NVIDIA Vera Rubin NVL72
The generation after Blackwell: a liquid-cooled rack of 72 Rubin GPUs and 36 Vera CPUs. NVIDIA says it is in full production, with production shipments beginning this fall.
NVIDIA · Next generationIn production
Vera Rubin
Rubin GPU · Vera CPU · rack-scale
- Memory
- 288 GB HBM4
- Bandwidth
- Up to 22 TB/s
- Rack
- 72 GPUs · 36 CPUs
What has been announced
As published by NVIDIA, checked 24 September 2026.
- System
- 36 Vera Rubin Superchips: 72 Rubin GPUs and 36 Vera CPUs per rack
- Cooling
- 100% liquid cooling, designed for a 45°C warm-water inlet
- GPU memory
- 288 GB HBM4 per GPU; 20.7 TB HBM4 per rack
- Memory bandwidth
- Up to 22 TB/s per GPU in NVIDIA's datasheet; the product page table lists 19.2 TB/s
- FP4 compute
- 3,600 PFLOPS NVFP4 inference (with sparsity) and 2,520 PFLOPS NVFP4 training (dense), per rack
- Interconnect
- Sixth-generation NVLink: 3.6 TB/s per GPU, 260 TB/s per rack
- Power
- Not published for the GPU or the rack
- Status
- In full production; NVIDIA says production shipments begin this fall
Configuration and site considerations
Vera Rubin raises the facility requirements, and its delivery timing is part of the plan.
- Plan for 100% liquid cooling with warm-water inlet up to 45°C.
- Power figures are not yet published; confirm the site's power plan with the system builder.
- Confirm the delivery schedule with the system builder as shipments begin.
Prepare this system for a financing review
Document the configuration, the supplier and the deployment so the request can be assessed against the actual equipment.**
System configuration and quantity
Supplier and purchase order
Or fleet details and ownership information for deployed servers.
Deployment site and colocation agreement
Power and cooling readiness
Customer contract or documented on-demand revenue
