Origin Lab just closed an $8M funding round to build a data marketplace specifically for world model training. The startup is positioning itself as the middleman between video game studios sitting on mountains of simulation data and AI labs that need it.
World models are AI systems that learn how physical spaces and objects work, like understanding gravity, collision, or how a door opens. Video games already simulate all of this stuff at scale, which makes game data incredibly valuable for training these models.
The pitch is straightforward. Game companies have been generating this data for years but haven't had a good way to monetize it beyond the games themselves. AI labs need high quality spatial and physics data but don't want to build entire game engines just to generate training sets.
Origin Lab handles the licensing, quality control, and delivery. Game studios get a new revenue stream from assets they've already created. AI companies get access to diverse, realistic simulation data without reinventing the wheel.
This matters because world models are becoming critical infrastructure for robotics, autonomous systems, and spatial AI applications. The companies that can train on the richest datasets will have a significant advantage, and game engines have been quietly generating some of the best spatial data available for decades.
