Nvidia and Dassault Systèmes combine digital twins and AI in industry world models

Nvidia is putting its AI infrastructure, models, and libraries together with virtual twin technology from 3D design software vendor Dassault Systèmes to build a shared architecture the companies describe as “science-validated industry world models.”

It’s a deepening of their existing collaboration with “a shared long-term vision for how industrial AI will be built, validated and deployed at scale,” the companies said.

Nvidia is multiplying its partnerships with developers of industrial design software. It recently expanded the integration options for its Omniverse development platform farther into physical AI applications, building the Omniverse libraries into Siemens Digital Twin Composer as well as integrating into products and services from Accenture, Microsoft, Ansys, Cadence, and others.

World models

Dassault Systèmes CEO Pascal Daloz talked up the partnership’s potential for manufacturing at his company’s recent 3DExperience World conference, saying that, today, “Industry is producing knowledge and know-how, and the knowledge and know-how are generating the objects.”

This s why Dassault Systèmes and Nvidia are bringing together virtual twins and 3D universes, he said: “3D universes are not applications, they are knowledge factories, factories where the knowledge is enriched. The know-how scales and the results are trusted.”

The companies are using AI to supercharge this enrichment process, he said — but much more than generative AI: Their approach uses “real-world AI, grounded in industry, engineering, and science. So it’s not only about large language models, it’s about what we call world models.”

LLMs don’t build satellites, design aircraft, or discover cancer therapeutics, he said, people do: “So the world model makes the virtual twin truly generative.”

As part of the partnership, Nvidia is adopting Dassault Systèmes model-based systems engineering (MBSE) to design its AI factories, and Dassault Systèmes will use Nvidia’s open models and libraries and run its AI factories on Nvidia infrastructure.

Everything will be software defined with real-time simulations powered by Nvidia, said Nvidia CEO Jensen Huang, who also spoke at 3DExperience World.

Nvidia currently designs and simulates its own data centers with Dassault Systèmes MBSE, and, he said, even runs the network and the supercomputers inside the virtual twin before breaking ground.

“That allows us to save tons of time and tons of money, and over time, of course, this data center has an AI that keeps it optimal,” Huang said. “So we’re going to have virtual twins of these AI factories running forever, training and updating our models.”

Functional partnership

The process of simulating a factory in a virtual twin is “extraordinarily complex,” noted Bill Curtis, IoT and edge analyst in residence at Moor Insights & Strategy. “Nvidia and Dassault Systèmes are defining an industrial physical AI computing stack: a shared, physics-validated digital-twin world model running on large-scale, AI-accelerated Nvidia infrastructure. This stack enables customers to design, validate, and operate factories, products, and even biological systems in virtual space before and after they exist physically.”

And, he said, “Nvidia’s adoption of Dassault’s software to design its own AI factories shows this is a functional partnership, not just marketing. Dassault is the engine for representing the physical world, and Nvidia is the infrastructure that it runs on.”

Info-Tech Research Group advisory fellow Scott Bickley said the partnership signals a shift from a generative or data-driven approach to digital twins to “one where digital twins are grounded in the laws of physical science, which can now be harnessed through the power of Nvidia’s accelerated AI Rubin compute and Omniverse stack. The ability to scale models that are validated and anchored by the laws of physics and biology opens the door to potential blockbuster discoveries across biology and materials research.”