Zenithon raises $10M to build faster models for physics-heavy engineering

Why it matters

Zenithon is trying to turn costly physics simulations into a repeatable software workflow. Its fusion research gives the founders a concrete starting point; independent performance evidence and paying customers will determine whether the approach extends across industries.

Zenithon has raised $10 million to build machine-learning models for physics-heavy engineering, Tech.eu reported. The report, published September 30th, does not specify when the financing was announced or closed, so the event date remains unverified. Co-founders Alex Higginbottom and Abetharan Antony are targeting problems in fusion, rockets and semiconductor manufacturing, where slow simulations can limit how many designs engineers evaluate.

The financing backs a specific bet: train models on expensive simulations and real-world experiments, then use them to predict how different designs might behave. Zenithon says its approach could let engineers explore up to one million design points in the time required for one conventional simulation. That is the company's performance claim; the materials available for this report do not provide an independent benchmark establishing it.

A fusion problem becomes a model-making business

Higginbottom came to Zenithon from physics and fusion analysis. ContactOut lists him as a fusion-energy techno-economic analyst at Woodruff Scientific from June 2023 to May 2025, after a research internship there, according to his public profile.

Zenithon identifies Antony as its CTPO, and public sources attribute a PhD in plasma physics to him. An earlier research project involving both founders was presented at a December 10th, 2025 International Atomic Energy Agency workshop. Its abstract described physics-informed neural operators as a surrogate for gyrokinetic plasma simulations, with the goal of producing near-real-time turbulent-transport predictions.

That work gives the company's broad "world models" pitch a concrete starting point. Zenithon's early technical example is a model intended to approximate a costly class of fusion-plasma simulation. The conference abstract discussed validation against unseen simulation data; it does not establish the million-design-point speedup or show that the system has learned from live experimental results at production scale.

The round buys iteration, not yet a disclosed customer base

Zenithon says it will spend the $10 million on hiring and compute, with a goal of releasing new models every three months and expanding in San Francisco and elsewhere in the United States. Tech.eu reported that the investors include BACKED, Lunar, Seraphim, MMC and SOSV, alongside founders and hyperscaler directors. The report does not identify a lead investor or disclose a valuation, financing structure or cumulative amount raised.

The company was incorporated in the UK on July 14th, 2025, according to Companies House. Its first stated application is narrow enough to test against a recognized engineering bottleneck, while its intended markets are wide: fusion reactors, aerospace and semiconductor fabs all involve costly physical systems and long design cycles. Each field has its own data, validation and deployment requirements. A model that accelerates one plasma calculation does not automatically transfer to rocket design or chip manufacturing.

The planned release cadence is central to the financing plan. The reported round gives Higginbottom and Antony money to build models and expand the team, but Tech.eu's report names no customers, commercial contracts or revenue figures. A key test will be whether successive models make a measurable difference inside engineers' existing workflows, where conventional simulation and experiments remain the reference for validating designs.

Zenithon's first technical step is to use machine learning to make a slow fusion calculation cheaper to explore, then test whether the method applies to other hard-physics problems. The founders have research experience close to that initial use case. Turning it into dependable tools across several industries is the work the company says the new capital will fund.