Zenithon AI Raises $10M to Build World Models for Extreme Physics

London-based deeptech startup Zenithon AI has raised $10 million to develop AI-powered world models designed to accelerate simulations for some of the most computationally demanding engineering problems, including fusion energy, rocketry and semiconductor manufacturing.

The funding round was led by Backed VC, Lunar Ventures, Seraphim Space, MMC Ventures and SOSV, with participation from angel investors.

Zenithon was founded in 2025 by Alex Higginbottom and Abetharan Antony after the two worked together in fusion research. The company is developing machine learning models aimed at reducing the time required to simulate extreme physics problems from months to potentially seconds.

Building AI Models for Extreme Physics

Zenithon's core idea is to use machine learning as an alternative or complement to conventional simulation software.

Traditional engineering simulations can require significant computational resources and long runtimes, particularly when researchers are modelling complex physical systems such as fusion plasmas.

For engineers, this creates a practical limitation. By the time a simulation is completed, the underlying design may already have changed, requiring another simulation cycle.

Zenithon is attempting to address this problem through what it calls "world models for extreme physics."

The company's models are designed to learn from both simulation data and real-world results. Rather than running every design through a conventional simulation, engineers can use the models to evaluate large numbers of potential designs and receive predictions along with uncertainty estimates.

According to Zenithon, its technology can evaluate up to one million different design variations in the time required for a conventional simulation to run once.

The company is initially concentrating heavily on fusion energy, where simulations of plasma behaviour can take months.

From Fusion Research to a Startup

Zenithon's origins are closely connected to academic fusion research.

Co-founder and CEO Alex Higginbottom was pursuing a PhD in machine learning when he became frustrated with the pace at which engineering progress could take place in fusion.

The article describes his decision to leave the PhD partway through the programme and build a company focused on applying machine learning to difficult engineering problems.

He subsequently joined forces with Abetharan Antony, a plasma physicist who had spent years calculating tokamak results manually.

The founders developed Zenithon's initial technology through academic work involving machine-learning surrogates for gyrokinetic transport calculations.

Gyrokinetic transport calculations are used to understand how heat and particles move through fusion plasma.

That research became the foundation for the startup's broader approach to using machine learning models for extreme physics.

$10M to Train Large World Models

Zenithon says the new $10 million will be split roughly evenly between hiring and computing infrastructure.

The company plans to use the capital to train larger models and release new versions approximately every three months.

The funding also supports expansion of the team in San Francisco and across the United States, while the company continues to grow its research presence in London.

Zenithon currently has a full-time team of 11 people and plans to increase that number to 17 over the next three to six months.

Higginbottom said the funding is intended to provide the compute capacity required to train large models for extreme physics.

The capital therefore represents more than conventional startup operating funding. A significant portion is intended to support the computational infrastructure required to develop the company's models.

Targeting Fusion, Rockets and Semiconductors

Although fusion is currently an important area of focus, Zenithon's ambitions extend beyond nuclear fusion.

The startup says its technology can address engineering problems across extreme physics, including rocketry and semiconductor manufacturing.

These industries share a common challenge: engineers often need to evaluate complex physical interactions across a large number of possible designs.

Running high-fidelity simulations for every variation can be expensive and time-consuming.

AI-based surrogate models can potentially allow engineers to explore a much larger design space before selecting candidates for more detailed conventional simulations.

Zenithon's stated objective is to make that process dramatically faster.

Its million-design comparison claim illustrates the scale of the company's ambition: instead of waiting for a single conventional simulation to complete before evaluating the next design, engineers could potentially use an AI model to explore many alternatives rapidly.

Proprietary Data as a Differentiator

Zenithon says one of its key advantages is the proprietary data used to train its models.

The company says its models are trained using both simulation data and real-world results.

This combination is intended to help the models make predictions that are grounded not only in theoretical simulations but also in observed engineering outcomes.

The company also provides uncertainty estimates alongside predictions, allowing engineers to understand the confidence associated with a particular result.

Zenithon positions this capability as an important part of using AI within engineering workflows, where inaccurate predictions can have significant consequences.

Rather than replacing detailed simulation completely, the technology can potentially help engineers identify promising designs before investing additional time and computing resources in higher-fidelity analysis.

A Growing Physics-AI Market

Zenithon is entering a market where significant capital is already flowing toward AI systems capable of modelling the physical world.

The article notes that companies such as AMI Labs and World Labs have each raised more than $1 billion to work on general-purpose physical reasoning and world models.

Zenithon's approach is considerably more specialised.

Instead of initially attempting to build a general-purpose world model for robotics, video or broad physical reasoning, the company is focusing on extreme physics and engineering applications.

The article identifies PhysicsX as Zenithon's main competitor. The London-based company has raised more than $400 million, including a $300 million financing at a reported $2.4 billion valuation.

Another company in the wider fusion software ecosystem is Fusionality, which focuses on plasma control and measurement for fusion operators.

Zenithon is therefore entering an increasingly competitive space but is attempting to differentiate itself through its concentration on simulation-heavy engineering problems.

Fusion Provides a Major Initial Use Case

Fusion energy is particularly relevant to Zenithon's strategy because of the computational complexity involved in designing and operating fusion systems.

Fusion companies need to understand plasma behaviour, reactor configurations and other highly complex physical interactions.

The article cites Fusion Industry Association figures showing that the fusion sector attracted a record $4.48 billion in private funding in the year ending July 2026.

As investment into fusion increases, the demand for faster engineering and simulation tools could also grow.

Zenithon believes its AI models can help shorten parts of the design cycle by allowing researchers and engineers to evaluate more possible configurations before committing to expensive physical testing or lengthy high-fidelity simulations.

The Road Ahead

Zenithon's immediate priority is to build its research and engineering team while increasing the scale of its AI models.

The company expects to grow from 11 employees to 17 within three to six months and plans to continue developing new models on a roughly quarterly schedule.

Its longer-term opportunity depends on whether AI models can deliver sufficiently reliable predictions for engineering environments where accuracy is critical.

The company is starting with a $10 million funding base and a specialised focus on extreme physics, rather than competing directly across every application of world models.

For Zenithon, the central proposition is straightforward: if AI can reduce months of engineering simulation to seconds for selected workloads, engineers could test dramatically more designs and potentially accelerate development in areas such as fusion energy, rocketry and semiconductor technology.