Worldmodeldata raises £7M to train AI on gaming environments
Worldmodeldata is turning licensed gameplay into training data for world models and physical AI, targeting a growing need for dynamic, provenance-aware datasets.
AI systems that understand the physical world need more than static text and images. They need data about how environments change over time.
What happened
Cambridge startup Worldmodeldata emerged from stealth with a £7M seed round led by Iona Star Capital.
The company is building licensed datasets from video-game environments for training AI world models and physical-AI systems. It says it sources gameplay data through agreements with developers and communities, including games built on Unreal and Unity, rather than scraping the web.
Why it matters
Video games can provide rich data about movement, objects, actions and cause-and-effect inside complex environments.
That makes them potentially useful for training models that need to reason about sequences and simulated worlds. The licensed-data model also matters as AI companies face growing scrutiny over where training data comes from.
Worldmodeldata is therefore attacking two bottlenecks at once: dynamic training data and provenance.
The bigger picture
The AI-data market is moving beyond web scraping. As world models and robotics systems develop, companies will need datasets that represent interaction, physics and temporal change.
Gaming environments could become one source of that training material, especially where data can be licensed cleanly and generated at scale.