General Intuition is trying to turn Medal's gaming clips into the training data AI agents lack, and investors are treating that dataset as a serious moat.
General Intuition's new fundraising talks are really a story about data. TechCrunch reported on June 18 that the lab is in talks to raise $300 million at roughly a $2 billion valuation, only eight months after it announced a $133.7 million seed round led mainly by Khosla Ventures, with General Catalyst and Raine Group also involved. That is not normal seed-to-Series-A behavior. It tells you investors are not just buying another model pitch. They are buying access to a feed of human play, mistakes, reactions and movement that most AI labs cannot easily copy.
The company came out of Medal, Pim de Witte's gameplay clipping platform. According to reporting by Alex Heath, Medal receives roughly 2 billion video uploads a year from tens of thousands of games. That is the useful part. A gaming clip does not just show pixels moving around a screen. It shows a player deciding when to turn, when to run, when to wait, when to dodge and when to fail. If you are trying to train systems that understand space and time, you want that kind of messy, repeated action at scale.
You can see why this is more interesting than a lab promising better chat. Large language models are strong at text, but they do not naturally understand what happens when an object falls, a character rounds a corner, or a drone has to move through a partially blocked space. General Intuition's bet is that games are one of the few places where you can observe decisions inside environments governed by rules. That does not make a game the same as the real world. It does make it a better training ground than another pile of scraped text.
De Witte is not doing this alone. Moritz Baier-Lentz, the former Lightspeed gaming investor, is part of the founding team, and Vinod Khosla has treated the company as a large bet on what comes after LLMs. The Verge reported in October 2025 that General Intuition was spun out after AI labs showed interest in Medal's data, with OpenAI reported elsewhere to have offered $500 million for the startup. Whether or not you believe every world model pitch, that sequence matters. When buyers appear before a company has even fully explained itself to the market, the asset is probably not ordinary.
General Intuition also is not raising into an empty field. The Financial Times reported this week that Odyssey raised $310 million in a round valuing the company at $1.45 billion, with Amazon among the backers and other participants including Nvidia, AMD, IQT and GV. Yann LeCun's AMI Labs announced a $1.03 billion seed round in March at a $3.5 billion pre-money valuation. World Labs, the spatial intelligence company founded by Fei-Fei Li, has also pulled in heavy capital. These companies are not all building the same thing, but they are circling the same weakness in today's AI stack: language alone will not give you reliable agents.
Frankly, that is the part founders should pay attention to. Investors spent the last few years rewarding whoever could put a better interface on top of foundation models. That trade is getting crowded. The new money is moving toward owners of scarce inputs: compute, distribution and data that teaches models something text cannot. Medal gives General Intuition one of those inputs. If the platform keeps growing, the dataset compounds without the company having to beg publishers, streamers or web platforms for access.
The first use cases are still early. General Intuition has pointed to non-player characters in games and search-and-rescue drones, two areas where scripted behavior breaks quickly once the environment changes. A better NPC is useful. A drone that can reason through an unfamiliar building is a much harder prize. You should be skeptical of any straight line from gaming clips to real-world autonomy, because the real world has friction, weather, bad sensors and consequences games do not. But skepticism is not the same as dismissal. The best AI companies often start by finding the one dataset everyone else underestimated.
The risk is obvious. OpenAI, Google DeepMind, Anthropic and Meta can all spend their way into new data partnerships if they decide world models are the next priority. They can buy studios, license video, hire research teams and tolerate years of losses. General Intuition's defense is timing. It already has a live platform creating clips, a research lab built around that supply and investors willing to fund the gap before revenue catches up.
A $2 billion valuation before a broad commercial rollout is not proof the company will win. It is proof that gaming data has moved from entertainment exhaust to AI infrastructure. If General Intuition turns Medal's clips into models that can act across unfamiliar environments, the company will not look expensive for long. If it cannot, this round will become another reminder that rare data still has to become useful product.
Also read: The AI spending arms race is quietly ending the era of Big Tech buybacks, Accenture's revenue miss and guidance cut signal a structural reckoning for enterprise IT consulting, and Google DeepMind puts $10 million behind the safety problem that agentic AI has been ignoring
