Siemens Reference Design for DSX Vera Rubin NVL72

The data center industry is entering a new era defined by AI factories, purpose-built facilities designed to manufacture intelligence at unprecedented scale, where tokens have become the primary unit of output. Platforms such as the NVIDIA Vera Rubin (VR) NVL72 introduce a rack-scale, tightly integrated compute system that fundamentally redefines infrastructure requirements. Hence, these AI factories require a co-designed approach across compute, cooling, power, and control systems to maximize performance per watt and ensure predictable, scalable operations.

This Siemens UL Reference Design for DSX Vera Rubin NVL72 focuses on the following design considerations:

• Extreme power density and highly dynamic, synchronized load profiles driven by large-scale training and inference
• Co-designed compute, cooling, and power systems to maximize tokens per watt and overall system efficiency
• High availability requirements under concentrated, multi-megawatt rack-scale load blocks
• The need to scale capacity in large, discrete increments aligned to DSX VR deployment units without overbuilding or stranding capital

To address these challenges, Siemens, together with NVIDIA, nVent, and Fluence, has developed this DSX VR-aligned reference design that translates NVIDIA’s AI factory vision into deployable, industrialized, and intelligent power and controls architecture. An example implementation with 32 Scalable Units organized into 8 Deployment Units results in a total IT load of approximately 103 MW.1

This reference design is a system-level solution that integrates Siemens electrification, automation, digitalization, and industrialized delivery capabilities with nVent’s advanced thermal management solutions, and Fluence’s grid-interactive energy storage to enable faster, safer, and more token-efficient AI factory deployment.

You can also download the complete medium- and low-voltage schematics package here, and discover more Siemens data center solutions here.