General Intuition just raised $320 million on a wild bet: that embodied AI will follow the same path as language models. Instead of training robots from scratch with tons of real-world data, CEO Pim de Witte wants to create a general foundation model for physical AI. The company trained its model on millions of hours of video game data, including controller inputs—teaching it intuitive spatial-temporal reasoning. After fine-tuning on just eight minutes of real-world data, their model powered a quadrupedal robot that navigated dynamic environments zero-shot. De Witte argues that most specialized robot work will become redundant once general models emerge. The startup’s endgame isn’t to build robots but to be the base layer for all physical AI. Why it matters: This approach could cut the time and cost of building robots dramatically, democratizing access to embodied AI and potentially accelerating everything from warehouse automation to personal assistants.