Supermicro introduced a Data Center Building Block Solutions (DCBBS) Blueprint for HPC and AI built around the NVIDIA Vera Rubin NVL4 platform, packaging compute, networking, liquid cooling, power distribution, and deployment services into a repeatable “Scalable Unit” aimed at research institutions and supercomputing centers. The company is pitching the blueprint as an end-to-end path from project planning and site survey through manufacturing, testing, and on-site deployment.
At the hardware level, the NVIDIA Vera Rubin NVL4 Scalable Unit is specified as an eight-rack, direct-to-chip liquid-cooled configuration designed around 362 kW per compute rack. Supermicro’s described build uses customized 52U, 750 mm-wide enclosures, with each rack housing 36 NVIDIA Vera Rubin NVL4 nodes. Across a full Scalable Unit, that totals 288 nodes, up to 1,152 NVIDIA Rubin GPUs, and 576 NVIDIA Vera CPUs, with the unit sized at 3.2 MW and described as scalable from 3.2 MW to 1 GW by multiplying the base block.
For cooling, the blueprint calls for Supermicro’s DLC-2 Direct Liquid Cooling stack. Each Scalable Unit includes three in-row cooling distribution units (up to 1.8 MW each) arranged in a 2+1 redundant configuration, along with direct-to-chip copper cold plates and vertically mounted cooling distribution manifolds. Supermicro also names its SMC PG25-A coolant, describing it as engineered for chemical and thermal stability and noting “ultra-high electrical impedance.”
On the networking side, the blueprint specifies an NVIDIA Quantum-X800 InfiniBand compute fabric deployed across dedicated networking switch racks, with fully liquid-cooled options available. For rack power and management, each compute rack is described as using eight 72 kW power shelves to deliver busbar power for the 362 kW rack envelope, plus two top-of-rack management switches per rack for out-of-band control.
For data center teams, the interesting part isn’t just the node counts—it’s that Supermicro is explicitly wrapping facility survey, integration work, and on-site commissioning into the same packaged design that also includes 362 kW liquid-cooled racks. That combination pushes decisions about floor loading, clearances, and power and cooling tie-ins to the front of the project, which is where high-density liquid-cooled builds typically succeed or fail.
Charles Liang, Supermicro’s president and CEO, said, “The institutions that accelerate infrastructure deployment will lead the next generation of breakthroughs.”
Supermicro said configurations for HPC and AI based on the NVIDIA GB200 NVL4 are also available for immediate deployment.
Source: Supermicro
