General Intuition is a New York artificial intelligence lab building foundation models that learn to act by training on video game footage. Its bet is that gameplay, recorded at massive scale and paired with the exact inputs a player made, is the fastest path to AI that can operate in the physical world. The company drew wide attention in 2026 when it raised $320 million in Series A funding at a $2.3 billion valuation, one of the largest early rounds in the emerging field of physical AI.
What General Intuition does
The company trains models on hundreds of millions of hours of gameplay clips that carry embedded action labels, meaning each frame is paired with the buttons a player pressed at that moment. Founder and chief executive Pim de Witte argues that this pairing of pixels with inputs, not the footage alone, is the real advantage, because it lets a model learn cause and effect rather than infer actions from video.
That dataset came from Medal, the gaming clip-sharing platform de Witte built, which General Intuition was spun out of. Inside the company, a world model the team calls "the gym" generates playable scenes frame by frame instead of rendering them with a traditional game engine, and the agent that learns inside it is the product the company intends to sell.
From game controller to robot
General Intuition's goal is a single model that can move from playing a game to navigating a simulation to controlling a real machine. In demonstrations, an agent that had played a shooter-style game for a hundred hours was shown driving a four-legged robot after only minutes of real-world fine-tuning, a sign of how the company hopes gameplay can substitute for slow, expensive physical data collection.
De Witte has framed the company as a platform rather than a product business, comparing his ambition to the role model providers play for software developers. The wager is that a foundation model for action and world simulation could become infrastructure that startups in gaming, logistics, manufacturing, and robotics build on top of.
Funding and backing
The Series A was led by Khosla Ventures, with participation from General Catalyst, Bezos Expeditions, Eric Schmidt, Innovation Endeavors, and others, bringing total disclosed funding to $454 million after a $134 million seed at launch in 2025. Much of the capital is earmarked for compute through a deal with CoreWeave to pre-train the next version of the model.
Investors point to the company's proprietary access to human action data through Medal as a defensible edge. The team has also taken an unusual public stance on values, ruling out lethal autonomy and launching Nerve, a marketplace that pays gamers for tasks such as data labeling and robot teleoperation.
Frequently asked questions
What does General Intuition do?
It is a New York AI lab that trains foundation models on action-labeled video game footage, aiming to build a single model that can play games, navigate simulations, and control real machines such as robots.
How much did General Intuition raise?
The company raised $320 million in a Series A round at a $2.3 billion valuation, led by Khosla Ventures, bringing its total disclosed funding to $454 million.
Who founded General Intuition?
It was founded by Pim de Witte and spun out of Medal, his gaming clip-sharing platform, which supplied the gameplay dataset behind the company's models.
Why train AI on video games?
Gameplay clips include action labels that record what happened on screen and which buttons the player pressed, letting the model learn cause and effect more directly than from video alone, and potentially replacing slow real-world data collection.
What is "physical AI"?
Physical AI refers to systems built to act in the real world, such as robots and embodied agents, rather than only generating text or images. General Intuition is among the startups attracting large investment in the category.