UK Startup Raises $8.6M to Turn Video Games Into AI Training Data

Jan de Vries · 2026-07-06

A Cambridge-based startup called Worldmodeldata just landed $8.6 million (about 8 million euros) in Seed funding to do something pretty wild: turn video games into training data for the next generation of physical AI. The round was led by Iona Star Capital, and the company is coming out of stealth mode with big ambitions. You know how AI has gotten really good at describing the world? Think chatbots that can write essays or generate images. That's cool, but it's not enough. The next big leap is AI that can actually act in the physical world, like robots that navigate warehouses or self-driving cars that handle unexpected obstacles. And for that, these systems need a totally different kind of training data. ### What Are World Models, Anyway? World models are basically an AI system's internal understanding of how the world works. Instead of just reacting to inputs, they learn how things look, interact, and change over time. This lets them predict what's going to happen next and plan actions accordingly. Think of it like a kid learning that if they drop a glass, it'll shatter. They don't need to actually break a hundred glasses to figure that out. Rhea Loucas, the founder and CEO of Worldmodeldata, explains it like this: "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." The problem is, that kind of data doesn't really exist yet. ### Why Video Games Are the Perfect Sandbox Here's the clever part. Video games are safe, controlled environments where you can generate massive amounts of action-conditioned data. Every time a player moves a character, jumps over a gap, or dodges an enemy, there's a clear relationship between action and consequence. That's exactly what physical AI needs to learn. Worldmodeldata captures this data directly from game engines, with humans in the loop to ensure quality. They're essentially manufacturing the missing ingredient that every company building physical AI will eventually need. ### The Bigger Picture: A Funding Frenzy in Europe This Seed round isn't happening in isolation. It's part of a wider pattern in 2026 around AI data infrastructure, world models, and physical AI across Europe. Just this year, we've seen some massive adjacent rounds: - AMI raised $967 million in Paris for world-model AI systems - Encord, based in London, secured $54 million in Series C funding for data infrastructure - Stanhope AI, also in London, raised $7.3 million for adaptive AI in robotics and defense These are serious numbers, and they signal that investors are betting big on the idea that AI's next frontier isn't just about smarter chatbots, it's about systems that can actually interact with the real world. ### Why This Matters for the US Market You might be wondering why a US audience should care about a UK startup. The answer is that this isn't just a European trend. The same data scarcity problem exists everywhere. As Gerry Buggy from Iona Star puts it, "AI spent the last few years learning to describe the world... The next era is about acting in it, and you cannot learn to act from passive video or text." For US companies building autonomous vehicles, warehouse robots, or medical devices that need to operate in complex environments, this approach could be a game-changer. Instead of expensive real-world testing, you can generate training data at scale from video games. ### The Team Behind the Vision Worldmodeldata was founded in 2025 by serial entrepreneur Rhea Loucas. The company also has Lord Richard Allan on board as Chairman. He's a UK technology policy specialist and former VP of Public Policy at Meta. That kind of experience is valuable when you're dealing with the regulatory implications of physical AI. Richard Allan says the company is "proud to anchor ourselves in the UK's AI ecosystem, a strategic choice driven by the urgent push for sovereign AI capabilities." He adds that this isn't just about improving AI model training, but "building an essential foundation for deploying AI in sectors where the demand is vast but the solutions remain limited." ### The Bottom Line Worldmodeldata is tackling a critical gap that large language models simply can't bridge. They're creating the training data that will power the next generation of AI systems that can actually see, move, and act in the physical world. And they're doing it by turning something as familiar as video games into a serious tool for innovation. Whether you're building physical AI yourself or just watching from the sidelines, this is a space worth keeping an eye on.