General Intuition has raised $220M at a $6.2B valuation to train AI models on video game footage. Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst backed the round.
The startup spun out of Medal, a video sharing app whose users capture and post gameplay clips. That archive of human play became its training set. Instead of filming a robotic arm welding car parts by hand, General Intuition generates synthetic footage of the task.
Its newest model, MIRA, released in June, is built to render far longer scenes than conventional video generators, which typically manage only seconds or minutes. Long sequences matter for factory automation, where a robot has to hold one task together across an entire shift.
World models have attracted heavy capital this year as labs hunt for training data that does not come from humans typing. General Intuition’s answer is to mine a decade of consumer gameplay, where players already solved navigation, timing and coordination problems.
The company is betting synthetic video becomes the default feedstock for physical AI, and that whoever owns the pipeline from gameplay to simulation owns a piece of every robot program built on top of it.
