€8M Bet on Video Games: How One UK Startup Is Powering the Next Generation of AI

Did you ever imagine that the future of AI might not be built on internet data alone, but on video games?

That’s the vision behind Worldmodeldata, a Cambridge-based AI startup that has emerged from stealth with €8 million (£7 million) in Seed funding led by Iona Star Capital. The company is building a massive library of video game-generated training data to power the next generation of world models and physical AI. As someone who grew up playing video games, I never would have imagined that the hours spent exploring virtual worlds could one day contribute to advancing AI.

Founder and CEO Rhea Loucas, sees world models represent a major shift in AI development, but they require internet-scale, action-conditioned data that simply doesn't exist today. Instead of relying solely on real-world data, Worldmodeldata believes video games provide the perfect environment to generate the massive datasets AI needs. It’s a fascinating idea because games naturally create millions of interactions that would be expensive, time-consuming, or even impossible to replicate in the real world.

The funding also highlights Europe's growing commitment to AI infrastructure, with companies like Encord and BeyondMath helping strengthen the UK's position in physical AI. To me, this shows that investors are looking beyond flashy AI applications and starting to back the infrastructure that could power future breakthroughs.

Worldmodeldata plans to expand its data library to 1 million hours by the end of 2026, a huge jump from today's largest datasets of around 40,000 hours. The new funding will support product development, team expansion, and new licensing partnerships. It's certainly an ambitious target, but it also raises an interesting question: Can AI trained primarily in virtual worlds truly understand the unpredictability of the real one? Even the most realistic games follow rules, while real life is far less predictable. It'll be interesting to see how well these models transfer what they've learned beyond the screen.

Still, I think this is one of the most creative approaches to AI training I've seen recently. It reminds us that innovation often comes from looking at familiar things in completely different ways. Sometimes the hobbies we enjoy outside of work end up inspiring technologies that could shape the future.

💬 Join Our Startup WhatsApp Community

Don't build your business alone. Join thousands of founders, entrepreneurs, and startup professionals sharing funding opportunities, industry news, events, partnerships, and practical advice every day.

👉 Join the community today and stay connected with the startup ecosystem.

Do you think video games could become one of AI's most valuable training grounds, or will real-world data always have the advantage?