$10m · Other round · Software · London, UK
London-based Zenithon AI has raised $10m in initial funding led by Backed. Seraphim, Lunar Ventures, MMC Ventures, SOSV and angel investors also participated, according to Seraphim’s announcement. The capital will support model development, team growth and commercial expansion in the US.
Zenithon is developing physics-grounded machine-learning models for engineers working on systems where conventional simulations can take days or weeks. The company is initially focused on fusion energy, with longer-term applications in spacecraft design, advanced propulsion and semiconductor manufacturing. Its goal is to let engineering teams explore substantially larger design spaces without running a full high-fidelity simulation for every variation.
Faster answers still have to earn engineers’ trust
The commercial value is not simply shorter runtime. Fusion, aerospace and semiconductor teams make expensive decisions from simulation outputs, so a learned model must fit existing workflows, quantify uncertainty and stay accurate outside its training data. If Zenithon can make early design exploration much faster while preserving a route back to conventional simulation and experiments, customers can test more ideas before committing scarce compute, equipment or laboratory time.
That validation burden is also the company’s main execution dependency. Zenithon’s founders presented work on physics-informed neural operators for rapid gyrokinetic turbulence modelling at an International Atomic Energy Agency workshop in 2025, describing validation against unseen simulation data and near-real-time predictions as the objective. Turning that research direction into a repeatable commercial product will require evidence that speed gains survive new geometries, operating conditions and customer datasets.
Zenithon was founded in July 2025 by CEO Alex Higginbottom and CTO Abetharan Antony. Seraphim says the 11-person team plans to grow to 17 employees, while the company’s own site lists machine-learning and physics specialists alongside advisers from fusion and neural-operator research. The funding gives Zenithon more capacity to develop the models and work with customers, but adoption will hinge on measurable design-cycle gains rather than the scale of the model alone.
Explore this theme
Why the future of AI belongs to models that simulate reality — Sifted · 12 February 2026. This analysis explains how world models differ from language models and why data, compute and validation remain constraints in physical applications.
Sources checked:
Seraphim Space · Sifted · Zenithon AI · IAEA workshop paper · Tech.eu Funding Explorer
