4 key trends shaping industrial AI as NVIDIA and Dassault Systèmes expand their partnership

NVIDIA and Dassault Systèmes announced a major expansion of their long-standing technology partnership at 3DEXPERIENCE World in Houston last week. The announcement positions industrial artificial intelligence (AI) as the next important step beyond generative AI technologies for industry.

Speaking during a press briefing ahead of the conference, executives from both companies described a shared vision for industrial AI, grounded in physics, engineering, and decades of industrial knowledge. The new industrial AI architecture is enabled through virtual twins and industry world models from Dassault Systèmes and large-scale AI infrastructure from NVIDIA.

Here are four key trends for industry emerging from the expanded NVIDIA–Dassault Systèmes partnership, highlighting how industrial AI is evolving across design, engineering, and operations.

1. Physical AI is the next evolution in industrial AI.

With the explosion of ChatGPT and other generative AI models, agentic AI has been used to build systems with agency to use as artificial assistance for knowledge work. The next major phase of AI, according to NVIDIA, will not be driven by large language models alone, but by AI systems that understand more clearly how physical systems behave. For manufacturing, this shift requires precise, physics-based digital representations of real-world assets, processes, and entire factories.

“The real value of AI is going to express itself when we apply it to the physical world in the era that is coming that we call physical AI,” said Rev Lebaredian, vice president of Omniverse and simulation technology at NVIDIA. “With physical AI grounded in the laws of physics, AI that understands the physical world and how things in the world operate, we can unlock incredible use cases.” It will be valuable in design and engineering stages, material sciences, and general robotics.

“We first have to model the world inside a computer. We need to represent the physical world accurately so that we can design, build, and operate things in the real world,” Lebaredian said. That requirement places virtual twins and engineering-accurate simulations at the center of the partnership. 

2. Virtual twins move from visualization to operational decision support.

Dassault Systèmes pioneered the concept of virtual twins nearly 40 years ago. “Our journey has always been about representing the world we live in, making visible what is invisible, and capturing knowledge and expertise for the benefit of industry and society. From molecules to medicine, from materials to satellites, from cars to airplanes, from laboratories to factories, we help engineers design almost everything in the world before they physically exist,” said Florence Hu-Aubigny, executive vice president of research & development at Dassault Systèmes. 

The 20th century was a world where industry produced objects, Hu-Aubigny said. In the 21st century, industry produces knowledge and know-how that generates objects. That industry know-how and knowledge is where Dassault Systèmes excels.

Virtual twins are more than a digital replica, said Hu-Aubigny. “It is a scientific, multidisciplinary, multi scale, virtual plus real representation, fully testable under any real condition before anything exists,” she adds.

Virtual twins integrate modeling, simulation, real-world data, and domain expertise, allowing companies to validate performance before physical assets are built, then continue testing and optimizing them once they are in operation. “The virtual twins embed modeling, simulation, data science, and deep domain expertise across the full life cycle, from conception to usage and generation,” Hu-Aubigny said.

Modeling and simulation have long had a place in industry. Now physical AI, specifically built for industrial use with 30 years of industrial data and knowledge behind it, will take it from simulation to broader industrial decision support.

World foundation models, or neural networks, are commonly used to train consumer AI systems for things like self-driving cars. World foundation models can be trained for many different purposes, but industry world models, Hu-Aubigny said, are trained specifically on decades of its data and industrial expertise and serve as the foundation for industrial AI.

“Industry world models go further. They embed the first principles of physics, engineering laws, and system constraints with four decades of industrial knowledge and know how we’ve accumulated with our clients,” she says. These AI-powered industry models combine multi-discipline modeling and simulation, spanning materials, components, machines, factories, and the entire industrial ecosystem.

That embedded knowledge enables AI systems to reason about cause and effect, constraints, and tradeoffs, which is critical for engineering, manufacturing, and asset-intensive environments.

3. Industrial AI aims to democratize advanced simulation and decision-making.

Both companies emphasized that historically high-fidelity simulation and physics-based modeling have been accessible only to a small group of specialists, and their industrial AI is intended to change that.

“What we’ve been seeing for the past few years with the introduction of these generative AI tools for general knowledge work has been really amazing,” Lebaredian said. “But largely these AIs have been restricted to things that we have in the knowledge space.”

By embedding industrial knowledge into AI systems, he said, those capabilities become broadly accessible in the industrial space. “Anybody who is building and designing anything for the real world can now have a team of assistants that has deep knowledge about how things are built to help them,” he added.

Dassault Systèmes calls these assistants virtual companions or AI-powered agents that understand intent, reason using industry world models, and orchestrate actions across design, engineering, and operations.

Hu-Aubigny said: “It will be democratized to the people who really need this knowledge and know-how when needed in the design, development, validation, commissioning, whatever phase we are.”

“The future lies in agile software-defined production systems built on modular, autonomous equipment,” she adds. 

4. AI factories are next.

The partnership also addresses the growing complexity of modern factories, culminating in AI factories, among some of the most complex systems being built today.

Lebaredian explained why simulation is no longer optional. “Building anything complex in the physical world becomes more and more difficult as the complexity increases,” he said. “The only way to really do this correctly is by simulating them at the design phase, and well before you actually do the deployment.”

Drawing from NVIDIA’s own experience, he added, “We don’t know of any other way to create something as complex as our GPUs without simulating them extensively and completely.”

The same approach, he said, applies to manufacturing systems, logistics operations, and autonomous factories. Validating behavior virtually before deployment reduces risk, improves reliability, and shortens time to operation.

With its OUTSCALE brand and the latest NVIDIA AI infrastructure, Dassault Systèmes will deploy and operate AI factories for customers on three continents. NVIDIA is adopting Dassault Systèmes’ model-based systems engineering (MBSE) to design the AI factories, starting with the NVIDIA Rubin platform and integrating into the NVIDIA Omniverse™ DSX Blueprint for large-scale AI factory deployment.

Agility and rapid iteration define the workflow, and physics-informed AI simulation models will push it from concept to production even faster. This infrastructure will largely advance biology and materials research, AI-driven design and engineering, and other autonomous production systems. Early industry partners include Bel Group, which uses the virtual twin factory to improve nutritional profiles and sustainability of their dairy formulations and packaging. For the automotive industry, Lucid will use the technology to power its vehicle and powertrain engineering.