The development of AI systems is currently progressing faster than the expansion of the necessary physical infrastructure. While computing power is growing in ever shorter cycles, data center operators are faced with the challenge of adapting power supply, cooling and control systems accordingly. The infrastructure provider Vertiv now wants to close this gap with digital twins.

The company has announced progress on a production-ready digital twin feature for Vertiv SmartRun. The solution will be integrated into the Nvidia Omniverse DSX Blueprint and is designed to help make AI factory infrastructure more configurable, repeatable and simulation-ready.

Planning as a complete digital system

With the increasing power densities of AI applications, traditional planning processes are increasingly reaching their limits, according to Vertiv. Until now, teams for power supply, cooling, control and implementation have often worked with separate documentation and handover processes.

The 'Vertiv SmartRun Digital Twin' is intended to replace this approach with a model-based process. Infrastructure can be designed, simulated and validated as a coherent system before construction. Configurations and dependencies are mapped in a virtual environment so that potential problems become visible at an earlier stage.

According to the company, this reduces design changes in late project phases and integration risks. At the same time, the time to operational readiness is to be shortened and collaboration between different specialist departments improved.

"AI infrastructure can no longer be planned generation by generation," says Scott Armul, Chief Product and Technology Officer at Vertiv. "To deliver more tokens per second per megawatt, power, cooling, control and deployment workflows must be designed as a single, interlocking system. The Vertiv SmartRun Digital Twin helps translate Vertiv's infrastructure expertise into configurable, simulation-ready building blocks that enable faster and safer AI factory planning. By extending this approach to Vertiv OneCore Rubin DSX, Vertiv is helping customers translate future compute requirements into deployable physical infrastructure before those requirements reach full deployment scale."

Part of a larger roadmap

According to the company, SmartRun's digital twin is the first step in a multi-stage strategy for digital twins of AI factories. The aim is to maintain the technical logic of a system throughout its entire life cycle - from configuration and simulation to commissioning and subsequent optimization.

In future, digital models will not only be used as planning tools, but will also form the basis for the operation and further development of complex infrastructure.

Nvidia relies on full-stack co-design

Integration into Nvidia Omniverse DSX is a central component of the concept. The platform is intended to enable digital twins of AI factories on a gigawatt scale and bring together different simulation models.

"AI factories require full-stack co-design across compute and physical infrastructure," said Vladimir Troy, Vice President of AI Infrastructure at Nvidia . "Nvidia Omniverse DSX Blueprint helps the ecosystem create, simulate and optimize digital twins for gigawatt-scale AI factories using OpenUSD, SimReady assets, and power, thermal and operational simulations. Incorporating Vertiv SmartRun into this workflow can help customers evaluate infrastructure decisions earlier and prepare for multiple generations of accelerated computing power."

Demonstration at Computex

Vertiv wants to show what such an approach can look like in practice at Computex Taipei 2026. SmartRun will be presented there both as a physical infrastructure system and as a configurable digital twin.

The demonstrator is based on Dassault Systèmes' model-based system engineering capabilities on the '3DEXPERIENCE' platform and is connected to Nvidia Omniverse DSX workflows. The goal is a common digital foundation for configuration, simulation, validation and subsequent optimization of the infrastructure.

"Digital twins make it possible to represent complex infrastructure systems with the intelligence of their configuration rules, dependencies and technical intent," said Stéphane Sireau, vice president of the high-tech industry at Dassault Systèmes. "At Computex, Vertiv, Dassault and Nvidia will demonstrate how Vertiv's AI factory infrastructure is evolving from document-based design workflows to an industrialized, model-based systems engineering approach optimized for speed, quality and system-level performance."

Infrastructure becomes a bottleneck

The push shows how much attention in the AI industry is now shifting to the physical infrastructure. While new generations of chips and AI models regularly make the headlines, the provision of energy, cooling and data center capacity is increasingly becoming a decisive factor. Digital twins should help to plan and test these systems more quickly and adapt them to the growing requirements of future AI factories.