Archive · 24 Sep 2026
Supermicro Begins Shipping NVIDIA Vera Rubin NVL72 Liquid-Cooled Racks
Supermicro says it is now shipping special computer racks for NVIDIA’s Vera Rubin AI platform.
These racks use liquid instead of only air to remove heat.
Each rack contains 72 Rubin graphics processors and 36 Vera processors.
The parts are connected so the rack can work like one large machine.
Supermicro also supplies cooling equipment, power systems, networking, and installation services.
The company tests each rack with its complete cooling system before delivery.
A larger blueprint can combine 16 racks into one scalable unit.
Supermicro says this approach is designed to help customers build large AI data centers more quickly and efficiently.
Supermicro announced that it is shipping NVIDIA Vera Rubin NVL72 racks integrated with its DCBBS and DLC-2 liquid-cooling systems.
Each rack combines 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs into one liquid-cooled machine.
The racks include 18 compute trays and nine NVIDIA NVLink switch trays, providing 216 TB/s of scale-up bandwidth.
Supermicro says its in-row cooling distribution units are rated at 1.8 megawatts each and can be deployed with N+1 redundancy.
A DCBBS Blueprint based on one scalable unit includes 1,152 NVIDIA Rubin GPUs across 16 compute racks and 331 TB of HBM4 memory.
- Who
- Supermicro, also identified as Super Micro Computer, Inc., announced the shipment; the racks use NVIDIA technology.
- What
- The company is shipping liquid-cooled NVIDIA Vera Rubin NVL72 racks integrated with its Data Center Building Block Solutions and DLC-2 cooling stack.
- When
- The announcement was dated September 24.
- Where
- The announcement came from San Jose, California; the systems are intended for data-center deployments.
- Why
- Supermicro says direct liquid cooling is needed to manage the higher heat loads of advanced AI systems and support efficient deployment.
- Rack configuration
- 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs per rack
- Rack memory
- 20.7 TB of HBM4 memory and up to 54 TB of LPDDR5X
- Compute layout
- 18 1U compute trays, each with four Rubin GPUs and two Vera CPUs
- Interconnect
- Nine sixth-generation NVIDIA NVLink switch trays provide 216 TB/s of scale-up bandwidth
- Cooling capacity
- Supermicro in-row cooling distribution units are rated at 1.8 MW each, with N+1 redundancy
- Scalable unit
- One blueprint-based unit includes 1,152 NVIDIA Rubin GPUs across 16 compute racks and 331 TB of HBM4 memory
- Deployment services
- Supermicro says its team supports site surveys, design, integration, testing, delivery, deployment, and ongoing support