AI-powered vehicle design refers to the use of artificial intelligence models, specifically physics-based neural networks, to predict and simulate the aerodynamic behavior of high-performance vehicles. Rather than relying solely on traditional computational methods that can take hours or even days to evaluate a single design change, AI surrogate models can approximate those same results in seconds. When paired with emerging quantum computing capabilities, this approach has the potential to reshape how engineers develop everything from race cars to commercial aircraft. On April 30, 2026, IBM and Italian race car manufacturer Dallara announced a collaboration that puts this concept into practice, combining IBM’s expertise in AI and quantum computing with Dallara’s five decades of motorsport engineering.
In this article, we’ll discuss what this partnership entails, how the technology works, and why it matters beyond the racetrack. We’ll look at the specific AI model IBM developed for the project, called the Gauge-Invariant Spectral Transformer (GIST), and break down how it was tested against traditional aerodynamic simulation methods using Dallara’s proprietary race car data. We’ll also explore the role quantum computing could play in the future of vehicle design, and why a collaboration between a tech giant and a legendary chassis builder could signal broader changes for the automotive and aerospace industries.
TL;DR Snapshot
IBM and the Dallara Group have announced a collaboration to develop physics-based AI foundation models that can dramatically accelerate the aerodynamic design process for high-performance vehicles. In early testing, their AI model completed evaluations in roughly 10 seconds that traditionally took hours using computational fluid dynamics (CFD), while identifying the same optimal designs with comparable accuracy. The partnership also lays the groundwork for integrating quantum computing into vehicle design workflows in the future.
Key takeaways include…
- AI models trained on Dallara’s proprietary aerodynamic simulation data can reduce design evaluation time from hours to seconds, potentially cutting full design exploration workflows from days to minutes and allowing engineers to test far more configurations early in the development process.
- IBM’s new GIST architecture solves a key technical challenge in applying AI to mesh-based physics simulations, encoding not just the points of a 3D model but also how those points connect, which is critical for accurately predicting forces on finely detailed race car components.
- The collaboration is exploring quantum and hybrid quantum-classical computing as a next step, aiming to tackle aerodynamic simulation problems that remain difficult even for today’s most powerful classical systems.
Who should read this: Engineers, automotive enthusiasts, AI researchers, motorsport fans, and anyone interested in the intersection of advanced computing and physical design.
Why Race Cars Are the Perfect Testbed for AI-Powered Design
High-performance motorsport is an environment where tiny engineering margins translate into enormous competitive advantages. As noted in IBM and Dallara’s joint announcement, Dallara designs and supplies chassis for some of the world’s most demanding racing series, including IndyCar (where average track speeds can exceed 230 mph), Formula 2, Formula 3, Super Formula, and Indy NXT, and contributes to top-tier programs like Formula E, WEC, and IMSA. This breadth gives Dallara a unique ability to validate simulation results against real-world vehicle performance across a wide variety of conditions.
Engineers in these programs rely heavily on computational fluid dynamics (CFD) to predict how subtle geometry changes affect forces like downforce and drag. CFD is a simulation method that solves complex mathematical equations describing how air flows around and interacts with a vehicle’s surfaces. It’s incredibly powerful, but also computationally expensive. As the IBM Research blog post on the collaboration explains, aerodynamic engineers can spend several hours evaluating a narrow surface adjustment, and weeks to months designing an entire car as they iterate through geometry changes, operating conditions, and performance tradeoffs.
This creates a bottleneck. The more configurations an engineer can test, the more likely they are to find an optimal design. But when each test takes hours, the number of configurations that can realistically be explored is severely limited. That’s exactly the kind of problem where AI can make a meaningful difference, by “learning” the physics well enough to deliver fast, accurate approximations that let engineers explore the design space more broadly before committing to expensive, full-fidelity simulations.
How IBM’s GIST Model Works
At the technical core of this collaboration is a new AI architecture developed by IBM Research called the Gauge-Invariant Spectral Transformer, or GIST. According to IBM’s preprint study, GIST is a graph-based neural operator designed to process the complex 3D mesh data that underpins modern vehicle design.
Most physical products today, from eyeglasses to airplanes, are designed using 3D meshes: dense networks of points, links, and surfaces that define a product’s shape and contours. Previous AI approaches for predicting aerodynamic forces treated these meshes as simple point clouds, essentially ignoring the connections between points. As the IBM Research blog notes, this approach might work for the geometry of a typical passenger car, but race cars feature finely detailed aerodynamic components where even the smallest design element can have an outsize effect on performance.
GIST addresses this by encoding both the coordinates of mesh points and their connecting links, capturing the full topology of the surface. This matters because points on opposite sides of a thin surface may be physically close together but experience completely different aerodynamic forces. Without understanding the connectivity, an AI model can’t distinguish between these scenarios.
To manage the computational cost of processing such complex graph data, IBM’s researchers used random projections, a compressed sensing technique, to generate efficient graph embeddings. They also designed the architecture to be “gauge-invariant,” a concept borrowed from physics meaning that the model’s predictions remain consistent regardless of the arbitrary mathematical choices used to represent the data. As IBM researcher Mattia Rigotti explained in the IBM Research blog, coordinate choices are like gauge choices, while physical relationships are gauge-invariant. The team carried that insight directly into the model’s design.
