Schneider Electric is developing AI-driven data centre infrastructure through its collaboration with NVIDIA and AVEVA. Announcements made at NVIDIA GTC in San Jose highlight developments aimed at supporting AI data centre design and operations.
The company introduced a reference design tailored for the NVIDIA Vera Rubin NVL72 racks, integrating with its controls reference designs and focusing on infrastructure requirements. The design includes:
- A new power distribution approach using a 480 VAC supply voltage.
- Support for a 45°C TCS loop supply temperature to improve efficiency.
- A new IT room structure with centralised networking and storage clusters, supporting closer physical proximity and power delivery for GPU racks.
- Improved token performance by aligning data centres to different GPU rack operational points, optimising power efficiency through redundancy.
- The design is validated using ETAP models for electrical systems and ITD CFD models for airflow.
Schneider Electric’s subsidiary AVEVA, together with NVIDIA, has also introduced a digital twin architecture designed to support the development of gigawatt-scale AI factories and improve GPU efficiency. This collaboration embeds AVEVA’s software within NVIDIA’s Omniverse DSX, supporting simulation and design processes.
Through simulation, AVEVA models power distribution, thermal dynamics, and airflow, enabling a performance-optimised design before physical construction. This approach is intended to reduce engineering cycles and improve deployment accuracy.
In addition, Schneider Electric has integrated NVIDIA Nemotron models into an agentic AI system for alarm management. This capability is designed to interpret alarms across systems, identify root causes, and recommend corrective actions in real time.
The system uses real-time IoT data to support faster issue resolution, reduce unnecessary technician dispatches, and improve operational resilience.
