Supermicro Starts Shipping NVIDIA Vera Rubin NVL72 Racks

Supermicro NVIDIA Vera Rubin NVL72 rack-scale AI systems shipping, photorealistic editorial image of tall liquid-cooled AI server racks on a clean data center production and deployment floor, realistic cables coolant manifolds and technicians, no text, no logos, landscape news cover composition, polished realistic editorial quality

Supermicro has begun shipping NVIDIA Vera Rubin NVL72 racks, moving NVIDIA’s next-generation rack-scale AI platform into customer deployments.

The systems combine compute, networking, power delivery and direct liquid cooling. Supermicro says one DCBBS Scalable Unit contains 1,152 NVIDIA Rubin GPUs and 331 TB of HBM4 memory across 16 compute racks.

72 Rubin GPUs in one rack

A Vera Rubin NVL72 rack integrates 72 Rubin GPUs and 36 Vera CPUs. The system uses NVIDIA NVLink switching to connect the compute trays into a rack-scale machine rather than treating each server as an isolated system.

Each rack carries roughly 20.7 TB of HBM4. At the larger Scalable Unit level, Supermicro balances 16 compute racks with storage, networking, power and cooling infrastructure.

Cooling from the chip to the cooling tower

Supermicro’s DLC-2 direct liquid-cooling system covers the thermal path from cold plates and manifolds through coolant distribution units and cooling towers. Its in-row CDUs are rated at 1.8 MW each and can be deployed with N+1 redundancy.

The approach reflects the same infrastructure shift BitcoinVersus.tech has been tracking across data centers: AI hardware can no longer be separated from the megawatts, cooling capacity, networking and physical infrastructure required to operate it.

From 5 MW to gigawatt scale

Supermicro says its Data Center Building Block Solutions can scale from approximately 5 MW deployments toward gigawatt-class AI campuses. The company provides site surveys, project design, integration, testing, deployment and ongoing support alongside the hardware.

For data-center operators, the important development is that Vera Rubin is moving beyond previews and reference designs. Production rack systems are now entering the deployment stage, bringing HBM4, rack-scale NVLink and much denser liquid-cooled compute into real facilities.

Sources: Supermicro and StorageReview.

BitcoinVersus.Tech Editor’s Note:
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please donate here: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb

6 responses to “Supermicro Starts Shipping NVIDIA Vera Rubin NVL72 Racks”

  1. […] convergence of compute and infrastructure through ABB’s 800 VDC AI data-center power system, Supermicro’s Vera Rubin rack systems and Huawei’s grid-interactive AI data […]

    Like

  2. […] same convergence from the facility side through ABB’s 800 VDC AI data-center power system, Supermicro’s Vera Rubin rack systems, and Samsung’s AI-RAN networking and robotics […]

    Like

  3. […] trend BitcoinVersus.tech has tracked through ABB’s 800 VDC AI data-center power system and Supermicro’s NVIDIA Vera Rubin racks: compute growth is increasingly constrained by how quickly operators can deliver electricity, […]

    Like

  4. […] covered the same infrastructure convergence through ABB’s 800 VDC AI power architecture, Supermicro’s Vera Rubin NVL72 systems, and Synopsys and TSMC’s advanced AI-chip design […]

    Like

  5. […] BitcoinVersus.tech recently covered Giga Computing’s 700 kW GAIFA validation facility, ABB’s 800 VDC AI power architecture and Supermicro’s Vera Rubin rack systems. […]

    Like

  6. […] deployment of up to 100,000 NVIDIA Vera Rubin GPUs. BitcoinVersus.tech has separately tracked Vera Rubin NVL72 rack deployments and the growing electrical challenge behind flexible AI data-center […]

    Like

Leave a comment