Building an AI data center is no longer just a computing problem. It’s an energy problem. Siemens, Nvidia, and Fluence just dropped their answer: a UL-aligned reference architecture designed specifically for Nvidia’s DSX Vera Rubin AI factory, built around the NVL72 platform, with a total facility capacity of 136 MW and an IT load of 100 MW.
What the architecture actually includes
The reference design covers the full electrical stack. Utility connections operate at a nominal voltage of 34.5 kV, with medium- and low-voltage distribution running all the way to individual rack interfaces.
The design meets Tier III concurrent maintainability standards, meaning any component can be taken offline for maintenance without disrupting the data center’s operations.
Fluence’s Smartstack battery energy storage platform is baked directly into the design to handle grid interconnection, load shaping, and power variability. AI workloads are notoriously spiky. Training runs can hammer the grid for hours, then drop to near-idle. Fluence’s batteries smooth out those peaks and valleys, letting the facility draw power more evenly from the grid while still delivering burst capacity when the GPUs need it.
