General Intuition raises $320M to train robots through videogames

Signals Inbox·June 26, 2026·World Models

General Intuition has raised $320 million to train AI on something robot labs badly need: huge amounts of human action data. Its source is unusual. Billions of gaming clips show not only what players saw, but exactly which buttons they pressed next. The bet is that this can teach machines how actions change a world before those machines spend years learning through expensive physical robots.

Q1What actually happened?

General Intuition officially announced a $320 million Series A to build models that can perceive, predict, and act in virtual and physical environments. The round values the company at $2.3 billion and brings its disclosed funding to roughly $454 million since it launched in October 2025.

Q2Why are videogames useful for training robots?

Because games contain endless examples of people seeing something, choosing an action, and watching what happens next. General Intuition gets more than video from Medal, its gaming clip platform. It also gets action labels showing which keys or controller buttons players pressed and when. That gives the model a clearer link between intent, action, and consequence.

Q3Why not just train robots in the real world?

Real-world robot data is painfully slow and expensive. A machine must be purchased, operated, repaired, supervised, and kept away from dangerous mistakes. Games let models experience huge numbers of actions cheaply and safely. General Intuition is betting that this virtual experience can reduce how much physical training a robot needs later.

Q4Has it actually worked on a robot?

In an early demonstration, the company used the same underlying model to control a game agent and a quadruped robot. It said the robot needed only eight minutes of real-world data before navigating an unfamiliar office through a camera. That is a striking result, but it is still a demonstration, not proof that the system works reliably across factories, homes, and thousands of machines.

Q5What makes this different from normal simulation?

Most robot simulations are designed by engineers and cover a limited set of environments. General Intuition starts with real human behavior spread across countless games, goals, maps, mistakes, and playing styles. Medal says users upload billions of clips each year. The company believes that variety can produce broader instincts than training one robot on one task inside one simulated warehouse.

Q6Why is the funding round so large?

Because this is a frontier-model bet, not a normal robotics app. Most of the money will pay for compute and the next round of pretraining through CoreWeave. The company wants to sell its model through an API to robotics, simulation, and gaming businesses. In other words, it wants to supply the robot brain rather than build every robot itself.

Q7So what is the real signal?

The valuable training data for physical AI may already exist outside robotics labs. Text created the first wave of foundation models. General Intuition thinks recorded human actions could power the next one. The tension is that gaming data is cheap and enormous, but real machines obey messier physics than videogames. This $320 million round is funding the test of whether that gap can actually be crossed.