Zenithon, a London-based startup building artificial intelligence models for some of the hardest engineering problems on Earth and beyond, has disclosed $10 million in funding. The company announced the funding on 30 September 2026, with backing from Backed, Lunar Ventures, Seraphim Space, MMC Ventures, SOSV and angel investors.
Co-founder Alex Higginbottom said the $10 million figure combines two financings raised over the past year, so it represents newly disclosed funding rather than a single round that closed this week.
World models for extreme physics
Zenithon develops what are known as world models: AI systems that learn how a physical environment behaves and can predict what will happen when conditions change. Rather than applying such models to video games or robotics, as many other companies do, Zenithon focuses on engineering problems involving extreme physics, including fusion reactors, rocket engines and semiconductor manufacturing processes.
These are areas where engineers rely heavily on computer simulation. Designing a fusion reactor, for example, requires modelling how superheated plasma behaves inside powerful magnetic fields. Designing a rocket engine involves simulating combustion, fluid flow and heat transfer at extreme temperatures and pressures. In chip manufacturing, engineers must model processes such as etching and deposition at the scale of individual atoms.
Conventional physics simulations for such systems can be extremely slow and expensive, sometimes requiring days or weeks of supercomputer time for a single design. That limits how many options engineers can explore.
Zenithon says its models can explore as many as one million design configurations in the time conventional methods might need for a single simulation. That is a company claim that will need to be validated by customers and independent testing, but if it holds, it could change how engineers search for better designs.
Founded by fusion researchers
The company was founded by Alex Higginbottom and Abetharan Antony, both of whom have backgrounds in fusion research. That experience gives them first-hand knowledge of the bottlenecks that slow progress in one of the most demanding fields of engineering.
Fusion energy, which aims to reproduce the process that powers the sun to generate almost limitless clean electricity, has attracted billions of dollars of private investment in recent years. The United Kingdom is one of the leading centres for fusion research, with a national programme to build a prototype fusion power plant and a cluster of private fusion companies. Faster, more accurate simulation is widely seen as one of the keys to accelerating progress.
Where the money will go
Zenithon is a small team of about 11 people, and expects to grow to around 17 in the coming months. The company plans to spend roughly half of its funding on computing power and half on hiring. It aims to release new models approximately every three months.
That balance reflects the economics of building AI models: training them requires significant computing resources, while improving them requires specialists who understand both machine learning and the underlying physics.
Investors with space and deeptech focus
The investor group reflects Zenithon's focus on hard technology. Seraphim Space is one of the most active investors in space technology globally. Lunar Ventures and Backed back technically ambitious European startups, while MMC Ventures is one of the UK's established early-stage investors. SOSV is known for its deep technology accelerator programmes in areas such as climate, health and hardware.
Part of the world model boom
Zenithon's funding comes amid a surge of investor interest in world models, which many in the AI industry see as the next major frontier after large language models. Language models learn from text; world models learn how physical systems behave, which is essential for applications in robotics, autonomous vehicles, scientific discovery and engineering.
In the same week, New York-based General Intuition announced it had raised $220 million at a $6.2 billion valuation to build action and world models trained on video game footage. Large technology companies and AI labs have also been investing heavily in the field.
Zenithon's approach differs from many of these companies. Rather than building general-purpose models of the visual world, it focuses on narrow, high-value domains where accurate physics matters more than visual realism. That specialisation could make its models more useful to engineering teams, but it also means its market is smaller and more technical.
The opportunity in engineering software
Simulation software is a large and established market, dominated by companies whose tools are used across aerospace, automotive, energy and semiconductor industries. These companies have begun integrating AI into their products to speed up simulations, and they have the advantage of deep customer relationships.
AI-native startups such as Zenithon are betting that a fundamentally new approach, built around learned models rather than traditional numerical solvers, can deliver step-change improvements in speed. If the models are accurate enough, engineers could use them to rapidly narrow down promising designs before running detailed conventional simulations on a small number of candidates.
Why the UK is a natural base
London offers Zenithon access to a deep pool of talent in both machine learning and physics. The city hosts major AI research labs, and the UK's universities and national laboratories have long been at the forefront of fusion and plasma research. The country's space sector, supported by a growing cluster of satellite and launch companies, provides another potential customer base. For a company working at the intersection of AI and extreme physics, that combination of expertise is difficult to find elsewhere in Europe.
Challenges ahead
The biggest challenge for AI-based simulation is trust. In safety-critical fields such as rocket engines and fusion reactors, engineers need confidence that a model's predictions are reliable, particularly in conditions outside the data it was trained on. Demonstrating accuracy and understanding the limits of AI predictions will be crucial for adoption.
Data is another constraint. High-quality training data for extreme physics often comes from expensive experiments or simulations, and much of it is proprietary.
Why it matters
If Zenithon can deliver on its promise, it could help accelerate progress in some of the most important technologies of the coming decades: clean fusion energy, cheaper access to space and more advanced semiconductors. Each of those fields is limited in part by how quickly engineers can test new ideas.
With $10 million in disclosed funding, a specialist team and a growing appetite among investors for AI that understands the physical world, Zenithon is positioned to show whether world models can move from impressive demonstrations to practical engineering tools.
