Nvidia and Schneider Electric are testing a digital twin for AI data center operations.

Nvidia and Schneider Electric are testing digital-twin technology that connects AI workloads with data center power and cooling models, allowing operators to evaluate infrastructure changes before taking them live.

The pilot brings together AVEVA, an industrial engineering and operations software company; Electrical Transient and Analysis Program (ETAP), which develops software for modeling and simulating electrical power systems; Nscale, an AI infrastructure and cloud provider; and Start Campus, developer of the SINES data center campus in Portugal. The project focuses on Nscale's operations at SIN01, a data center on the SINES campus.

Built using Nvidia's Omniverse DSX Blueprint, a framework for creating digital twins of AI data centers, the system combines facility engineering and operational data with power and cooling simulations. Operators can use the models to evaluate how infrastructure changes could affect AI workloads before making physical modifications.

For operators, the challenge is determining whether power and cooling systems can support denser AI hardware without costly redesigns, deployment delays or stranded capacity. Digital twins could help evaluate configurations before operators commit to equipment purchases or physical modifications.

"The nearer-term value is lower commissioning risk and faster deployment," said Stephen Sopko, practice lead for semiconductor and deep-tech research at HyperFrame Research.

**Digital twins target deployment risk**

Digital twins can help operators rehearse configurations and identify problems before hardware arrives, potentially shortening validation cycles and reducing physical rework, Sopko said.

The SINES pilot extends the approach to facility engineering by connecting AI workload requirements to electrical distribution and liquid-cooling models. Operators could use those simulations to evaluate the effects of higher rack density before making physical changes.

CoreWeave provides a separate example of this kind of pre-deployment approach. The company uses Nvidia's DSX Air to simulate AI factory environments and validate infrastructure configurations before physical systems arrive. The cloud-based platform models computing, networking, storage and software orchestration.

That case illustrates how simulation can help operators validate configurations before deployment. Public disclosures do not establish that CoreWeave uses DSX Air to control power and cooling systems in a live facility.

**Linking workloads to facility systems**

Schneider Electric calls the pilot system Immersive Twin. It combines the company's energy management technology with AVEVA industrial software, ETAP electrical engineering tools and third-party technologies.

The system uses OpenUSD, a framework for describing and exchanging 3D data, to connect digital models and assets. The companies intend to support the AI factory lifecycle, from design through operations, and make it easier to integrate equipment and models from different vendors.

Interoperability could help operators connect facility engineering tools with AI infrastructure models, Sopko said. Full bidirectional closed-loop control, however, remains a longer-term goal.

Proving that digital twins improve AI computing efficiency will require more than simulation results. Operators should compare time to first token, commissioning defects, power usage effectiveness (PUE), thermal events and measured computing output against a baseline using comparable workloads and hardware, Sopko said.

The companies have not published results showing that the SINES pilot has improved token throughput, energy consumption, cooling efficiency or deployment time. Nor have they disclosed validation results comparing the digital twin's predictions with measured operating conditions.

**Portugal campus targets next-generation AI**

The pilot focuses on SIN01, part of Start Campus's SINES Data Campus, a development designed for up to 1.2 gigawatts of capacity.

The companies said the campus is expected to house more than 66,000 Nvidia Rubin GPUs. They did not provide a deployment schedule or specify how many Rubin GPUs are currently installed at SIN01.

The project builds on collaboration among Schneider Electric, AVEVA, ETAP and Nvidia, including work through the Alliance for OpenUSD on digital twins and AI factory simulation.

For operators planning high-density AI deployments, digital twins offer a way to test infrastructure configurations before committing capital or modifying a live facility. Whether the SINES pilot delivers measurable gains in efficiency, reliability or deployment speed remains an open question.