Engineering simulation delivers deep insights to product development teams, producing glorious full-field data on every variable you might be interested in. Today’s simulation software makes this work almost magically easy, considering the mind-boggling amount of physics modeling and non-linear maths that you are solving at the touch of a button.
Let’s think for a moment about how we got here. Early simulation processes were complicated and long-winded, often involving several different tools with “questionable” interoperability, non-intuitive command line interfaces, text input and output files, etc. Not for the faint-hearted!
More recently, we have become used to integrated environments in which we can do all the work we need to in one place, not to mention those environments are far more visual and interactive. We can easily share models with colleagues and accelerate model setup with easy-to-use templates and well-established best practices. But while those simulations have become much faster over time thanks to advances in hardware and software, they still take a significant amount of time to solve.
Time for the Next Revolution in Simulation
Fast forward to 2025, and we are now riding a new wave of innovation in CAE. AI is set to change the course of engineering, particularly engineering simulation. There is great potential to speed up the performance predictions by orders of magnitude by making inferences using an AI model that is trained on past simulation data.
But here’s the twist: as powerful as this promise is, the landscape is complex. The pace of change is bewildering, and the choice of tooling is broad. If you have experimented with AI surrogate modeling, the chances are that the disjointed and complicated user experience I just described might be coming back to haunt you. There is a natural uncertainty about whether it is worth the time and investment to explore these ideas. How do you even get started? Will it really help you to design a better product?
Foundation models are here to help. They provide a much more accessible way to test the value-add of Physics AI, as well as a way to accelerate you towards successful AI adoption.
1. Foundation Models Already Know About Your Application
Of course, AI models need data, first and foremost. The problem is that most engineering organizations have not organized their simulation data with AI model training in mind. It’s often fragmented across teams, poorly labeled, or saved in formats that would require a lot of manual work to repurpose.
Is it then a misconception that companies are sitting on a ‘gold mine’ of data that they could be using for AI? It depends. The big question is the readiness of that data. In most cases, even where datasets exist, they require significant rework to make them suitable for model training. Whether you are creating a dataset from scratch or compiling and re-running past simulations to obtain the right training data, it can be a major investment to build up the right dataset.
This is where foundation models can give you a much-needed leg-up. They are already pre-trained on a broad set of representative data for the application you are trying to solve, so they provide a baseline of model accuracy that you can then build on top of.
2. Foundation Models Work Out of the Box
Up until now, Physics AI has been an expert-only domain. In order to train and use an AI model, you need to know how it goes together, what tool chain you need, and how to get the most out of it. Foundation models, on the other hand, make the user experience much more similar to how we use LLMs. All of us are perfectly happy to pick an LLM and start prompting it to see what it can do, with almost zero knowledge of how to train and build one ourselves.
Why should this be important to your strategy? Making models widely available and easily accessible is essential for building trust and belief in the technology. You have to see it to believe it. But it isn’t just about testing the capability. The people in your company (or even outside it) who would benefit the most from the instantaneous insights that an AI model can provide will likely not be AI experts. They don’t need to know how to build one themselves, just how to get value from it.
3. Foundation Models Change the Starting Point
As well as encapsulating a large amount of training data and a validated Physics AI methodology, foundation models effectively transport you to a ‘base camp’ of AI-powered exploration. It’s a smarter place to start: a point where the hardest parts of the climb (grappling with data, architecture selection, model training) have already been handled. From there, engineering teams can run inferences, test hypotheses, and get value without the uphill slog.
In practice, this means that you could refine a foundation model with a relatively small number of additional simulations to create a model that has a good grasp of the overall design space, plus the more unique characteristics of your company’s products.
No Specialized Setup. No Steep Learning Curve. Just a Fast, Meaningful Path To Value.
At SimScale, we believe AI should be intrinsic, not a bolt-on.
Because we’re a cloud-native platform, your simulation data is already where it needs to be—accessible, centralized, and structured. That alone removes a huge barrier to AI adoption.
We also provide the tooling to train and deploy AI models directly inside the platform. No exporting, no file juggling. Whether you’re building your own surrogate model or using one of ours, it all happens in the same place where your simulations run.
And we’re not stopping there.
We’re actively developing foundation models for key engineering applications. Our first, a pump performance prediction model built with NVIDIA PhysicsNeMo, was announced at NVIDIA GTC 2025. This model, which is available now, gives pump designers and hydraulic engineers an ideal AI playground in which to explore the capabilities of Physics AI and how it can be integrated into their workflows. Watch the video below and click here or watch our recent webinar to learn more.
Start Your AI Ascent With SimScale Today
With foundation models, your engineering teams will be able to access Physics AI at the point of application, not development.
If you are keen to learn how foundation models can help with the problems you and your team are working on, please get in touch!
