Signal A gaming platform's video archive becomes a training asset worth rejecting a 500 million dollar offer

Summary

Building world models requires rich, multimodal training data from real environments, and companies that control unique data sources gain an advantage. CB Insights illustrates this with Medal, a gaming platform based in the Netherlands. Medal collects around 2 billion video clips a year from 10 million monthly users, which the firm describes as a substantial data moat. After rejecting a 500 million dollar acquisition offer from OpenAI, Medal spun out a world models lab called General Intuition. The spinout uses Medal's gameplay videos to train agents for spatial reasoning, adaptive game bots and autonomous navigation. The case shows how unusual data archives can become strategic assets in this field.