A startup spun out of a gaming clip platform has landed one of the largest artificial intelligence funding rounds of the year. General Intuition has raised $320 million in a Series A round at a $2.3 billion valuation, betting that hundreds of millions of hours of player-annotated gameplay can teach AI agents how to move and act in the real world.
TechCrunch first reported the round on June 25. It was led by Khosla Ventures, with General Catalyst, Jeff Bezos, Eric Schmidt, former Formula 1 driver Nico Rosberg, and researchers from Google DeepMind and MIT also participating. The raise brings the company's disclosed funding to roughly $454 million, following a $134 million launch round in October 2025. For more context on this story, see our ongoing artificial intelligence updates.
The company's thesis is unconventional but increasingly fashionable: that the data sitting inside a consumer gaming product may be one of the most underrated training sources for physical AI.
From Clip-Sharing to Robotics Data
General Intuition was built out of Medal, a clip-sharing platform for gamers founded by Pim de Witte. According to Startup Fortune, de Witte turned down a reported $500 million acquisition offer from OpenAI in late 2024, concluding that the footage beneath his platform was worth more as the foundation for a new AI company than as a line item inside another lab.
Medal has roughly 10 million monthly active users who upload about 2 billion video clips a year. The critical detail is what those clips contain beyond video. A Medal clip can carry the controller input, the button press, and the mouse movement that produced the action on screen. That is precisely the kind of data that robotics labs struggle to capture at scale: not only what happened, but what a human did to make it happen.
The company's argument is that if a model can watch a player navigate a cluttered fight and predict the next frame from their inputs, it may be learning something genuinely useful about space, timing, intention, and consequence. A warehouse robot does not need to win a match. It needs to move through an awkward physical environment without colliding with a shelf, a cart, or a person.
Why Investors Are Paying Attention
The investor list reads as a who's-who of physical AI believers. Jeff Bezos previously backed Physical Intelligence, the robotics software company that raised $400 million in 2024 at a valuation above $2 billion. Eric Schmidt has spoken openly about robotics as one of the next major technology markets. Google DeepMind has spent years advancing world models and agents that reason about physical space.
The wider market gives de Witte room to make the bet. Business Insider recently reported that venture funding for robotics climbed from $4 billion in 2019 to $26 billion in 2025. Companies including Figure AI, Physical Intelligence, Skild AI, and Agility Robotics have all attracted significant capital as money shifts toward machines that can act in the physical world rather than merely answer questions in a text box. Nvidia chief executive Jensen Huang has repeatedly described embodied AI as the next wave.
The Compute Bet and What Comes Next
Most of the new capital is earmarked for compute. TechCrunch reported that the next version of General Intuition's model will be trained on CoreWeave infrastructure, with a public API planned by the end of summer 2026. The API will be the first real test of whether the Medal thesis holds up outside of a pitch deck and a demonstration.
The obvious objection is that games are not the real world. A game engine enforces clean rules that a factory floor never does. Collision detection is precise, objects behave according to code, and a robot arm operating near a human worker must contend with weight, friction, poor lighting, and latency. General Intuition's counter is that scale changes the equation. The company is not claiming a video game is a factory. It is betting that hundreds of millions of genuine human decisions, attached to visible outcomes, can help a model build the spatial-temporal instincts that robotics still lacks.
A Narrower, More Interesting Story
The useful point is narrower and more interesting than the headline. One of the most potentially valuable datasets for physical AI may have been hiding inside a consumer gaming product while the largest labs were looking somewhere cleaner. Medal's clips contain real human choices made under pressure, not trajectories generated by a simulator searching for tidy paths. For an agent learning to anticipate what a person might do next, that distinction matters.
General Intuition still has to prove the hard part. A large dataset can secure funding, but it cannot by itself make a robot useful. If the summer API shows that gaming-trained agents generalize to real-world tasks, de Witte's decision to walk away from a $500 million OpenAI offer may look prescient. If it does not, the round will stand as an expensive reminder that the world is harder to model than a game.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
