The impressive growth of digital innovators like Google, Salesforce, Uber, and Airbnb is sometimes attributed to modern agile software development practices, allowing them to rapidly experiment with and continuously update their products, services, and features at scale. However, applying these same approaches to makers of physical products like planes, electric cars, and manufacturing equipment has been much more challenging.
One essential challenge is that the data about physical products is more nuanced, making the testing and feedback process more complex. Agile developers deploying test-driven development approaches for new software can often write a simple test first to help focus their programming efforts and ensure the code works as intended. This is more complex with physical products that have to contend with the vagaries of the real world influenced by multiple physical phenomena.]
Our own recent coverage on How Digital Twins Get Faster and More Connected focused on the role of NVIDIA's GPU-accelerated simulation in testing physical product designs faster. I recently connected with Thomas Von Tschammer, co-founder and Managing Director of Neural Concept US, who argues it will also usher in an age of more agile engineering practices:
We're seeing more and more traditional OEMs across automotive, aerospace and micro-electronics adopt innovative AI tools by moving towards more agile approaches for product design. Companies increasingly recognize the need for tools that integrate seamlessly rather than relying on siloed, proprietary systems.
Customers including Subaru, Mitsubishi Chemical Group, Airbus, Bosch, GE Renewable Energy, and LG Electronics are all starting to use the tech to streamline product development and innovation. Neural Concept offers a 3D AI and integration platform to interconnect different simulation tools and workflows developed by PLM and simulation vendors like Siemens, Dassault, and PTC. It works with these tools to fine-tune and run the appropriate AI surrogate that accelerates simulation. Omniverse Blueprint helps speed processing at the hardware level.
One of the biggest challenges legacy firms face is figuring out how to modernize their designs for traditional products to take advantage of new technologies or adapt to environmental constraints. Take cars, for example. The essential components in a modern gas car have seen incremental improvements of similar components like motors, alternators, carburetors, and mechanical brakes, further bolstered by electronic controls. Over the years, the simulation and testing tools co-evolved to support these incremental improvements.
But then electronic vehicles come along, allowing engineers to rethink the entire car from the ground up with electric motors, batteries, and regenerative braking. Startups like Tesla and BYD figured out how to do this profitably. Yet most legacy vendors have needed help to design and mass produce profitable electric vehicles. One issue is that the legacy workflows for designing cars have been rigid and often built for specific tasks like designing combustion engines, which struggle to adapt to new challenges in developing electric vehicles.
Tschammer explains:
The traditional engineering simulation workflow has long been rigid and sequential. Typically, engineers start by managing and modifying product designs using CAD tools, which are tailored for specific tasks. These designs are then prepared and converted into formats compatible with solvers. After running simulations, the results are visualized, often using separate tools, before being shared across teams like design, analysis, and manufacturing. Each step often involves incompatible file formats and manual data handling, creating silos and inefficiencies. Additionally, data management becomes a challenge, especially when multiple teams iterate on the same design with little integration between their tools.
In contrast, NVIDIA Omniverse Blueprint helps transform these workflows by introducing a modular, open and more interconnected approach that allows organizations to meet evolving engineering challenges faster by using the right tool for the task. This is already changing how engineers conceptualize, design, and validate products.
One aspect of this lies in streamlining multi-physics workflows that combine simulations for aerodynamics, structural integrity and hydrodynamics into a unified process. Tschammer says:
This integration is essential for optimizing performance and achieving higher accuracy but remains challenging due to the complexity of combining diverse tools and physics models. Neural Concept and NVIDIA address this by creating platforms that streamline interoperability, allowing engineers to manage multi-physics challenges more efficiently.
For example, Neural Concept has been working with sailing team SP80 and a team of Swiss Federal Technology Institute of Lausanne (EPFL) engineers to design and build a sailboat capable of reaching 80 knots (150 kph), which will attempt to beat the current record of 65 knots in 2025. A key innovation to enable this is the development of a ventilating hydrofoil with a wedge-shaped profile instead of traditional tear-drop designs. This profile creates a stable air bubble around the foil's low-pressure side, significantly reducing drag and allowing the boat to exceed the 50-knot barrier. Since this had never been attempted before, the project required integrating hydrodynamic simulations with structural analysis to ensure both performance and durability.
Another aspect lies in making interconnecting different models for designing products and running simulations easier. This is akin to the pre-written code modules used to streamline the coding process in software development. The engineering world has been embracing model-based systems engineering approaches that use domain models to exchange information, feedback and requirements more rigorously than is possible with describing everything in documents. But things have traditionally gotten complicated when going across domains. Tschammer explains:
Traditionally, these systems were rigid and disconnected, making integration difficult. Neural Concept's platform and NVIDIA's open approach enable greater flexibility by leveraging modern programming languages and automated workflows. This fosters a shift from sequential, static processes to agile, modular systems where new tools can be added or removed as needed. AI workflows can even replace certain components, enabling large-scale MBSE with reduced effort and complexity.
Building an agile engineering workflow must start with advanced IT infrastructure that supports efficient data flow, high-performance computing, and hybrid private cloud environments. But Tschammer says that transitioning from legacy systems to these new infrastructures requires overcoming inertia, organizational resistance, and budgetary constraints.
Cloud platforms can bring digital agility and scalability but pose challenges in cost planning and integration with existing systems. Tschammer recommends that companies focus on quick wins by implementing manageable changes demonstrating clear value to build momentum and justify larger investments to make this transition smoother.
He also observes that younger companies often adapt faster due to their familiarity with modern tools and programming languages, like Python. He notes:
This creates a growing divide between agile innovators and traditional companies stuck in rigid systems. To close this gap, organizations must embrace a cultural shift, modernize their teams' tools, and invest in scalable infrastructure to remain competitive.
These days, when people talk about AI, they point to the buzzy generative AI stuff. However, there are still significant opportunities for many other variants to consider. Neuro Concept's essential innovation has been finding different ways to improve physics-inspired AI surrogate models that can run physical simulations significantly faster than the traditional approach.
But then, when the simulations stop being the slowest link in the chain, it's important to reevaluate all of those other links. That's where solving the integration challenges becomes more important. You have to get the right data between the respective tools, the AI surrogate models, and the different types of users. NVIDIA Blueprint helps improve interoperability at the physical computing level but still relies on the simulation tool vendors to fill in the gaps.
It's telling that Neuro Concept was founded in Switzerland, which has a history of operating as a neutral party in many domains. It's promising that they use this same concept to improve data integration for physical product design.
