General Intuition raised a $320 million Series A at roughly a $2.3 billion valuation, announced June 25, 2026, led by General Catalyst; total funding to date is about $454 million. The company trains AI on action-labeled gameplay footage sourced from Medal.tv, building 'action models' that decide what to do and 'world models' that predict what happens next.
The bet is a specific data advantage: gameplay clips carry embedded labels recording exactly when a player pressed which button, so they encode decision-making, not just pixels. That is the ingredient missing from web text and expensive to gather from real-world teleoperation, and it is why world-model approaches keep attracting nine-figure rounds.
If robots can learn transferable skills from games, the industry's most painful bottleneck — collecting real demonstration data — gets cheaper. The size of the round, alongside NVIDIA's Cosmos world-model push, marks world models as one of the biggest research bets in physical AI. The open question is how cleanly game-world skills transfer to physical bodies.
Key Facts
- $320M Series A at a ~$2.3B valuation; announced June 25, 2026
- Led by General Catalyst; ~$454M total funding to date
- Trains on action-labeled gameplay footage (via Medal.tv)
- Builds both action models and world models
- Aimed at real-world agents and robots
Frequently Asked
How much did General Intuition raise?
A $320 million Series A at roughly a $2.3 billion valuation, announced June 25, 2026, led by General Catalyst, bringing total funding to about $454 million.
What is the company's approach?
It trains AI on action-labeled gameplay footage — clips that record exactly when players pressed which buttons — to build action models (what to do) and world models (what happens next) for real-world agents and robots.
Why does gameplay data matter?
Gameplay encodes decision-making, not just images, which is missing from web text and costly to gather from real-world teleoperation — a potential shortcut around robotics' data bottleneck.