General Intuition Raises $320M at a $2.3B Valuation to Train AI Agents on Video Games

General Intuition raised a new $320 million round at a $2.3 billion valuation, TechCrunch reported on June 25, 2026. The company says it trains AI agents on gameplay footage from its parent company Medal, building a single model that responds to a Fortnite screen while also handling real-world physics. Khosla Ventures led the round and Jeff Bezos and Eric Schmidt joined, putting serious money behind the bet that gameplay can become training data for robots. This article organizes only the verifiable facts on the basis of TechCrunch's primary reporting.

Who Put In the Money

General Intuition raised $320 million in a round led by Khosla Ventures, earning a $2.3 billion valuation. Counting the $134 million seed from October 2025, total disclosed funding reaches $454 million. Participants included General Catalyst, Jeff Bezos, Eric Schmidt, former F1 champion Nico Rosberg, and researchers from Google DeepMind and MIT. CEO Pim de Witte, 31, co-founded the company with Eloi Alonso, Adam Jelley, and Vincent Micheli.

Why Gameplay Footage Becomes a Robot Textbook

General Intuition trains on millions of hours of gameplay footage held by its parent company, Medal. The clips carry action labels such as button presses and their timing, so the model learns spatial-temporal reasoning that links on-screen change to control inputs, the company says. De Witte said, "We have a single model that can respond to Fortnite information on the screen and take action, but also to real-world dynamics." The resulting world model picks up physical rules on its own, learning that walls are solid, ladders can be climbed, and shadows lengthen.

The Weight of That 8-Minute Number

General Intuition says moving a game-trained agent onto a quadrupedal robot required only 8 minutes of real-world robotics data for fine-tuning. The company is testing the approach on drones, quadrupeds, and driving simulations, and plans to open a developer API by the end of summer. It also launched Nerve, a marketplace that pays gamers to label data and teleoperate robots. The idea is that game footage can partly replace the expensive real-world data collection that robotics usually demands.

Why This Bet Looks Large (Interpretation)

Robotics' biggest bottleneck is data, not algorithms. Collecting data by physically running robots in the real world is slow, expensive, and risky. General Intuition's logic reads as an attempt to route around that bottleneck using games, a body of large-scale data that already exists. The presence of names like Bezos and Schmidt can be read as a bet that a world model could reshape the very supply chain of training data for robotics. In other words, this round is a wager on how data is sourced, not on any single robot's performance.

What to Watch From a Korean Vantage Point (Interpretation)

Korea is a rare market that holds game IP, vast play logs, and robotics and mobility manufacturing capacity all at once. If the view that gameplay footage is a robot-learning asset holds up, one can read room for domestic game studios' play data to combine with the robotics ambitions of automakers and appliance firms. That said, a structure like Nerve, which commercializes gamers' control data, is exactly where Korea would need to newly settle questions of data rights and consent. The design of data contracts, more than the technical transfer, looks like the crux.

Where to Read Cautiously Anyway

General Intuition's performance figures are largely company claims, and independently verified benchmarks have not yet been published. Whether an 8-minute fine-tune that drove a quadruped will generalize across varied robots and environments remains to be seen. The gap between in-game physics and real-world physics, along with the early stage of commercial revenue, are also worth weighing. The $2.3 billion figure is a valuation, not the amount raised in this round, and should be read as such.

Sources: TechCrunch