A fusion reactor, rocket engine or semiconductor process can take days to simulate properly, which means even the best engineering teams are forced to explore only a fraction of the designs they could theoretically build. Zenithon is betting that AI can turn that bottleneck into an advantage.
The UK AI startup has raised $10 million from BACKED, Lunar Ventures, Seraphim Space, MMC Ventures and SOSV, alongside founders and hyperscaler directors, to develop what it calls world models for extreme physics. The company is targeting fusion, aerospace and semiconductor manufacturing, where small changes in design can have large physical consequences and real-world testing is expensive.
Zenithon’s approach is built around physics-grounded machine learning. Instead of replacing conventional simulation, its models are designed to learn how complex physical systems behave and predict the outcome of different designs much faster. Zenithon says its technology can search one million design points in the time required for a single traditional simulation, while also learning from experimental data as engineers test real hardware. That performance figure is company-reported.
The company was co-founded by Alex Higginbottom and Abetharan Antony, with a team drawn from machine-learning and physics research backgrounds. Its advisers include Meta board director Charlie Songhurst, former Commonwealth Fusion Systems CTO Dan Brunner and NYU assistant professor Zongyi Li, whose work includes neural operators for scientific computing.
The new capital will be split between compute and hiring, with Zenithon planning to expand in San Francisco and across the US while releasing new models roughly every three months.
What makes the story important is not simply faster simulation. If Zenithon’s models perform reliably in production, engineers could explore far larger design spaces before committing to expensive experiments or hardware.
For fusion, aerospace and semiconductor teams, that could change the economics of iteration itself: less time waiting for one answer, and far more chances to discover a better one.
