Faced with Ansys and Dassault, Beyondmath wants to impose a new standard in multiphysical simulation

Reducing weeks of scientific calculation a few seconds is Beyondmath’s promise. Founded in 2022 to Cambridge by Alan Patterson (CEO, ex-google) and Darren Garvey (CTO), this British startup offers a radical alternative to traditional simulation software, by combining generative and fundamental artificial intelligence. Their ambition is to replace the supercomputers to $ 50 million and expert armies with an accessible, fast and resource -thrifty platform.

At the heart of their solution, an AI capable of learning complex physical laws from large corpus of industrial and scientific data. Unlike conventional simulators that resolve equations by iterative digital calculation, Beyondmath directly generates the most likely responses to engineering problems, within a reduced time of a 1000 factor. It is intended for the automotive, aeronautical, batteries, data centers, where simulation constraints slow down the marketing.

The company was equipped in 2024 with an NVIDIA DGX H200 system, cutting -edge IA architecture designed to cause very large -scale models. This technological choice, rare for a startup startup, wants to provide real -time multiphysical simulation in production environments.

An engineering in rupture

Beyondmath’s starting point is in an observation shared by many manufacturers, namely that current simulation software is too expensive, too complex and too slow. Reference solutions like Ansys, Comsol or Simulia (Dassault Systèmes) require heavy infrastructure, sharp expertise, and processing times incompatible with rapid innovation cycles.

Beyondmath offers a complete bypass of this logic. By leading to AI models on phenomena such as thermodynamics, fluid mechanics or electromagnetic fields, the startup does not seek to reproduce conventional solvents, but to exceed them. The promise is twice both drastically reduce the costs of access to simulation, and integrate scientific calculation into iterative workflows, including in the design phase.

Thought in Cambridge, the company quickly established its headquarters to Londonwhere it now structures its technical and commercial teams. Today, it has around fifteen employees from Deepmind, Oxford, Cambridge or Imperial College, divided between AI, HPC and computational physics.

A fundraising to structure industrialization

In August 2024, Beyondmath raised 8.5 million dollars with Up. (lead), Insight Partners And Inmotion Venturesthe investment fund of Jaguar Land Rover. This Seed series financing aims to strengthen technical infrastructure, speed up recruitments and develop the first deployments in the automotive, energy and industrial sectors.

This lifting confirms the growing interest in technologies capable of shaking up conventional simulation tools. The entry of ENDUCTION VENTURES also highlights a strategic interest in concrete cases, in particular in the thermal optimization of components or the shortening of the prototyping cycles.

A global competition still open

Beyondmath’s positioning is part of a global dynamic. In the United States, intrinsic (alphabet subsidiary) explores uses of simulation in robotics. Others, such as Abacus.ai, apply the generative models to chemistry or materials. In Europe, Neural Concept (Switzerland) constitutes a direct competitor, already integrated at Airbus or BMW. The British Toffeeam targets additive manufacturing with topological optimization algorithms.

In France, despite a strong tradition in simulation (CEA, Mines Paris, Inria), no actor has yet emerged with such a transversal generative proposal. Some initiatives explore molecular modeling or thermal simulation, but the ecosystem remains embryonic. France has scientific bricks, but it lacks entrepreneurial interfaces and risk tolerance to bring out actors like Beyondmath. Especially since Dassault Systems plays a central role in the structuring of French simulation. But his vertical and integrated model slows down the emergence of more radical approaches, such as generative ia simulation. The blocking is not intentional, but perfectly illustrates the case with a systemic effect of technological hegemony.