Today at SAE’s WCX 2025 event in Detroit, San Mateo, CA-based scalable engineering simulation platform startup Luminary Cloud announced what it says is the world’s first physics AI (artificial intelligence) foundation model and dataset designed for automotive SUV aerodynamic analysis. Developed in collaboration with Honda and Nvidia, the initial Shift-SUV model will enable automotive designers and engineers to perform real-time interactive aerodynamic analysis during the initial design process.

“This project represents a fundamental shift in automotive design methodology,” said Pete Schlampp, CEO at Luminary Cloud. “By providing instantaneous, physics-informed aerodynamic feedback, our open-source foundation model enables designers and engineers to collaborate more effectively at the earliest stages of development, saving time and money—while also advancing innovation.”

While there are a few open source simulation databases for sedans based on the DrivAer platform developed at the Technische Universität München’s Institute of Aerodynamics and Fluid Mechanics, Luminary Cloud says automakers lack a strong foundation model and simulation database for SUVs—the fastest-growing automotive segment globally, accounting for 48% of global sales in 2023, according to the IEA. The new foundation model, combined with Luminary’s physics AI virtual wind tunnel, addresses a critical need in the automotive industry by bridging the gap between design aesthetics and engineering performance.

“The automotive industry has long needed a solution that accelerates styling and engineering development cycles,” said Fong Loon Pan, Principal Aero Design Lead at Honda. “The Physics AI solution built with the Luminary Cloud Shift-SUV foundation model and Blender integration provides immediate insights into vehicle performance that designers and engineers need.”

The Shift-SUV aerodynamic dataset was generated by simulating several thousand geometry variants of the AeroSUV open-geometry model with Luminary’s high-fidelity CFD (computational fluid dynamics) platform. The foundational model is trained on this dataset using Nvidia’s PhysicsNeMo leveraging its DoMINO architecture for automotive external-aerodynamics—enabling Physics AI-based simulation inference.

The AI-based workflow is said to be better than traditional design approaches in that it avoids cumbersome workflow steps, providing designers with instant aerodynamic performance insights. Luminary Cloud’s model is deployed in a virtual wind tunnel, allowing designers and engineers to test designs earlier in real-world simulations.

At launch, the model is trained on about one thousand simulations, with a goal of advancing to 25,000 by the end of the year. The company will roll out new data monthly, with a staged release of pre-trained models at specific open-source dataset size milestones.

“AI and accelerated computing are revolutionizing engineering design and how products are brought to market,” said Tim Costa, Senior Director of CAE and CUDA-X at Nvidia. “Luminary’s innovative approach to develop an open-source foundation model for SUVs leveraging PhysicsNeMo and Omniverse Blueprint exemplifies the power of AI physics for automotive design.”

Luminary Cloud says that the foundation model offers several key benefits for the automotive industry, such as access to difficult-to-find or -generate open-source, industry-relevant data. The company will release both the dataset and foundation model, enabling widespread adoption and customization.

For the largest open-source physics AI SUV aerodynamics foundation model database, the high-fidelity, transient DDES (delayed detached eddy simulation) CFD simulations include various vehicle configurations, body types, and flow conditions to increase accuracy in AI model prediction. Contributions from Honda provide design variation guidance to ensure industry-relevant design parameters, along with Nvidia’s accelerated GPU computing and Omniverse Blueprint for real-time digital-twin CAE resources to accelerate simulation database generation and model training with PhysicsNeMo.

Automotive OEMs, suppliers, and research institutions that are interested in this foundation model and dataset can access it by contacting shift.luminarycloud.com.

 

Introducing real-time engineering

In 2019, the Sutter Hill Ventures-incubated Luminary was founded by Jason Lango, an expert in high-performance systems who also founded Bracket Computing, which was acquired by VMWare in 2018, and Juan Alonso, the founder of Stanford’s Aerospace Design Laboratory and a former director of NASA’s aeronautics research programs. Alongside a team of experts in physics, numerical methods, GPU and cloud computing, and machine learning/AI, Lango and Alonso were on a mission to democratize and modernize high-performance computing for engineering and science.

“Jason and Juan are bringing the power of GPUs and the elasticity of the cloud to one of the most complex engineering functions,” said Mike Speiser, a Managing Director of Sutter Hill Ventures who serves on the board of Luminary Cloud.

With Sutter Hill Ventures leading its $115 million funding, Luminary Cloud launched out of stealth in March 2024 with its CAE (computer-aided engineering) SaaS (software as a service) platform. The platform aims to empower smarter and faster design cycles and allow engineers to develop better products in a fraction of the time.

“While software engineering has become more agile thanks to advances in cloud technologies, physical engineering hasn’t kept pace, despite increasing pressure to deliver advanced products faster and more efficiently,” said Lango, former CEO and now Board Member and Advisor of Luminary Cloud. “Our platform empowers R&D teams to analyze, design, and innovate faster than they ever could before, which is crucial when facing strong and globally competitive markets that demand better products in shorter time frames.”

Digital simulations are a crucial step of the R&D process as they allow engineers to create virtual prototypes and understand how they perform in realistic environments—including considerations such as air and water flows, temperature and pressure distributions, and aerodynamic drag—and make modifications and improvements long before a physical product comes to market. However, today’s CAE solutions aren’t delivering the rapid iteration needed to meet market demands, consumer expectations, and regulatory pressures, claims the company. Instead, most engineers use legacy software that runs in on-premise infrastructure—which can be challenging and expensive to scale—resulting in a slow design process where each simulation takes days or weeks.

Luminary’s platform makes it possible to run high-fidelity simulations 100 times faster than legacy vendors by leveraging the raw speed of GPU- and cloud-based processing. Its proprietary simulation platform is powered by massively parallel Nvidia GPU clusters in the cloud. With hyper-fast and accurate simulations, engineers can iterate and test a variety of scenarios, answer more questions, and use the insights to optimize product design.

The company makes the end-to-end engineering workflow easier and faster via a modern web-based user experience. Enhanced by a Lumi AI-based engineering design copilot, it cuts down the time engineers spend in setup and simulation so they can spend more time analyzing and optimizing.

 

Aero customers lead the way

Luminary’s customer base spans industries, but key earlier adopters have been aerospace companies like Joby Aviation, Rune Aero, nTop, Sceye, and Piper Aircraft. Others include Trek Bikes and Cobra Golf, a subsidiary of Puma.

“With Luminary, we can quickly distinguish between ideas we should pursue and those we should abandon,” said Gregor Mikić, Flight Research and Physics Lead and Chief Aerodynamicist at Joby Aviation. “We are able to take complete aircraft configurations and run them in complex simulations in a matter of minutes, saving a significant amount of time.”

Last month, during Nvidia’s GTC, Rune Aero announced it had upgraded its aircraft development process using a Luminary Cloud-powered interactive virtual wind tunnel. The autonomous aircraft startup for the growing middle-mile air cargo market says its virtual wind tunnel adoption enabled its aerospace design engineers to quickly analyze many aircraft design prototypes, iterate designs, and visualize the impact in real time. It dramatically reduces future technical risk and costs compared to relying solely on conventional physical wind tunnel tests.

“At the early stages of aircraft development, getting fast, accurate, and cost-effective design feedback is essential,” said Nadine Auda, Co-founder of Rune Aero. “Luminary’s virtual wind tunnel allows Rune Aero to test configurations earlier in the process, reducing our early development costs by over 80% compared to traditional wind tunnels and enabling faster, smarter design decisions. Rune’s optimized aircraft configuration resulting from these tests improves aerodynamic lift-to-drag ratio, combined with advanced propulsion, this doubles payload and range while cutting fuel burn by 50%, ultimately lowering operating costs for our customers.”

Also during GTC, computational design startup nTop announced that by connecting its parametric geometry generation, Luminary’s GPU-native simulation and simulation management platform, and the PhysicsNeMo via NIM microservices, its engineers can now create and analyze hundreds of design variations in a single day—a process that previously took weeks to months of manual effort across disconnected systems.

“Physics simulation has long been a critical bottleneck in real-time design optimization,” said Brad Rothenberg, CEO and founder at nTop. “The integration between Luminary and nTop, powered by NVIDIA hardware, brings us significantly closer to solving this challenge. Through Luminary’s API, we can seamlessly push nTop geometries to Luminary for physics calculations and automatically return results to nTop in a robust, continuous loop. This makes it now possible, and actually easy, to train up physics-based AI models used to accelerate performance predictions; we’re now closer than ever to real-time design optimization.”