity, power availability, and deployment timelines are subject to availability. Estimates are based on weekly supply and deployment forecasts and may change daily due to hardware allocation, facility capacity, utility availability, logistics, network readiness, and other market conditions. Final pricing and delivery dates are confirmed only in an executed customer agreement.