As artificial intelligence evolves beyond language processing and image recognition, developers are increasingly working on systems capable of understanding how the physical world behaves. These emerging AI models, often referred to as world models, are designed to predict how environments change over time, allowing machines to reason, plan, and make decisions in dynamic real world situations. Training such systems requires enormous volumes of high quality environmental data, something that remains difficult and expensive to collect. Cambridge startup Worldmodeldata believes modern video games can provide an effective solution, and the company has now secured fresh funding to accelerate the development of its unique AI training data platform.
Worldmodeldata has raised £7 million in seed funding in a round led by London based venture capital firm Iona Star Capital as the company officially emerged from stealth.
The investment will support platform development, team expansion, and new licensing partnerships while accelerating the creation of one of the world’s largest structured datasets for world model training.
Building the Data Layer for World Models
Founded by serial entrepreneur Rhea Loucas, Worldmodeldata is developing a specialised platform that aggregates and organises gameplay data generated within modern video games.
The company focuses on supplying structured datasets for organisations building world models, physical artificial intelligence systems, autonomous technologies, and robotics.
Unlike conventional AI datasets collected from text or images, gameplay data captures dynamic environments where objects, characters, and physical interactions continuously evolve.
The company believes these simulated environments provide valuable training material for AI systems that must understand how the real world changes over time.
Turning Video Games Into AI Training Data
Worldmodeldata sources gameplay information through licensing agreements with game developers and gaming communities rather than relying on web scraping.
Its platform works with games developed using widely adopted engines including Unreal and Unity.
By securing licensed access to gameplay environments, the company can organise large volumes of structured information suitable for machine learning while respecting intellectual property rights.
The resulting datasets allow artificial intelligence developers to train systems using highly diverse virtual environments that closely resemble many real world scenarios.
This approach offers a scalable alternative to collecting equivalent physical world data.
Supporting the Next Generation of AI
The company believes world models represent one of the most important developments in artificial intelligence.
Unlike conventional AI systems that primarily analyse static information, world models attempt to predict how environments evolve over time while supporting planning and decision making.
Potential applications extend across numerous industries.
Autonomous vehicles could use world models to anticipate traffic behaviour and pedestrian movement.
Robotics developers may train machines to navigate unfamiliar environments more effectively.
Simulation platforms, industrial automation, and physical AI research could also benefit from richer environmental datasets.
Strengthening Leadership and Growth
As the company begins its commercial expansion, technology policy specialist Lord Richard Allan has joined the board as chairman.
His appointment adds significant experience in technology policy and digital governance while supporting the company’s long term strategic development.
Worldmodeldata plans to continue expanding relationships with game developers and AI companies as demand for specialised training datasets increases.
The company believes responsible licensing partnerships will remain central to its growth strategy.
Scaling Towards One Million Hours of Data
The newly secured funding will enable Worldmodeldata to accelerate product development while expanding its engineering team and licensing activities.
A key objective is to build a library containing one million hours of structured AI training data by the end of next year.
As artificial intelligence continues advancing towards systems capable of understanding and interacting with the physical world, demand for high quality training data is expected to grow rapidly. By transforming licensed gameplay into structured datasets for world model development, Worldmodeldata aims to provide foundational infrastructure for the next generation of intelligent machines across robotics, autonomous systems, and advanced AI research.