Tech • AI • Robotics
How Dassault Systèmes Is Building AI That Understands Physics | NVIDIA AI Podcast Ep. 296
9/10TL;DR
Dassault Systèmes is advancing industrial AI through physics-based “world models” and autonomous “Virtual Companions” that simulate, design, and optimize products with scientific accuracy.
KEY POINTS
Shift to agentic industrial AI
Dassault Systèmes is transforming its 3DEXPERIENCE platform from a traditional SaaS model into an agent-as-a-service architecture, embedding AI at its core. The platform supports 400,000 customers, 45 million users, and 15 million scientists and engineers, enabling design and simulation across industries from aerospace to healthcare. This shift focuses on automating complex engineering workflows while keeping humans in control.
Virtual Twins as simulation engines
Central to the strategy are Virtual Twins, which combine real-world data with scientific modeling to create multiscale, multidisciplinary simulations. These allow products to be tested under realistic conditions before physical production. Engineers can validate performance, safety, and manufacturability entirely in a digital environment.
Industry world models grounded in science
The company’s industry world models differ from conventional generative AI by embedding physics, engineering laws, and material science. Instead of predicting outcomes from patterns alone, these systems understand causality, such as why an aircraft flies. They integrate industrial standards, regulations, and domain-specific knowledge, enabling AI to operate with technical accuracy and industry context.
Three-layer AI architecture
The system relies on three pillars: industrial knowledge (rules, standards, processes), virtual world understanding (AI operating on Virtual Twins), and industrial reasoning (agent-driven decision-making). Together, these enable AI systems to interpret complex engineering problems and generate validated solutions grounded in real-world constraints.
Virtual Companions as AI coworkers
Virtual Companions act as task-specific AI agents that translate intelligence into action. Examples include AURA (business expert), LEO (engineering specialist), and MARIE (scientific expert). These agents understand user intent, execute workflows, and ensure compliance with regulations while protecting intellectual property.
Trust, traceability, and human oversight
To mitigate risks such as hallucinations, the platform enforces human-in-the-loop decision-making at critical stages. It also introduces IP Lifecycle Management (IPLM), providing full traceability, auditability, and lineage of AI-generated outputs. This ensures transparency in how decisions and designs are produced.
Deep integration with NVIDIA technologies
The platform integrates NVIDIA AI, including NIM microservices, Omniverse, CUDA-X, and Nemotron models, to accelerate training, simulation, and inference. Performance gains include a 30% improvement in document processing and 20% better reasoning performance in AI agents. The collaboration spans over 25 years, evolving from graphics acceleration to industrial AI.
Hybrid model strategy
Dassault employs a hybrid AI approach, combining proprietary models with external ones such as Nemotron and Mistral. Model selection depends on performance, regulatory compliance, and data sovereignty, especially for sensitive industries. Techniques like fine-tuning and retrieval-augmented generation adapt models to specific industrial contexts.
Real-world use case in aerospace
A notable application involves reconstructing aircraft without original designs by generating digital parts from scans. Using LEO, engineers can automatically create optimized, manufacturable 3D components by analyzing geometry, physics, and kinematics. This dramatically accelerates reverse engineering and design validation.
Toward autonomous, self-improving systems
Future systems aim for continuous, proactive optimization, where AI agents monitor factories, supply chains, and projects in real time. Virtual Twins will act as training environments, running millions of simulations to refine solutions. This creates a closed-loop system where digital models evolve and improve autonomously.
CONCLUSION
Dassault Systèmes is positioning physics-based AI and simulation-driven agents as the foundation of industrial innovation, aiming to deliver autonomous yet controllable systems that enhance engineering precision and productivity.
Explain this