General Intuition raises $220M to train robot brains on 3 billion gaming videos per year

General Intuition has raised $220 million at a $6 billion valuation, with the round anchored by proof that gaming data can transfer to real-world robot control.

The core bet

Pim de Witte, General Intuition's co-founder and CEO, describes the company as building the largest video game recorder in the world, capturing how humans control characters inside games and using that data to pre-train robot foundation models. The logic is direct: robots and game controllers share the same input language. When a player uses a controller to move a character in first-person view, the perception-to-action loop is structurally identical to teleoperating a robot. General Intuition maps those inputs to robot actions across pixel space, turning what would otherwise be an expensive robotics data problem into a data abundance problem.

The analogy to LLMs is deliberate. Just as language models are pre-trained on predicting the next text token from internet-scale text, General Intuition pre-trains on predicting actions from frames, using the roughly 3 billion gaming videos people capture and upload every year. The result, de Witte argues, is a foundation model capable of controlling any robot that ships with a game controller, which he says covers the majority of robots currently in the market.

“We built the largest video game recorder in the world, and so we get a lot of data on how people control embodiments inside video games. LLMs go text to text. We go frames to actions, and then we can scale that into extremely general models that can control every robot that ships with the game controller, which is the majority of robots right now.”

Transfer, proved

What unlocked this round was demonstrating that the transfer actually works. De Witte says recent demos have shown the model can generalize from game environments to physical-world robotics, validating the core hypothesis that pixel-to-action scaling holds across different environments even when the pixels change significantly.

Near-term focus

General Intuition is targeting quadrupeds and drones in the near term, not humanoids. De Witte says humanoids remain mechanically unstable and their supply chains haven't scaled, making them a harder near-term commercial target. The company spent 2025 scaling with customers in simulation and plans to shift toward physical-world transfer in 2026. De Witte's view is that competitors betting exclusively on physical-world robotics are likely to miss an inflection point, because every robotics manufacturer runs simulation testing before real-world deployment, and that's where General Intuition is building its customer base first.

Gaming as a second market

Beyond robotics, de Witte sees a natural product emerging in gaming itself. Player liquidity is the central economic problem for any live game, and AI companions that can fill squads, make callouts, and match a player's skill level would address that directly. He expects game developers to push toward richer, interactive AI-populated environments because the economics favor it, and suggests AI companions will start out looking like better NPCs before the use case matures. He frames the need for clear labeling of AI participants as an open question the industry will have to work through.

On gamer skepticism toward AI more broadly, de Witte says the hostility was rational: GPU prices rose, game studios delayed releases, and developers were laid off while publishers directed investment toward AI with little visible benefit to players. He thinks that's starting to change, pointing to modding becoming significantly easier and new types of game interactions becoming possible on top of these models.

General Intuition's commercial thesis depends on hitting the simulation-to-physical transfer milestone in 2026 and being positioned as the foundation layer when robotics deployment scales. The $6 billion valuation is a large number to grow into, and the key variable is whether pixel-space transfer holds as robustly in messy physical environments as the current demos suggest.