Supermicro released DCBBS reference designs for NVIDIA Vera Rubin enabling standardized GPU deployments from 5MW to 1GW scale.
Super Micro Computer introduces Data Center Building Block Solutions (DCBBS) Blueprints based on the NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8 platforms, designed for gigawatt-scale AI data center deployment. The Blueprints begin with a scalable unit of 1,152 GPUs that can be multiplied to virtually any size and include end-to-end design and delivery with a dedicated team of experts covering the full deployment lifecycle. According to Charles Liang, president and CEO of Supermicro, "The NVIDIA Vera Rubin NVL72 platform sets a new standard for AI factory performance, and our DCBBS Blueprints give customers a proven, end-to-end path to build at any scale — from 5MW to 1GW." He added that the company's experience with deploying some of the earliest and largest liquid-cooled AI factories is "built into every Blueprint — so our customers can move from design to fully operational faster than ever before."
The DCBBS Blueprints address the practical implementation challenges of advanced AI infrastructure by starting from a fundamental constraint: available power. The Blueprints feature a balanced bill-of-materials for power envelopes ranging from 5MW to 1GW, providing the correct ratio of cooling capacity, power delivery, compute nodes, management nodes, high-performance storage nodes, context memory storage platform nodes, and networking. This approach ensures optimal performance while avoiding bottlenecks such as network oversubscription, power capacity limitations, thermal throttling, or other limitations. Supermicro has successfully deployed the world's largest liquid-cooled AI factories featuring over 100,000 GPUs, and this proven track record is embedded in the Blueprint methodology.
Each scalable unit at the heart of the DCBBS Blueprints contains 1,152 NVIDIA Rubin GPUs with 331TB of HBM4 GPU memory. The Vera Rubin generation doubles GPU memory bandwidth, GPU-to-GPU NVLink bandwidth, and per-GPU networking bandwidth compared to NVIDIA Blackwell, providing the architectural foundation for training and inference of frontier AI models with multiple trillions of parameters. The Blueprints include 2x context memory storage platform racks streamlined for long-context inference, agentic working memory, and retrieval workloads, while ensuring alignment with NVIDIA's latest reference architecture.
The single-vendor accountability model eliminates complexity inherent in typical AI infrastructure buildouts, which involve more than a dozen distinct supplier relationships across compute, storage, networking, racks, cooling distribution, cooling towers, power infrastructure, battery backup, cabling, and services. The DCBBS Blueprints for NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8 are now available for customer engagements with deployments scheduled for the second half of 2026, aligned with NVIDIA Vera Rubin general availability. Supermicro will demonstrate both platforms at Computex booth N0824 from June 2-6, 2026, in Taipei, Taiwan, with additional demonstrations at NVIDIA GTC Taipei.