Cambridge startup raises £7m for gaming AI training data

TL;DR:

• Cambridge-based Worldmodeldata has raised £7m in seed funding as it emerges from stealth.
• It aggregates licensed gameplay from titles built on engines such as Unreal and Unity to train “world models” — AI that learns how environments look, interact and change.
• The round was led by London VC Iona Star Capital, with a goal of a one-million-hour data library by the end of 2026.

Worldmodeldata is betting that the next wave of AI will be built not on scraped web text but on data that captures how the physical world behaves over time. World models — an AI system’s internal understanding of cause and effect — underpin efforts to build robots and agents that can predict outcomes and plan actions safely. The catch, the company argues, is that such models are only as good as the data they learn from, and that data barely exists at scale.

Games as a controlled world

Its answer is video games. Modern titles generate rich records of human behaviour and interaction in complex, dynamic environments — and, crucially, they are safe and controllable. Worldmodeldata structures gameplay from Unreal- and Unity-based titles into datasets for frontier labs, robotics firms and physical-AI developers. Rather than scraping, it licenses the data, letting developers and the gaming community monetise their gameplay and assets.

“World models represent a fundamental paradigm shift in AI, but progress like this needs fuel: internet-scale data to help AI systems make predictions and reason in physical environments,” said founder and chief executive Rhea Loucas. “Video games, as safe, controlled environments, are the perfect setting for generating the action-conditioned data needed to train the next generation of AI.”

The round is a small but telling data point in a UK market where AI firms now command a record 44% of smaller-business equity funding. It also reflects a licensing-first stance on training data at a time when the provenance of AI datasets is under growing legal and ethical scrutiny.

Looking forward

The funding will go toward product development, hiring and securing data-sourcing agreements. If world models do become the backbone of physical AI and robotics, a supply of licensed, high-fidelity behavioural data could prove valuable infrastructure — and a rare example of a UK startup positioning upstream of the frontier labs rather than competing with them.