Schneider & NVIDIA unveil AI factory blueprints, twins

Schneider Electric has expanded its work with NVIDIA on data centre infrastructure, releasing new validated design blueprints for gigawatt-scale "AI factories" and a reference design for NVIDIA's latest rack-scale systems.

The announcement also includes a digital twin architecture developed with AVEVA, Schneider Electric's industrial software business, and early testing of an agentic AI model for data centre alarm management.

Vera Rubin design

A new reference design covers power distribution and cooling for NVIDIA Vera Rubin NVL72 racks. It integrates with Schneider Electric controls reference designs and outlines an approach to supporting NVIDIA's rack-scale architectures.

It specifies 480 VAC power distribution and a higher supply temperature of 45°C for the technology cooling system loop. It also describes an IT room architecture in which clusters of AI racks share centralised networking, storage, CPU, and support racks.

The layout keeps each NVIDIA rack-scale system physically close together, while separating higher-voltage distribution for the GPU racks. Schneider Electric and NVIDIA present this as a route to larger clusters and more predictable power delivery.

The design also addresses GPU rack "operating points", including MaxP and MaxQ. The companies link MaxQ operation to higher token output per watt under power constraints, and to optimising performance through redundancy.

Validation uses ETAP models for electrical system design and ITD CFD models for data hall layout and airflow. The models assess electrical behaviour and thermal performance under defined design assumptions.

Digital twin stack

AVEVA and NVIDIA have developed a lifecycle digital twin architecture for large-scale AI factories, centred on NVIDIA Omniverse and the NVIDIA Omniverse DSX Blueprint ecosystem.

Schneider Electric plans to create "SimReady assets and digital twins" in NVIDIA Omniverse, supported by AVEVA software. Under the arrangement, AVEVA engineering and operations software is embedded across the NVIDIA Omniverse DSX Blueprint and ecosystem.

The work targets the period between design and production output, which the companies call "time-to-token". It also focuses on facility economics measured in tokens per megawatt.

In the proposed workflow, designers assemble a system architecture in Omniverse. AVEVA then runs multi-domain simulations covering power distribution, thermal dynamics, airflow performance, and controls. These simulations support iteration across different load and environmental conditions before construction begins.

Schneider Electric links the approach to fewer engineering cycles and improved deployment accuracy, and positions it as a way to reduce design risk for facilities built around large clusters of high-density compute.

"As AI workloads scale in both size and complexity, the margin for error in data center design becomes incredibly small," said Manish Kumar, Executive Vice President, Secure Power & Data Centres, Schneider Electric.

"Delivering AI at scale requires tightly integrated electrical, cooling and digital architectures that can support both unprecedented performance demands while maintaining peak energy efficiency. By combining advanced software, digital twins and validated reference designs, operators can simulate and optimize infrastructure before a single rack is deployed. This approach reduces risk, accelerates deployment and ensures the efficiency and resilience needed to power the next generation of AI factories."

NVIDIA described the joint work as a response to the infrastructure demands of larger AI deployments.

"Gigawatt-scale AI factories demand a fundamentally new class of energy-efficient and highly predictable infrastructure," said Vladimir Troy, Vice President of AI Infrastructure, NVIDIA. "Together, NVIDIA and Schneider Electric are providing the power, cooling, and digital twin architectures needed to accelerate time-to-token for our customers worldwide."

Alarm management

Schneider Electric has also tested NVIDIA's Nemotron model for agentic AI in a new alarm management service. It positions the work as addressing the challenge of interpreting alarms across multiple systems and identifying root causes in complex data centre environments.

The approach ingests streaming IoT data from multiple systems and uses software tools to analyse conditions, diagnose issues, and recommend actions. The system is designed to operate alongside expert technicians and reduce unnecessary dispatches.

The Nemotron work builds on earlier joint activity with partners focused on digital twins and modelling. Examples cited include Switch's LDC EVO operating system, presented as working with NVIDIA Omniverse libraries and OpenUSD for real-time visibility across systems.

Separately, ETAP has integrated electrical modelling into NVIDIA Omniverse. Schneider Electric, ETAP, and AVEVA have also joined the Alliance for OpenUSD, aligning on interoperable digital twin assets and simulation-ready 3D models.

Schneider Electric has backed an industry shift towards 800 VDC power architectures under an NVIDIA-led transition as rack power densities increase. It has also released reference designs tied to NVIDIA Mission Control and NVIDIA GB300 NVL72 that incorporate ETAP and EcoStruxure IT Design CFD models, focusing on simulating power and cooling scenarios for different deployments.

Further work between Schneider Electric, AVEVA, and NVIDIA will focus on validated design patterns, simulations, and operational software as data centre operators plan larger AI clusters and higher electrical and cooling loads.