General Intuition, the New York startup teaching AI agents how to move through space and time, is in talks to raise around $300 million, according to TechCrunch AI. The deal would push the company’s valuation to just over $2 billion. TechCrunch AI reports the round has already pulled in big names, including Jeff Bezos and Eric Schmidt, alongside existing backers Khosla Ventures and General Catalyst.
What stands out here is the speed. General Intuition spun out of Medal, a platform for uploading and sharing video game clips, just eight months ago with a $134 million seed round. Going from seed to a $2 billion valuation in under a year is rare, even by current AI standards.
What the company actually builds
Most of the buzz in the “world model” space is about selling simulations. General Intuition takes a different route. It builds world models to train AI agents, then sells the agents. The models are the means, not the product.
The edge comes from data. The startup trains its embodied AI on Medal’s library of roughly 2 billion video clips a year, pulled from 10 million monthly active users. That footage is first-person, interactive gameplay, which the company argues is the ideal material for teaching machines spatial-temporal reasoning: the ability to perceive a scene, anticipate what happens next, and react in real time.
Founder Pim de Witte, who co-founded Medal, leads the company with co-founders Eloi Alonso, Adam Jelley, and Vincent Micheli, researchers focused on world modeling and simulation.
Why the dataset matters
That clip library is the real asset, and rivals know it. Per TechCrunch AI, OpenAI previously tried to acquire Medal, and other major AI labs have come knocking too. When the biggest labs are circling your training data, that tells you something about its scarcity.
Here is why first-person gameplay is valuable. Training agents to operate in physical or simulated space needs examples of cause and effect over time: move here, this happens; turn there, that appears. Gameplay footage captures exactly that loop at massive scale, far cheaper than recording real-world robotics data.
The competitive picture
The world model race is crowded and getting louder:
- Runway, Decart, and World Labs have all shipped world models recently.
- Google’s Genie 3 started folding in Google Maps data to push toward real-world simulation.
- Most of these players target gaming and robotics training as near-term commercial uses.
General Intuition is betting that owning a hard-to-replicate dataset gives it a defensible path while competitors fight over similar ground. The agents-as-product approach also sidesteps the question of who pays for raw simulation tools.
What comes next
The fresh capital is earmarked for compute. According to a source cited by TechCrunch AI, the company plans to scale up its capacity and ship a new product by the end of summer or early fall.
A few things worth watching:
- The product reveal. A late-summer launch turns the pitch into something you can judge. Until then, the valuation rests on the dataset and the team.
- Data as moat. If proprietary gameplay data proves as decisive as the company claims, expect more labs to chase exclusive footage deals.
- The funding signal. Bezos and Schmidt backing a pre-product world model startup at $2 billion shows how aggressively capital is flowing toward embodied AI and robotics foundations.
This is significant because it marks a shift in where the world model fight is heading. The early contest was about generating convincing simulations. General Intuition is arguing the bigger prize is the agents those simulations can train. If that thesis holds, the company’s quiet pile of gameplay clips may turn out to be one of the more valuable training sets in the field.
More details are available in the original TechCrunch AI report.