Nexus Compute GB200 NVL72 Grace-Blackwell Rack
A liquid-cooled rack as one GPU for trillion-parameter training and inference.
We help you choose, configure, and deliver the right system — no obligation.
Configuration at a Glance
Tailored per engagement. Full technical overview below.
Configuration Options
Core specifications for this system. Every component is configurable to your workload — request a quote for a tailored build.
72x NVIDIA Blackwell GPUs + 36x Grace CPUs
36x Grace (72-core Arm Neoverse V2), 17.3TB LPDDR5X
13.5TB HBM3e unified pool
Overview
The NVIDIA GB200 NVL72 connects 36 Grace CPUs and 72 Blackwell GPUs through the largest NVLink domain ever built, presenting an entire liquid-cooled rack as a single, exascale-class accelerator. Nexus Compute specifies, integrates, tests, and warranty-backs the rack as a coherent system sourced through authorized channels, coordinating power, cooling, and fabric so it arrives commissioned rather than as parts.
Who This Solution Is For
Business Benefits
Rack acts as one GPU
A 72-GPU NVLink domain lets trillion-parameter models train and serve as if on a single massive accelerator.
Real-time giant-model inference
Unified GPU memory and NVLink bandwidth deliver real-time inference on models too large for conventional nodes.
Delivered as a system
We integrate compute, liquid cooling, and fabric and commission the rack so it performs as an engineered whole.
Typical Business Use Cases
Trillion-parameter foundation model training
Real-time inference on mixture-of-experts and giant LLMs
AI-factory and SuperPOD scale units
Sovereign and large-scale research compute
Industry Applications
Technical Overview
The GB200 NVL72 unifies 72 Blackwell GPUs and 36 Grace CPUs across 18 compute trays, linked by a fifth-generation NVLink Switch fabric providing 130TB/s of total GPU bandwidth and roughly 1.44 exaFLOPS of FP4 compute. The fully liquid-cooled rack shares a 13.5TB HBM3e pool with NVLink-C2C joining each Grace CPU to its Blackwell GPUs at 900GB/s.
| Accelerator | 72x NVIDIA Blackwell GPUs + 36x Grace CPUs |
| GPU Interconnect | 5th-gen NVLink Switch, 130TB/s aggregate |
| GPU Memory | 13.5TB HBM3e unified pool |
| FP4 Compute | ~1.44 exaFLOPS per rack |
| CPU | 36x Grace (72-core Arm Neoverse V2), 17.3TB LPDDR5X |
| Networking | Quantum-2/Quantum-X800 InfiniBand or Spectrum-X |
| Cooling | Liquid-cooled, rack-scale |
| Form Factor | Integrated rack (18 compute trays) |
| Power | Up to ~120kW per rack |
Specifications are indicative and configured to each engagement. Request a quote for a configuration tailored to your requirements.
Warranty, Support & Fulfillment
Every system ships from an authorized channel, configured and tested, with the documentation enterprise buyers need — backed by warranty and a dedicated account team.
Enterprise Warranty
Full manufacturer warranty with optional on-site, next-business-day support and extended coverage.
Authorized Channel
Sourced through Tier-1 distribution and OEM partners — never grey market. Asset & warranty records included.
Lead Time & Deployment
48-hour quotes, then configured, burn-in tested, and delivered on a committed schedule.
Nationwide Fulfillment
Coordinated logistics, rack-and-stack, and delivery wherever your infrastructure lives.
Frequently Asked Questions
What facility does an NVL72 rack require?
It is fully liquid-cooled and draws up to roughly 120kW, so it needs high-density power and a CDU or facility water loop. We assess and plan power, cooling, and floor loading early in the engagement.
Can I start with one rack and scale later?
Yes. A single NVL72 is a complete scale unit and additional racks interconnect over InfiniBand into a SuperPOD. We design the fabric so capacity grows cleanly.
Is it better for training or inference?
Both. The unified NVLink domain accelerates trillion-parameter training and enables real-time inference on the largest models. We tune storage and fabric to your primary objective.
Hardware Assistance
Configure the Nexus Compute GB200 NVL72 Grace-Blackwell Rack with Nexus Compute
Tell us your requirements and a hardware specialist will help you specify, configure, and quote the right system — typically within two business days. No obligation.