USD.AI

NVIDIA GB200 NVL72

A liquid-cooled, rack-scale system built from 36 GB200 Grace Blackwell Superchips, each pairing one Grace CPU with two Blackwell GPUs, for 72 GPUs in one NVLink domain.

Discuss GPU financingPrepare your documents
NVIDIA · Current generationShipping

GB200

Blackwell · rack-scale NVL72

Memory
186 GB HBM3E
Bandwidth
8 TB/s
Rack
72 GPUs · 36 CPUs

Specifications at a glance

As published by NVIDIA, checked 24 September 2026. Configurations vary by system builder.

System
36 GB200 Superchips: 72 Blackwell GPUs and 36 Grace CPUs, with one 72-GPU NVLink domain
Cooling
Liquid-cooled
GPU memory
186 GB HBM3E per GPU; 13.4 TB HBM3E per rack
Memory bandwidth
8 TB/s per GPU; 576 TB/s per rack
FP4 compute
1,440 PFLOPS with sparsity, 720 PFLOPS dense, per rack
Interconnect
Fifth-generation NVLink: 1.8 TB/s per GPU, 130 TB/s per rack
Power
Configurable up to 1,200 W per GPU; rack power not published
Status
Shipping; generally available in the cloud since February 2025
NVIDIA GB200 NVL72Blackwell datasheet

Configuration and site considerations

Like GB300, the GB200 NVL72 is bought and deployed as a whole rack.

  • Plan for a liquid-cooled rack and matching facility cooling.
  • Confirm rack power with the system builder; NVIDIA publishes per-GPU power only.
  • Document the rack count, the system builder and the delivery schedule, or fleet records for deployed racks.

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

Bring the equipment plan to a financing conversation

Discuss GPU financing

*Yield information: Current net and gross APYs are the staking yields reported by the USD.AI app at the displayed observation time. Projected gross APY is a forward-looking estimate and may rely on assumptions about loan deployment, borrower performance, reserve or escrow earnings, utilization and fees. Loan APRs describe loan interest rates and are distinct from staking APYs. Figures refresh periodically; a stale or unavailable source is identified beside the metric. No yield or projection is guaranteed. Actual performance may be lower, zero or negative. Withdrawals depend on the queue. Loan defaults, collateral losses, and smart contract or operational failures can cause losses.

**All financing subject to underwriting, eligibility and definitive documentation. Application submission begins a review. It does not establish a financing commitment. Calculator results and Compute IQ's sample deal are illustrative scenarios based on the inputs shown, not a financing offer or approval; the final loan amount depends on GPU condition and verification. Listing a system is not a statement of eligibility; equipment and transaction eligibility are reviewed individually.

This communication is not directed at residents of the United States, United Kingdom, or other jurisdictions where the distribution of such content may be restricted. The USD.AI Protocol is non-custodial, decentralized software providing structured lending infrastructure. The Protocol does not custody assets, execute trades, or act as a broker, dealer, or investment adviser. Nothing herein constitutes investment advice, a recommendation, or an offer or solicitation to purchase any digital asset. All interactions are executed directly between users and on-chain smart contracts, subject to applicable protocol terms. Participation involves risk including potential loss of funds. Do your own research.

Product names are trademarks of their owners and imply no partnership or endorsement. 3D hardware renderings are illustrative and may differ from the actual products.

© 2026 USD.AI. All Rights reserved