Zenithon AI Secures $10M to Accelerate Fusion and Rocketry Simulations with AI World Models

The development of next-generation hardware, from nuclear fusion reactors to advanced semiconductors, is currently bottlenecked by the computational weight of physics simulations. Calculating plasma behavior in a tokamak reactor or fluid dynamics in a rocket engine can take months, meaning engineers often move on to new designs before previous test results are finished. This lag forces deep tech companies to rely on slow, iterative physical testing rather than rapid digital prototyping.

To solve this, London-based Zenithon AI has raised $10 million to build large-scale AI world models specifically designed for extreme physics. Founded by machine learning researcher Alex Higginbottom and plasma physicist Abetharan Antony, the startup replaces traditional mathematical solvers with neural networks trained to predict physical interactions instantly. By grounding their models in physics and providing uncertainty estimates, Zenithon AI allows engineers to evaluate up to a million design variations in the time a legacy system takes to render just one.

The funding round, led by Backed VC, Lunar Ventures, Seraphim Space, MMC Ventures, and SOSV, consolidates two earlier raises and provides capital for massive compute resources. Unlike general-purpose AI models that struggle with strict physical laws, Zenithon's approach represents a growing sub-sector of scientific AI tailored for industrial R&D. If successful, this technology could shorten development cycles for clean energy and space exploration, shifting the hardware industry closer to the rapid iteration speeds of software development.