Compute · Rack-scale
NVIDIA GB300 NVL72
A fully liquid-cooled rack-scale system — 72 Blackwell Ultra GPUs and 36 Grace CPUs acting as one accelerator, for frontier-model training and agentic AI at the highest scale.
Specifications
One rack, one giant GPU
| GPUs per rack | 72 × Blackwell Ultra |
| CPUs per rack | 36 × Grace |
| GPU interconnect | NVLink 5, rack-scale fabric |
| Cluster interconnect | 800 Gb/s InfiniBand per GPU (Quantum-X800) |
| Cooling | Fully liquid-cooled |
| Availability | By reservation — contact sales |
Interconnect
The network is not the bottleneck
Each GPU gets 800 Gb/s of dedicated InfiniBand bandwidth, so all-reduce traffic across racks keeps pace with the NVLink fabric inside them. Distributed jobs scale the way the reference architecture says they should — we validate it per deployment.
Suited workloads
Frontier-model pre-training
Large-scale agentic AI
Massive MoE serving
Multi-trillion-token runs
Start your journey today
Tell us about your workload — an engineer, not a sales bot, will get back to you within one business day.