Zenithon, a London AI lab building world models for extreme physics, has raised £7.5M ($10M) in a seed round led by BACKED.
Fusion reactors, rockets and advanced chip fabs share a bottleneck. The physics inside them is so complex that a single simulation or experiment can take days, which leaves even strong engineering teams testing a small fraction of possible designs before they commit to expensive hardware.
Zenithon trains machine learning models on a mix of simulation output and real experimental data, so they learn how plasmas, shock waves and other extreme systems behave. The company says its models can search a million design points in the time a conventional solver takes to run one, letting engineers look well beyond familiar designs. Its first targets are fusion reactors, hypersonics and the most demanding steps in semiconductor manufacturing. Zenithon builds the full stack itself, from data generation and model architecture to training and delivery, and plans a new generation of models every three months.
The company was formally founded in July 2025 by chief executive Alex Higginbottom and CTO Abetharan Antony, a plasma physicist who previously modelled fusion plasmas at General Fusion, and came through the first cohort of King's College Cambridge's SPARK incubator. It now has a full-time team of 11. Its advisers include Dan Brunner, co-founder and former CTO of Commonwealth Fusion Systems, and Charlie Songhurst, a Meta board director.
Lunar, Seraphim, MMC and SOSV joined the round, alongside founders and directors from hyperscalers. The money will be split roughly evenly between hiring and compute. Zenithon plans to grow to 17 people over the next three to six months, adding research capacity in London and building a delivery team in San Francisco. BACKED partner Alex Brunicki said the firm believes Zenithon has assembled "the best team globally" for the problem.
Sources