IBM and Dallara expand investments in Simulation as the Key to Physical AI

According to a press release IBM has partnered with Dallara Group (a world-leading manufacturer specializing in the design, engineering, and production of racing cars for top-tier motorsports.) to develop physics-based AI models aimed at accelerating high-performance vehicle design — and more broadly, redefining how machines learn in the physical world. Or in the words of Alessandro Curioni IBM Fellow and VP, Algorithms and Applications, IBM Research.: "Some of the hardest engineering challenges come down to accurately simulating the physical world,"

At the core of the collaboration is a familiar bottleneck: AI training

Engineering teams today rely heavily on computational fluid dynamics (CFD) to model aerodynamics. These simulations are accurate but expensive, often taking hours for narrow analyses and weeks across full design cycles. IBM’s approach introduces domain-specific AI models trained on high-fidelity simulation data to dramatically compress these timelines.

In early results, AI models evaluated aerodynamic configurations in seconds instead of hours, identifying optimal designs with comparable accuracy to traditional methods.

The implication is not just faster workflows, but a fundamental shift: engineers can now explore far more design permutations earlier in the process, reserving expensive simulations for final validation.

Looking ahead, IBM and Dallara are also exploring how quantum computing could further increase simulation fidelity for complex physical systems.

TheMarketAI Take

This fits squarely into a pattern we’ve been covering: the future of physical AI depends on simulation.

Unlike software AI, which can train on the internet, physical AI faces a severe data constraint. Real-world data is expensive, slow to generate, and often limited in scope. That makes simulation not just helpful — but essential.

NVIDIA has pushed this approach aggressively with robotics. Now IBM is applying the same principle to physics-heavy domains like aerodynamics.

The logic is straightforward:

• The more realistic the simulation

• The more scalable the training

• The better the real-world performance

But the challenge remains the same: the sim-to-real gap.

Even the best simulations struggle to capture the full complexity of the physical world. Small inaccuracies can compound, especially in high-precision environments like vehicle dynamics or robotics.

What IBM and Dallara are showing is a hybrid path forward. Not replacing physics-based simulation, but augmenting it with AI to accelerate iteration.

Physical AI remains one of the hardest frontiers in technology.But increasingly, the answer looks less like better models — and more like better worlds to train them in.