On a quiet floor in a New York office, a quadrupedal robot is learning to walk. It doesn't use traditional sensors or pre-programmed paths. Instead, it relies on a brain trained by thousands of hours of Fortnite gameplay. This is the core of General Intuition, a startup that just secured a $2.3 billion valuation to prove that the best way to teach a machine about the real world is to let it play video games.

General Intuition announced a $320 million funding round on Thursday, led by Khosla Ventures. The capital arrives as the company attempts to solve one of robotics' hardest problems: generalization. Most AI models struggle when they leave the screen. They lack a sense of physical causality. General Intuition believes it has found the shortcut.

The Gameplay Shortcut

For years, AI researchers have treated video games as a sandbox. General Intuition is taking this further. By leveraging data from Medal, a platform where gamers upload clips, the company has access to millions of hours of footage. Crucially, this data includes action labels—the exact button presses a player made in response to what they saw on screen.

Most competitors try to infer intent from video alone. That is a mistake. By pairing visual input with concrete action data, General Intuition’s model learns the relationship between movement and consequence. It learns that walls are solid. It learns that gravity is constant. It learns the difference between the 'self' and the 'environment.'

From Virtual Walls to Real Floors

During a recent demonstration, the company’s model navigated a simulated environment frame-by-frame. It didn't clip through walls. It didn't ignore physics. It behaved with a spatial awareness that felt eerily human. When that same model was ported to a physical robot, the results were immediate. The bot required only eight minutes of real-world data to begin navigating the office floor. It bumped into a trash bin, corrected itself, and kept moving. It was learning.

This is the promise of the 'gym.' The company uses its virtual world model as a training ground for agents that will eventually operate in the physical world. If the model can master the complex, high-speed dynamics of a shooter game, it should, in theory, be able to navigate a warehouse or a living room.

The Heavy Hitters Behind the Bet

Investors are betting on the data. The $320 million round includes participation from General Catalyst, Jeff Bezos, Eric Schmidt, and researchers from Google DeepMind. Vinod Khosla, who led the round, views this as a fundamental shift. He calls it the emergence of 'intuition' in AI.

Scaling this will be expensive. The vast majority of the new capital is earmarked for compute capacity. The company has secured a deal with CoreWeave to power the pre-training of its next model. They are not building a game. They are building a brain.

What This Means for Developers

General Intuition plans to open its API to a wider audience by the end of this summer. For developers, this represents a potential leap in agentic capability. If the model proves robust, it could lower the barrier to entry for building autonomous systems that don't require months of custom training data.

Key Takeaways

  • General Intuition raised $320 million at a $2.3 billion valuation to train AI agents using video game action data.
  • The company argues that pairing visual data with specific button-press labels allows AI to learn physical causality better than video-only training.
  • The startup plans to release its API by the end of summer, aiming to provide a scalable training foundation for real-world robotics.

The Road Ahead

Scaling is the hurdle. While the demos are impressive, the jump from a controlled office environment to the chaotic, unpredictable real world is massive. The company’s next version of the model will face that test. By the time the API launches in a few months, we will know if this 'gameplay shortcut' holds up under pressure. The industry is watching. The robot is walking. It is time to see if it can run.