Worldmodeldata, a deeptech startup headquartered in Cambridge, is building a vast library of AI training data extracted from video games. The company announced it has raised a £7 million seed round to scale its operations and accelerate its goal of dominating this niche but critical market.
The seed round was led by undisclosed investors, with the company securing £7 million in financing. In a notable board appointment, a former Meta policy VP has joined Worldmodeldata's board, signaling a push for both strategic growth and governance in a space that often skirts ethical and regulatory lines.
Worldmodeldata's core proposition is capturing gameplay data to train AI models — a market with growing demand as developers seek high-quality, diverse datasets beyond static images or text. The startup claims it can reach 1 million hours of data before its rivals, though it faces competition from other specialized data providers and in-house efforts by major game publishers. The broader AI training data market is projected to be worth billions, but sourcing from video games raises questions about licensing and consent.
This funding round signals that investors are betting on synthetic environments as a rich source for AI training, particularly for autonomous systems and simulation models. However, the challenge will be navigating intellectual property rights and ensuring data quality at scale. If Worldmodeldata succeeds, it could become a key infrastructure player for AI companies seeking realistic, varied datasets that are otherwise hard to obtain.
Notably, the former Meta policy VP appointment suggests the startup is preemptively addressing regulatory hurdles. The company's ability to secure license agreements with game developers will be critical — a factor that could either accelerate growth or create bottlenecks against competitors with existing publisher relationships.