BeyondMath raises $18.5m to train AI that understands physics

Modern engineering software has long been built the same way: one model, one problem. Early large language models, for example, were typically designed to do one task — translate text, answer questions or search. In physics and engineering, the same logic applied. One tool for airflow, another for heat or electromagnetics, each tool highly specialised to an individual use case.

Now, with GPT-style models and a foundational approach, one model can be applied to a wide array of problems, and, for BeyondMath, this looks like a foundational model of physics that it believes could transform how engineers work.

The Cambridge-founded deeptech company is building AI models that learn the laws of physics directly, enabling design iteration at much higher speeds. 

By training a single, foundational model to understand how fluids, heat and forces behave, the company believes it can apply the same system across automotive, aerospace and energy — from Formula 1 cars to aircraft components and renewable infrastructure — dramatically reducing the time it takes to go from design to production.

“We are building a foundational model of physics that can be applied to a wide range of problems, as opposed to a model that can do one problem in a narrow way. And we are quite differentiated in the market in that sense,” co-founder and CEO Alan Patterson tells Pathfounders.

The result should be faster simulation — enabling thousands of iterations in seconds rather than days — reducing costs, increasing efficiency and reducing the environmental impact.

That approach has now attracted fresh capital. BeyondMath has closed a $10m seed extension led by Cambridge Innovation Capital, alongside existing investors including UP.Partners, Insight Partners and InMotion Ventures. The round brings its total funding to $18.5m, following an $8.5m raise in 2024.

But the ambition extends far beyond speeding up simulations.

“It's not just the number of iterations and optimisation, but how fast you can go from paper to production. Why does it take five years? Why doesn't it take six months? And how do you make that happen? And when you go into how these companies work and how we build and engineer and manufacture, bringing down these timelines at every point of the chain is hugely important,” Patterson says.

That capability is already being deployed at scale. BeyondMath’s customers include major automotive, aerospace and electronics manufacturers, and it has partnerships with NVIDIA and AWS. In the US, it is working with engineering firm Honeywell on a $19m, three-year project with the firm to reduce the time it takes to simulate thousands of iterations of complex aircraft components from days to seconds.

In Formula 1 — where competitive advantage is measured in fractions of a second — BeyondMath supports real-time testing of thousands of design combinations. Engineers can simulate changes down to the literal nuts and bolts to optimise aerodynamics.

“Almost everything we build or manufacture deals in some way with this problem,” Patterson says.

The company was founded by AI industry veterans Patterson and Darren Garvey, marking their fourth time working together. They first crossed paths in 2010 at Evi Technologies (then True Knowledge), which was acquired by Amazon and went on to form the basis of Alexa. Later roles at eBay and HomeX followed, before they left in 2022 to launch BeyondMath.

Having studied electromagnetic simulation in the 90s, Patterson remembers it taking a week to run a simulation. Fast forward to 2022 and he saw it was still a huge bottleneck. But with advances in AI and machine learning over the past five years, a lot more has become possible when it comes to training AI to conceptualise physics to this level.

“I think even now the hardware is really only just catching up to the scale of the problem. And also the techniques that you have to employ means it's only really possible to do this now,” Garvey says.

There was some scepticism when they started given the level of AI adoption, but now, “it feels like this year is going to be a huge year for physics AI,” Garvey says. “I think the markets now are starting to recognise the value it can have. People are starting to really see it.”

“BeyondMath is tackling one of the hardest and most valuable problems in engineering,” Edward Inns, Principal at Cambridge Innovation Capital, said in a statement. “By combining first-principles physics with modern AI, the team has built a platform that can redefine how complex systems are designed across multiple industries.”

The new funding will be used to scale commercial deployments, increase its customer base across Europe, the US and Japan, and nearly double the company’s team by the end of the year. And, as Patterson notes, it will also mean “as ever with AI companies: more compute, more GPUs”.

Company info: