NVIDIA GB200 NVL72
NVIDIA · Blackwell (Grace+Blackwell)
Shipping to select customers
trainingfrontierblackwellgracerack-scale
Memory
13.8TB HBM3e (72 GPU system)
FP8 TFLOPS
324000
Power
27000 (system)W
Released
Jun 1, 2025
Technical Specifications
| Manufacturer | NVIDIA |
|---|---|
| Architecture | Blackwell (Grace+Blackwell) |
| Memory | 13.8TB HBM3e (72 GPU system) |
| Memory Bandwidth | 8.0 TB/s per GPU |
| FP16 TFLOPS | 162000 |
| FP8 TFLOPS | 324000 |
| FP4 TFLOPS | 648000 |
| Interconnect | NVLink5 (1.8TB/s) + NVSwitch |
| Power (TDP) | 27000 (system)W |
| Released | Jun 1, 2025 |
| Status | Shipping to select customers |
Availability
Limited — Dell, Supermicro, Cisco shipping systems; volume ramp H2 2026
Pricing
Retail (estimated)
3000000-3500000 (per NVL72 rack)
Cloud Pricing ($/hr)
| coreweave | $TBD/hr |
|---|
Key Specs
Memory: 13.8TB HBM3e (72 GPU system)
Bandwidth: 8.0 TB/s per GPU
FP8: 324000 TFLOPS
Power: 27000 (system)W
Datacenters Using This GPU
- xAI Memphis Supercluster
- Quintus-X AI Campus
- Crusoe West Texas AI Campus
- Microsoft West Des Moines AI Campus
- Microsoft Cheyenne Data Center
- Google Storey County Data Center
- AWS Umatilla Data Center
- Meta Temple Data Center
- Oracle Mesa Data Center Campus
- Oracle West Jordan Data Center
- CoreWeave Illinois AI Campus
- Lambda Cloud GPU Infrastructure
- UAE G42 Stargate Abu Dhabi
- Sakura Internet Ishikari AI Campus
- Singapore AI Compute Cluster