In practical testing, IBM and Dallara trained the GIST model on Dallara’s proprietary simulation data for a conceptual Le Mans Prototype 2 (LMP2)-like race car. They compared the model’s predictions against traditional CFD across multiple configurations of the rear diffuser, the underfloor component that helps generate efficient downforce. According to the joint press release, the traditional CFD approach took a few hours to calculate all configurations, while the AI model completed the same evaluations in about 10 seconds, identifying the same optimal design with roughly the same error margins. Applied to a typical complete set of hundreds of geometry configurations, this speedup could cut days of work down to minutes.
The Quantum Computing Horizon
While AI is delivering results today, IBM and Dallara are also looking further ahead to quantum computing. According to the their announcement, the collaboration will evaluate where quantum and hybrid quantum-classical approaches can complement traditional simulation workflows in the near term, while identifying longer-term opportunities for practical use in automotive and motorsport design.
Quantum computing excels at certain types of problems that are extraordinarily difficult for classical computers, particularly those involving the simulation of complex physical systems with many interacting variables. Aerodynamic design, where turbulent airflows interact with intricate vehicle geometries across constantly changing conditions, fits this profile well. While quantum computing isn’t yet mature enough to handle these problems at production scale, the partnership positions both companies to be ready as the technology progresses.
The combination of AI and quantum computing is where things get especially interesting. AI surrogate models can handle the rapid screening of design configurations today, while quantum-enhanced simulations could eventually provide even higher fidelity results for the most complex and difficult-to-model scenarios, creating a layered workflow where each technology handles the tasks it’s best suited for.
Beyond the Racetrack: Broader Implications
It’s tempting to view this partnership as a niche motorsport story, but the implications extend well beyond racing. As Dallara CIO Fabrizio Arbucci noted in the press release, more efficient designs could benefit all transport categories, from passenger vehicles to aircraft, and even other industries affected by aerodynamics. He pointed out that even a one to two percent reduction in drag across passenger vehicles could add up to meaningful fuel-efficiency gains at scale.
This is the kind of cascading impact that makes the collaboration noteworthy for people outside of motorsport. The same AI models being tested on race car diffusers could, in principle, be applied to optimize the shape of commercial trucks, improve the aerodynamic efficiency of wind turbine blades, or refine the design of next-generation aircraft. Any industry where physical simulation is a bottleneck in the design process stands to benefit from faster, AI-accelerated workflows.
The initial results from the partnership have been documented in a study titled “Faster by Design,” published on arXiv. IBM and Dallara also presented their findings at the AI & PDE Workshop during the International Conference on Learning Representations (ICLR) 2026 in Rio de Janeiro.
Frequently Asked Questions
IBM (International Business Machines Corporation) is an American multinational technology company headquartered in Armonk, New York. Founded in 1911, IBM is one of the world’s largest technology firms, providing services in hybrid cloud computing, artificial intelligence, consulting, and enterprise software. The company’s research division, IBM Research, has been at the forefront of breakthroughs in AI, quantum computing, and physical sciences, developing models that predict weather patterns, flood extents, solar dynamics, and now aerodynamic forces on vehicles.
Dallara is an Italian racing car manufacturer founded by engineer Giampaolo Dallara in 1972, headquartered near Parma, Italy. The company is the sole chassis supplier for several major racing series, including IndyCar, Indy NXT, Formula 2, Formula 3, and Super Formula. It also contributes to programs in Formula E, the WEC, and IMSA. Beyond motorsport, Dallara applies its engineering expertise to high-performance road vehicles, like the Dallara Stradale, and aerospace projects.
Gauge-Invariant Spectral Transformer is a new AI architecture developed by IBM Research for processing graph-structured simulation data. Unlike previous approaches that treated 3D vehicle models as simple collections of points, GIST encodes both the points and their connections, capturing the full surface topology. The “gauge-invariant” design ensures the model produces consistent predictions regardless of how the underlying data is mathematically represented.
CFD is a branch of engineering simulation that uses mathematical equations and computer algorithms to predict how fluids (like air or water) flow around and interact with physical objects. In vehicle design, engineers use CFD to simulate aerodynamic forces such as downforce and drag without needing to build a physical prototype. While highly accurate, CFD simulations are computationally intensive and can take hours to days to complete, depending on the complexity of the analysis.
A neural surrogate model is an AI system trained to approximate the results of a more expensive computational process, like a CFD simulation. Once trained on a dataset of simulation results, the surrogate can predict outcomes for new configurations much faster than running the full simulation. In the IBM-Dallara collaboration, the surrogate model completed aerodynamic evaluations in about 10 seconds that took traditional CFD several hours.
Quantum computing is still in its early stages within this partnership. IBM and Dallara are exploring how quantum and hybrid quantum-classical approaches might eventually enhance simulation fidelity for complex aerodynamic problems that push the limits of classical computing. While not yet producing production-ready results, this exploration positions both companies to integrate quantum methods into vehicle design workflows as the technology matures.
The International Conference on Learning Representations is one of the most prestigious academic conferences in the field of machine learning and artificial intelligence. IBM and Dallara presented their research findings at the AI & PDE Workshop during ICLR 2026, which took place in Rio de Janeiro, Brazil.
Other Enterprise AI Articles You May Be Interested In
Nvidia Nemotron 3 Nano Omni: A Unified Open Model for Vision, Audio, and Text
NetSuite SuiteConnect 2026: New AI Coding Agent Skills for SuiteCloud Developers
Meta’s $2B Manus AI Acquisition Blocked: What Founders, VCs, and Tech Leaders Need to Know
DeepSeek V4 Changes the AI Pricing Game: Full Breakdown and Analysis
