The short version: At NVIDIA GTC 2026, Schneider Electric announced tools built with NVIDIA and AVEVA for designing and running AI data centers: a validated power and cooling reference design for NVIDIA’s newest racks, digital twins for testing a facility before it is built, and an AI system that helps operators trace alarms to their cause.
The three pieces
- A reference design for Vera Rubin NVL72 racks. It covers power and cooling for NVIDIA’s rack-scale systems, including 480 VAC power distribution, a 45°C supply temperature on the technology cooling loop, and an IT room laid out around clusters of AI racks. Schneider validated it with ETAP models for the electrical design and CFD models for layout and airflow.
- Digital twins. The partners are using simulation to test designs before deployment, which matters when one mistake in a high-density hall can delay a multi-hundred-megawatt project.
- AI for operations. A system that reads alarms across multiple building systems, pinpoints likely root causes and recommends fixes using live IoT data.
Why it matters
Data Centre Review noted the move puts Schneider much earlier in the design process and deeper into daily operations than a hardware supplier usually sits. That reflects how fast AI campuses are going up and how little room there is for redesign. Liquid cooling at these temperatures and power densities is still new to many design teams, so validated designs and simulation shorten the learning curve.
What changes in the day-to-day work
AI data centers are pushing rack densities far past what most facilities were designed for, and almost all of the newest racks need liquid cooling. That puts pressure on every discipline on the project. Reference designs and simulation do not remove that pressure, but they change where engineers spend their time.
- Electrical engineers start from a validated power architecture instead of a blank page, then spend their effort on utility coordination, short-circuit and arc flash studies, and adapting the design to the site’s service and equipment lead times.
- Mechanical engineers work from a tested cooling concept but still size the coolant distribution, heat rejection and piping for local climate, water limits and redundancy targets.
- Design teams can run layout and airflow scenarios in a digital twin before steel goes up, which catches clashes and hot spots when they are cheap to fix.
- Operations staff get help sorting a flood of alarms down to a likely root cause, which matters when one cooling fault can take down millions of dollars of GPUs.
The common thread is that AI and simulation move effort earlier in the project and shift it from drafting toward analysis and judgment. The engineers who adapt fastest are the ones comfortable working from models and data rather than marked-up drawings.
Who should pay attention
Hyperscale and colocation developers, MEP design firms that work on mission-critical projects, general contractors building AI campuses, and operators converting older halls for high-density racks. Our US data center map tracks where many of these projects are going up.
Why reference designs matter right now
Every new generation of AI hardware has raised the power and cooling bar, and design teams have had to relearn the basics each time. A validated reference design gives owners and engineers a known starting point that has already been checked with electrical and airflow models, instead of every firm working out the same problems on its own.
That speed matters because the bottleneck on many AI campuses is not land or money but people and equipment. Utility interconnections, switchgear and cooling plant all have long lead times, and there are not enough experienced mission-critical engineers to go around. Anything that cuts design rework frees those engineers for the site-specific problems that a reference design cannot solve. Our data center workforce shortage statistics cover the wider picture.
What it means for hiring
Reference designs speed up projects, but every site still needs engineers who can adapt them to local utilities, codes and equipment.
- Electrical engineers with ETAP and medium-voltage distribution experience are in heavy demand on AI campuses.
- Mechanical engineers who have designed liquid cooling, CDUs and direct-to-chip systems are among the hardest data center roles to fill.
- Controls and critical facilities engineers who can work with AI-driven monitoring will run these sites once they open.
Building a data center team? See our data center recruiting page, our guide on how to hire data center engineers and the US data center map.
Interview questions for data center engineers
- How would you adapt a vendor reference design to a site with a different utility service and climate?
- What studies do you run in ETAP or a similar tool before issuing an electrical design for a high-density hall?
- Describe a liquid cooling system you designed or commissioned. What redundancy did you build in?
- How have you used CFD or a digital twin to change a layout before construction?
Frequently asked questions
What did Schneider Electric announce with NVIDIA and AVEVA?
At NVIDIA GTC 2026, Schneider Electric announced a validated reference design for NVIDIA Vera Rubin NVL72 racks, digital twin work for testing designs before deployment, and an AI system that helps operators trace alarms to root causes.
Why do AI data centers need liquid cooling?
AI racks pack far more power into each rack than traditional servers, and air alone cannot carry that much heat away efficiently. Liquid cooling moves heat directly from the chips into a coolant loop.
Which data center engineers are hardest to hire?
In our experience, mechanical engineers with liquid cooling design experience and electrical engineers who have designed medium-voltage systems for large campuses are the toughest searches, followed by commissioning and critical facilities engineers.
More stories like this are on our AI for Engineers hub.
