General Intuition raised $320 million at a $2.3 billion valuation, betting that video game data is the key to building the next generation of physical AI.
CEO Pim de Witte argues that robotics is about to have its ChatGPT moment—and Vinod Khosla is investing hundreds of millions on that thesis. The idea: build one foundation model capable of understanding movement and spatial reasoning across any environment, rather than teaching each robot from scratch with millions of hours of real-world footage.
General Intuition trained its model on millions of hours of video game data—tracking controller inputs, button presses, timing—to teach machines how humans intuitively understand space and time. They then fine-tuned that model on just eight minutes of real-world robot footage. A quadrupedal robot could navigate an office, avoid people, and adapt to dynamic objects with zero additional training.
Robotics companies currently waste resources building bespoke models for individual tasks, environments, and machines. De Witte argues most of that work will become obsolete within the next few years. A pre-trained foundation model can reach 80% of the way there in minutes instead of requiring millions of hours of expensive real-world data.
Foundation models democratized AI. When GPT-3 launched, any startup with an idea could fine-tune a powerful model instead of starting from zero. If General Intuition succeeds, the same transformation happens to robotics—and the barrier to entry for building the next autonomous vehicle, warehouse robot, or humanoid drops significantly.
De Witte is not interested in building robots himself. General Intuition wants to be the foundation—the base layer that robotics startups build on top of, much like how every AI company now starts with OpenAI, Anthropic, or Meta's models.
Whether this actually works is an open question. With Khosla's backing and a working demo, the robotics world is watching. The ChatGPT moment for physical AI may be arriving now.
