Supermicro says liquid-cooled Vera Rubin NVL72 racks are shipping

Supermicro says it has begun shipping NVIDIA Vera Rubin NVL72 racks integrated with its direct-liquid-cooling equipment and data-centre deployment services.

Each NVL72 rack contains 72 NVIDIA Rubin GPUs and 36 Vera CPUs, according to the company. Supermicro describes a wider package that includes cold plates, manifolds, power shelves, cooling distribution units, rear-door heat exchangers and facility-side cooling towers.

A shipping rack is not a finished data centre

The announcement is useful because it moves beyond an accelerator specification to the infrastructure needed to install it. Higher compute density concentrates electrical demand and heat in the rack. The facility must provide suitable power distribution, water quality, flow, pressure, heat rejection, monitoring and redundancy before the system can operate as intended.

Supermicro says its in-row cooling distribution units are rated at 1.8 MW and can be deployed with N+1 redundancy. It also offers design blueprints for larger clusters. Those cluster figures should not be treated as the specification of one rack or as evidence that every customer site is ready to accept the equipment.

For buyers, the practical questions include the facility connection point, coolant specification, allowable temperatures, leak detection, maintenance responsibility, acceptance testing and the lead time for electrical and mechanical works. Cabling, networking, floor loading and commissioning also sit outside the accelerator bill of materials.

Supermicro says it tests racks with the complete liquid-cooling stack and offers support from site survey through deployment. The announcement does not identify customer installations or provide independent operating data. “Shipping” confirms a supplier availability claim, not a completed deployment or measured efficiency at a customer site.