Zenithon Raises $10M for Physics World Models

Zenithon has raised $10 million to build world models aimed at environments where physics becomes too complex or expensive to explore through conventional simulation alone.

What happened

The London-based AI lab raised $10 million from investors including Backed, Lunar, Seraphim, MMC and SOSV.

The company is focused on world models for extreme physics rather than broad consumer or gaming environments.

Its thesis is that models capable of learning how difficult physical systems behave could become useful in areas such as aerospace, science and advanced engineering.

Why it matters

World models are gaining attention because they may allow AI systems to reason about how an environment changes over time rather than simply recognise patterns in static data.

That is especially valuable in scientific domains where physical experiments are expensive, dangerous or difficult to repeat.

The challenge is proving that learned simulations remain accurate enough to be useful outside the training distribution.

The bigger picture

A growing group of AI startups is specialising world models around specific physical domains.

That mirrors what happened in enterprise software: general-purpose tools emerged first, followed by more specialised products with stronger domain knowledge.

Zenithon's funding is small compared with the largest world-model rounds, but its focus on extreme physics makes it strategically interesting as AI research moves closer to engineering and scientific applications.