I've spent a lot of time with Physics AI companies over the last few months, and I'm surprised more people aren't talking about the field, especially here in the UK.
The funding is there and the technology is moving quickly. The constraint is people, and it's something I see every day in my work at DeepRec.ai.
This piece covers what Physics AI is, where the money's going, the roles companies are hiring for and where the strongest candidates tend to come from.
What is Physics AI?
Physics AI uses machine learning to predict how physical systems behave, so engineers can test a design in seconds rather than days.
Engineers rely on simulation to design planes, cars, chips, cooling systems and much more. Computational fluid dynamics (CFD) models how air and liquid move, finite element analysis (FEA) models stress and heat in solid structures, and multiphysics simulation combines several of these at once. The results are accurate, but a single high-fidelity run can take days.
Physics AI models learn from that simulation data and return an answer in seconds. These AI surrogate models stand in for the full solver during design, so a team can explore thousands of options rather than a handful. Physics-informed neural networks go a step further by building the governing equations into how the model learns, while neural operators learn to map a design straight to its simulated result.
The wider discipline is often called scientific machine learning, and it sits within the broader AI for Science movement. The commercial idea is simple. Faster AI simulation means faster engineering.
Where the money is going in Physics AI
Investors have noticed. PhysicsX raised $300 million in June at a valuation of around $2.4 billion. In May, Mistral acquired Emmi AI. Mistral's co-founders then put their own money into NP Company in Paris, which competes with the business Mistral had just bought.
Make of that what you will, but I read it as a market that's still wide open. Nobody owns the category yet, and the companies that build the strongest teams will be best placed to lead it.
Why talent is the bottleneck in Physics AI
From where I sit, the money isn't the bottleneck. Finding the people is.
Physics AI needs engineers who understand the physics and the numerical methods properly, and who can also build and scale deep learning models. That's a very small group.
A lot of excellent machine learning engineers have never worked with simulation data. They know how to train large models, but not how a mesh is built, why a solver converges or what makes a boundary condition physically realistic.
A lot of excellent simulation engineers have never trained a model at scale. They know the physics inside out, but haven't worked with distributed training, large data pipelines or modern ML tooling. The people who can work across both are rare.
Physics AI roles companies are hiring for
Job titles in this space are still settling, which makes the market harder to search. These are the ones I see most often.
Engineering roles
- Simulation Engineer (AI/ML)
- CFD Machine Learning Engineer
- Computational Fluid Dynamics (CFD) Engineer
- CAE Machine Learning Engineer
- FEA Engineer (Machine Learning)
- Multiphysics Simulation Engineer
- Digital Twin Engineer
- Computational Engineer
- Machine Learning Engineer
Research roles
- Research Scientist, Physics AI
- Research Scientist, Scientific Machine Learning
- AI Research Scientist, Simulation
- Engineering Simulation Scientist
- Machine Learning Researcher, Physical Sciences
- Computational Physicist (Machine Learning)
- Applied Scientist, Physics-Based AI
- Research Engineer, AI for Science
The engineering roles tend to focus on building models and getting them into real design workflows. The research roles push the methods forward, from new surrogate model architectures to physics-informed machine learning.
Where the best Physics AI candidates come from
The strongest people I've spoken to have come from some fairly unexpected places.
- Computational physics PhDs
- F1 and aerospace CFD teams
- High performance computing (HPC)
- Scientific machine learning research
Plenty of them wouldn't describe themselves as "AI people". A computational physicist or a CFD engineer in a motorsport team may have exactly the right foundation without machine learning appearing anywhere in their job title. If you're only searching for AI job titles, you'll miss most of them.
That's why we look past the title. We focus on the problems someone has solved, the simulation tools and numerical methods they've used and the models they've built.
The UK's chance to lead in Physics AI
I think the UK has a real shot at leading here. We already have deep engineering talent across aerospace, motorsport and academic research, and much of it works with simulation every day.
Keeping that talent is another matter. UK companies building Physics AI teams will need to give simulation engineers and computational physicists good reasons to build their careers here.
Hiring for Physics AI
Physics AI is the area I'm focusing on at DeepRec.ai. If you're building a Physics AI team, or you're one of the people who can work across simulation and machine learning, I'd love to hear from you.
Get in touch with Ed Killin at DeepRec.ai, or see how we support simulation engineer recruitment and scientific machine learning researcher recruitment.