en.Wedoany.com Reported - Cambridge-based Worldmodeldata has announced the completion of a £7 million (approximately €8 million) seed funding round, aimed at converting player actions and scene data from video games into artificial intelligence training datasets. The round was led by London's Iona Star Capital, with Lord Richard Allan, Meta's former head of European public policy, joining the company as chairman.
The company targets a new type of AI system known as "world models." Unlike traditional chatbots that only process text, world models attempt to predict how an environment changes after an agent takes action within it. Training such models requires a special type of data: actions and their resulting environmental feedback, aligned on a frame-by-frame basis.
Worldmodeldata claims that video games are perfectly suited to record this information. The company sources footage and engine data from games built on Unreal and Unity engines (rather than through web scraping), packaging video, player inputs, and underlying 3D states into structured datasets. Its target customers are laboratories researching physical AI, robotics, and autonomous vehicles.
The company highlights a key figure in its promotion: by the end of 2026, its data scale will reach one million hours of game data. The company states this scale is 25 times larger than the largest publicly available dataset, though this comparison is self-claimed and has not been independently verified. Currently, its actual operations are small-scale, with no final customer contracts signed and no revenue. The team consists of about ten people (including advisors and contractors). The seed round actually closed last December, seven months before this week's public announcement. Founder Rhea Loucas stated that the company "decided to remain silent for a while" to clarify laboratories' specific data needs.
There are already several competitors in this space. San Francisco's Origin Lab raised $8 million in May and has signed over 20 publisher agreements. New York's General Intuition has raised $454 million, but its data is used internally to train its own models. Loucas positions Worldmodeldata as a neutral data supplier, selling data to all customers, with a role closer to a company providing stress tests for AI systems rather than a model builder.
Worldmodeldata is essentially a bet on the European market. The company chose to remain in Cambridge, with Allan viewing it as part of the UK's push for sovereign AI. Whether a ten-person team can achieve the one-million-hour data target before better-funded competitors remains uncertain. For now, Worldmodeldata is selling "shovels" in a "gold rush" that has just begun, with customers hoping the "gold" will be an AI that ultimately grasps causality.
