Supermicro의 NVIDIA Vera Rubin NVL72 GPU 시스템이 1.8 MW 인-로우 액체 냉각 및 1,152-GPU 확장 가능한 단위를 갖춘 채 현재 출하 중입니다.

Supermicro is now shipping NVIDIA Vera Rubin NVL72 racks integrated with its Data Center Building Block Solutions (DCBBS) framework and its DLC-2 direct liquid cooling stack. Each rack follows the VR NVL72 design that NVIDIA launched at CES 2026, combining 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs in a single liquid-cooled unit that operates as one machine. The configuration spreads across 18 1U compute trays, each holding four Rubin GPUs and two Vera CPUs. Nine sixth-generation NVIDIA NVLink switch trays connect these compute trays and deliver 216 TB/s of scale-up bandwidth. Each rack carries 20.7 TB of HBM4 and up to 54 TB of LPDDR5X.

"We have spent years building the liquid-cooling stack, the manufacturing capacity, and the deployment teams for exactly this moment," said Charles Liang, president and CEO of Supermicro. "Our customers can now order a Scalable Unit and receive production-ready systems with end-to-end integration, because we design and build every layer between the cold plate and the cooling tower."

NVIDIA designed the Vera Rubin platform around direct liquid cooling, requiring heat to move through a complete fluid distribution loop, from cold plates through manifolds to the cooling tower. Supermicro provides the entire pathway from its portfolio, which encompasses cold plates, manifolds, hose kits, rack power shelves, in-row CDUs, in-rack CDUs, Liquid-to-Air sidecar CDUs, rear door heat exchangers, and facility-side cooling towers. Supermicro first outlined its liquid-cooling strategy for Vera Rubin NVL72 in January.

The shipping NVL72 package pairs the racks with Supermicro in-row cooling distribution units (CDUs) rated at 1.8 MW each, deployed with N+1 redundancy, and optional rear door heat exchangers to capture residual heat. Supermicro tests and validates every rack with the full liquid cooling stack, accelerating time-to-online once systems are deployed.

Supermicro's DCBBS Blueprint defines a balanced bill of materials for a given power envelope, scaling from 5 MW to gigawatt capacity—the same approach used for its Vera Rubin NVL4 DCBBS Blueprint in June. One Vera Rubin NVL72 Scalable Unit comprises 16 compute racks with 1,152 Rubin GPUs and 331 TB of HBM4, sized to match corresponding cooling capacity, power delivery, high-performance storage, context memory storage, and networking.

Supermicro manages complete integration, including networking and cabling to the NVIDIA Reference Architecture across the AI compute fabric, converged fabric, and out-of-band management. The company provides end-to-end project management from site survey and design through integration, testing, delivery, deployment, and ongoing support. Supermicro's rack-scale compute capabilities are also integrated into Cisco's Secure AI Factory, which added Vera Rubin NVL72 support in August.