- Zenithon AI has raised $10 million to develop AI world models that will speed up simulations in the fields of fusion, rocketry, and semiconductor chips.
- The London-based startup aims to test one million engineering designs in the time it previously took to run a single simulation.
- The fundraising round was led by Backed VC, Lunar Ventures, Seraphim Space, MMC Ventures, and SOSV, with some angel investors also participating.
While pursuing a PhD in machine learning, Alex Higginbottom concluded that a greater challenge lay beyond the scope of his thesis. Since simulating plasma in a fusion reactor takes several months, the engineers will have already altered the designs by then.
“I left my PhD because I was frustrated with the rate of progress in fusion energy. I wanted to take the strongest machine learning researchers in the world and apply them to these problems that really mattered, and a startup was the best way to do that,” Higginbottom recalled of his decision to leave halfway through his doctoral studies in a conversation with Tech Funding News.
He joined forces with Abetharan Antony, a plasma physicist who had, for many years, calculated tokamak results by hand.
Zenithon AI has raised $10 million to develop machine learning models for extreme physics. This involves engineering challenges in rockets, nuclear reactors, and semiconductor manufacturing, where simulations are time-consuming and testing many designs is difficult.
The funding round was headed by Backed VC, Lunar Ventures, Seraphim Space, MMC Ventures and SOSV, with assistance from angel investors. Higginbottom explained that this initial funding wraps two funding rounds into one over the past year, while keeping the door open for potential follow-on extensions.
“We’re using $10 million to train large world models for extreme physics… Without it, we wouldn’t be able to train such large models and have such strong capability,” said Higginbottom. The capital expenditure will be split evenly between headcount and compute, enabling Zenithon to release new models every three months and expand its team in San Francisco and across the U.S.
Alex Brunicki of Backed VC added: “We are delighted to be partnering with Alex and Abi as they develop SOTA Foundation Models for extreme physics… We believe Zenithon will have a profound impact on these industries and are thrilled to be partnering with this exceptional team.”
A substitute for conventional simulation software
Zenithon was established in 2025 in London by Higginbottom and Antony, who first met in the context of fusion research and developed their first product whilst working on academic projects involving machine-learning surrogates for gyrokinetic transport calculations, a method used to describe how heat and particles move in a fusion plasma.
Zenithon presents its technology as a substitute for conventional simulation software, which generally requires engineers to make assumptions before conducting tests.
“We are building the world’s first world models for extreme physics… What sets us apart from competitors is that we’re the only company with this vast amount of proprietary data,” Higginbottom notes.
Since its models have been trained on both simulation data and real-world results, engineers can examine more design options and obtain uncertainty estimates with each prediction.
The company states that its models can evaluate a million different design variations in the same time it takes a standard simulation to run once. Among its early customers are those working in the field of fusion energy, since reducing the time required for plasma simulation from months to seconds could accelerate reactor design.
A competitive landscape and what’s next
Zenithon’s $10 million funding is minor when compared to the total amounts invested in world models. For example, both Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs have each raised more than $1 billion for general-purpose physical reasoning in robotics and video. There is also the company Fusionality from Lausanne, which develops software for plasma control and measurement for fusion operators and operates in a more specialised area.
The main competitor of Zenithon is PhysicsX, another London-based startup that has raised more than $400 million, including a $300 million financing at a $2.4 billion valuation.
Zenithon, on the other hand, is starting with $10 million in seed funding and is focusing on extreme physics related to fusion, with the aim of developing deep expertise in a challenging simulation field.
“We are one of the leading AI labs in the UK now,” said Higginbottom, highlighting plans to grow their research team in London. Currently operating with a full-time team of 11, the startup plans to expand to 17 team members over the next three to six months.
The global market for simulation software was valued at approximately $26.5 billion in 2025 and is projected to grow by about 13% per year until 2033. Fusion raised a record $4.48 billion in private funding in the year ending July 2026, according to figures from the Fusion Industry Association.
It will be up to whether Zenithon can compete with larger companies such as PhysicsX, or whether it is acquired or comes under market pressure to demonstrate what the physics-AI sector might do in terms of future consolidation.