Dassault Systèmes and Vertiv set modeled targets for AI factory design

Dassault Systèmes and Vertiv announced the collaboration during NVIDIA GTC Berlin, extending work on Vertiv SmartRun, a configurable virtual twin, to wider AI factory infrastructure including Vertiv OneCore. OneCore is described as a hybrid-built data-centre solution integrating critical power, thermal management and supporting infrastructure. The announcement brings together Dassault Systèmes’ systems-engineering and simulation tools with Vertiv’s power and cooling designs, with the stated aim of assessing an integrated facility before construction.

The stated objective is to reduce complexity at the construction site and shorten the path from design kickoff to productive AI capacity. The release describes AI factories as linked systems in which power, cooling, networking, controls and computing affect one another. A change in one area can have consequences elsewhere, so the partners say engineers need to assess the integrated system before deployment.

The reference designs are based on NVIDIA’s Vera Rubin DSX AI Factory reference design and Omniverse DSX Blueprint. Dassault Systèmes supplies the 3DEXPERIENCE platform, model-based systems engineering, simulation and virtual-twin capabilities; Vertiv contributes configurable SimReady power and cooling assets and infrastructure expertise. The companies say those assets can be represented as system models and adapted to customer requirements. Their collaboration expands from the SmartRun virtual twin to solutions that include OneCore, connecting facility power, thermal systems and compute planning in one design process.

In the described workflow, engineers can compare configurations, propagate design changes across disciplines, run system-level and multiphysics validation, and generate engineering deliverables from a common environment. The model is intended to carry design intent through construction and commissioning and into later technology upgrades. This is relevant because a change to power or cooling can affect other parts of a densely coupled AI facility.

The target figures refer to different stages and measures: up to 10 times faster configuration and validation; up to a 50% reduction in the defined infrastructure deployment period; roughly 10 weeks earlier capacity readiness against Vertiv’s modeled conventional baseline; and up to 30% more cumulative token output in the first 12 months from project kickoff at the same installed capacity. Vertiv says the last difference comes only from earlier facility readiness, not from improved operating efficiency. These are scenario outputs based on the model and its assumptions, not measured customer results or guaranteed savings.

The release explicitly says the token-output difference comes solely from earlier facility readiness and does not represent improved operating efficiency or a measured customer result. The comparison therefore depends on the modeled baseline, project assumptions and definition of the deployment period. The percentages and timing should be read as scenario outputs supplied by Vertiv, not as independently validated forecasts or guaranteed savings.

At NVIDIA GTC Berlin, held from 20 to 22 October, the partners said they would demonstrate elements of the process, including how facility requirements inform infrastructure configuration and how teams can assess power and cooling constraints virtually.