Dassault Systèmes and Vertiv team up to simulate AI data centers before construction

Dassault Systèmes and Vertiv announced an October 22, 2026 collaboration to configure and test AI data-center infrastructure virtually before construction begins. The partnership combines engineering software with power and cooling expertise for NVIDIA DSX AI factories. Its promised benefit is a shorter path to usable computing capacity, though the projected gains remain modeled rather than measured customer results.

Testing the whole facility, not isolated parts

The companies describe AI factories as tightly connected systems: power, cooling, networking, controls and computing must be engineered together. A change in one area can affect the others. Their approach aims to expose those consequences in a virtual twin—a digital representation used to simulate the infrastructure—before teams encounter them during physical deployment.

Dassault Systèmes supplies its 3DEXPERIENCE platform, which combines systems engineering, simulation and virtual-twin capabilities. Vertiv contributes its infrastructure expertise and pre-engineered power and cooling designs. The platform can represent that equipment as one integrated digital system, allowing engineers to track design decisions and see how changes affect connected parts.

The collaboration builds on Vertiv’s SmartRun configurable virtual twin and extends the approach to a wider set of its AI infrastructure offerings. That includes OneCore, a hybrid-built data-center solution combining critical power, thermal management and supporting infrastructure. The scope therefore reaches beyond a single power or cooling component.

A configurable model instead of a fixed design

The starting material is Vertiv’s SimReady power and cooling assets, turned into configurable system models. Customer-specific configurations are based on the NVIDIA Vera Rubin DSX AI Factory reference design and NVIDIA Omniverse DSX Blueprint. Those provide the design framework; the collaboration supplies tools to evaluate and validate infrastructure configurations within it.

Rather than produce one fixed design, the companies describe a shared engineering environment that supports several connected tasks:

• Compare configurations and carry design changes across engineering disciplines, so teams can examine their effects across the infrastructure rather than only within one component.
• Validate the complete system and the interaction of different physical effects, including the power and cooling constraints that shape a facility’s configuration.
• Generate engineering deliverables from the same environment, with the stated aim of preserving design intent through construction, commissioning and later technology upgrades.

Earlier opening, not faster operation

Vertiv’s model estimates that the combined approach could bring capacity online approximately 10 weeks sooner than a modeled conventional baseline. At the same installed capacity, it projects up to 30% more cumulative token output—the amount of AI output generated—within 12 months of project kickoff. Both figures depend on the model’s assumptions.

That is a scheduling claim, not a computing-efficiency claim. The projected output gain comes solely from having the facility ready earlier, giving the same installed capacity more time to work within the comparison period. The release explicitly says it does not reflect improved operating efficiency or a measured customer result.

For NVIDIA GTC Berlin, held October 20–22, the companies planned a demonstration connecting facility requirements to infrastructure configuration. The showcase was intended to let teams evaluate power and cooling constraints virtually and show how Vertiv’s factory-built infrastructure supports physical deployment. It brings the software models and infrastructure assets together as an engineering workflow.