Medal spinoff General Intuition raises $320 million for AGI push

General Intuition raised $320 million at a $2.3 billion valuation to build AI models on action-labeled gameplay clips, a wager that game worlds can teach machines spatial reasoning better than the open internet. The startup was spun out of Medal, the clip-sharing platform founded by Pim de Witte, and is pitching a model family built to understand how things move through space and time.

The company says its systems learn from unique, action-labeled video datasets across countless environments, not just raw footage. That distinction matters to its thesis: video games supply feedback, goals and simulated physics, giving a model something the open internet usually does not. In General Intuition’s view, that kind of data can help AI systems improve perception, movement and decision-making in fast-changing 3D settings, skills the lab says are central to what it calls spatial intelligence.

Medal gives the effort a large data pipeline. The company says it has more than 12 million users and over 3 billion clips captured, a scale General Intuition is using as the backbone of its training data. Medal’s blog says General Intuition is organized as a Public Benefit Corporation, and it says users can opt in or out of having their data used for research, with the company arguing that player behavior and gameplay actions can be removed from future use if people choose not to participate.

The financing round drew a crowded list of backers. Reporting said Khosla Ventures led the deal, with Jeff Bezos, Eric Schmidt, General Catalyst, Nico Rosberg, and researchers linked to Google DeepMind and MIT also taking part. After the new investment, General Intuition’s total disclosed funding was reported at $454 million, following an earlier seed round of $133.7 million that closed around October 2025.

The company’s pitch has drawn attention because it sits at the intersection of two crowded AI bets: world models and robotics. General Intuition says it wants models that can act across space and time rather than only generate text or images, and earlier coverage said it is working on navigation in unfamiliar physical environments. That is the promise and the test. Game footage may be a cleaner training environment than the open web, but whether clip-labeled play can produce truly general intelligence, or only another expensive rebrand of AGI ambition, will depend on how far those virtual skills carry into the physical world.