Schneider Electric and NVIDIA Design Gigawatt-Scale AI Factories

[Erkan Teskancan]

## Schneider Electric and NVIDIA Design Gigawatt-Scale AI Factories

Schneider Electric and NVIDIA are developing validated reference designs and digital twin architectures to optimize the design, simulation, and operation of large-scale AI factories.

Traditional data center structures are approaching physical and efficiency limits with the growing demand for artificial intelligence. At the NVIDIA GTC event, Schneider Electric, in collaboration with NVIDIA and AVEVA, presented innovative solutions that synchronize high-density power architectures with real-time digital twin simulations. This collaboration offers a validated design that addresses the power, cooling, and operational challenges of large-scale AI installations.

### Innovations of the NVIDIA Vera Rubin NVL72 Reference Design

• Reference design specifically developed for the NVIDIA Vera Rubin NVL72 architecture
• Increased energy efficiency with 480 VAC high-voltage distribution
• Energy savings by allowing cooling circuit temperatures up to 45°C
• Physical proximity to optimize network communication of AI racks, as well as high-voltage power distribution
• Overcoming power constraints by prioritizing token/watt with "MaxQ" operating mode
• Guaranteed higher performance and redundancy compared to standard cooling methods

### Time Savings with Lifecycle Digital Twins

AVEVA industrial software, integrated with NVIDIA Omniverse DSX Blueprint, enables full simulation of AI factories before construction. This verifies power distribution, thermal dynamics, and airflow under realistic load conditions.

• Embedding engineering and operations software into the design environment
• Performance validation with multi-domain simulations
• Electrical grid and airflow modeling with ETAP
• Acceleration of engineering processes and reduction of error margins

### Autonomous Alarm Management with Agentic AI

Schneider Electric is developing an AI-based solution using NVIDIA Nemotron open models to reduce alarm fatigue in data center services. This system analyzes IoT data streams to interpret complex alerts and identify root causes.

• Real-time problem diagnosis as an alternative to traditional rule-based monitoring
• Autonomous recommendations supported by expert technicians
• Increased data center resilience

The collaboration between Schneider Electric and NVIDIA also includes the transition to 800 VDC architectures and the integration of 3D models with OpenUSD.

These developments open a new era in the industry by solving critical infrastructure problems for large-scale AI factories.