General Intuition’s $2.3B bet that video games can train AI agents for the real world

BitcoinWorld General Intuition’s $2.3B bet that video games can train AI agents for the real world General Intuition, a startup spun out of gaming clip platform Medal, has raised $320 million at a $2.3 billion valuation to pursue an ambitious thesis: that the action data embedded in millions of hours of video game footage can teach AI agents to navigate the physical world. The round, led by Khosla Ventures with participation from General Catalyst, Jeff Bezos, Eric Schmidt, and researchers from Google DeepMind and MIT, signals strong investor belief that gameplay offers a scalable shortcut to building generalized AI models for robotics and simulation. From Fortnite to factory floors During a visit to General Intuition’s New York office, co-founder and CEO Pim de Witte demonstrated the company’s technology in action. An AI agent had been playing a game similar to Fortnite for 100 hours straight, learning spatial-temporal reasoning — understanding how to move through space and time. The same model was then powering a quadrupedal robot that navigated the office, bumping into chairs and trash bins like a learning toddler. De Witte explained that just eight minutes of real-world data collected on a street was enough to fine-tune the model for the robot’s new environment. The key differentiator, according to de Witte, is the action labels embedded in Medal’s gameplay clips — records of exactly which buttons players pressed and when. Most competitors try to infer actions from video alone, which he argues is insufficient for building a model that understands causality and the distinction between self and environment. A data moat built on gaming General Intuition’s data advantage comes from Medal, de Witte’s previous company, which hosts hundreds of millions of hours of uploaded gameplay. This proprietary dataset provides the foundation for training world models — simulated environments generated frame-by-frame rather than rendered by a traditional game engine. In a demo, a world model correctly treated walls as solid objects, ladders as climbable, and shadows as dynamic, demonstrating an understanding of physics learned purely from gameplay. The company is not selling the world model itself; instead, it uses the simulation as a training environment, or “the gym,” to improve its agentic model. The ultimate product is an API that allows customers to deploy the model in gaming, simulation, and robotics. De Witte emphasized that General Intuition will not build a self-driving car company but aims to make it “10 times easier for the next person to build a self-driving car company.” Investor conviction and ethical boundaries Vinod Khosla, whose firm led the round, described General Intuition as a generational bet, comparing the potential emergence of intuition in world models to the quantum leap of reasoning in large language models.