General Intuition Raises $220 Million at $6.2 Billion to Turn Gameplay Footage Into Training Grounds for Physical AI

General Intuition, a world-model startup that spins video game footage into training data for embodied AI, announced on Tuesday that it has raised $220 million at a $6.2 billion valuation. The round was provided by Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst.

The company spun out a year ago from Medal, the free gameplay-capture tool co-founded by General Intuition's chief executive Pim de Witte, and Medal remains its data engine. Medal says it is on track for three billion video uploads a year, and the startup's pitch rests on what makes that footage unusual: it arrives paired with the player's controller inputs, giving models both what happened on screen and the action that caused it. (Those figures are the company's own.)

The new price is a steep climb. General Intuition raised a $133.7 million seed in October 2025 led by Khosla Ventures and General Catalyst, then a $320 million round on June 25 at a $2.3 billion valuation. Disclosed funding now totals roughly $673.7 million, and the valuation has nearly tripled in about three months — TechCrunch reported in August that the company was finalizing a round at a $6 billion pre-money valuation.

The company's showcase is MIRA, a research world model built with French lab Kyutai in collaboration with Epic Games. Trained on about 10,000 hours of gameplay, MIRA renders a playable four-player Rocket League match in real time at 20 frames per second, responding to the keys each player presses. It packs just 5.6 billion parameters and, because it works in latent space rather than raw pixels, General Intuition says it can run "infinitely without diverging" — no drifting or dissolving over long sessions — while rendering 720-by-576 video on a single Nvidia B200. Those performance figures are company-reported.

Monetization is beginning. General Intuition is testing a commercial version of its technology with a limited set of customers in robotics, simulation and entertainment, and opened a waitlist alongside the funding announcement. The company frames its work as two coupled systems: action models that choose what to do, and world models that predict what follows — the pair it believes can teach machines to act in physical space before those models are broadly available.

The valuation prices a wager, not a product. MIRA demonstrates that a model can learn the physics of one game; it does not yet prove that skills transfer to a robot in an unfamiliar factory, which is the claim the $6.2 billion rests on. In June, TechCrunch watched a General Intuition model steer a quadruped robot around the company's office after a small amount of real-world fine-tuning — encouraging, but far from solved. With AMD paying $8.2 billion for World Labs and Nvidia pushing Cosmos, the race to supply physical AI with synthetic experience is officially crowded, and General Intuition just priced itself near the front of it.