General Intuition raises $320M at $2.3B valuation to train AI agents using video game data

General Intuition, a New York-based startup, raised $320 million in a new funding round that values the company at $2.3 billion, according to TechCrunch’s report. The round was led by Khosla Ventures, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers affiliated with Google DeepMind and MIT. The company’s total disclosed funding now stands at $454 million, following a $134 million seed round in October of the previous year.

The startup was spun out of Medal, a company co-founded by de Witte that allows gamers to upload and share video game clips. General Intuition uses hundreds of millions of hours of uploaded gameplay to train its models in spatial-temporal reasoning—understanding how to move through space and time. The company argues that the key to its approach is the embedded action labels in gameplay clips, which record exactly what buttons players pressed and when. This data is used to train models that can generalize from gameplay to simulation to embodied robotics.

In a live demonstration, a quadrupedal robot powered by General Intuition’s model navigated an office environment after just eight minutes of real-world fine-tuning. The same model was shown playing a game similar to Fortnite for 100 hours straight. Kent Rollins, the company’s chief product officer, stated that the agent controlling the game was the same one powering the robot. The robot’s default mode was described as “exploration,” relying on a single camera feed to navigate and avoid obstacles, much like a learning toddler.

The company’s world model—a simulated environment generated frame-by-frame—was also demonstrated. In one test, the model correctly identified walls as impassable and ladders as climbable, and showed dynamic lighting effects such as shadows lengthening as the sun moved. General Intuition refers to this environment internally as “the gym,” positioning it as a training ground rather than a final product. The company plans to sell the agentic model itself, arguing that the action data embedded in gameplay helps the model discern the “self” from the “environment,” enabling richer causal understanding.

General Intuition is not alone in exploring the use of video games and human action data to build general agents. However, the company’s approach differentiates itself by focusing on embedded action labels rather than inferring actions from video alone. The company’s proprietary dataset, derived from Medal’s platform, is a core asset that investors cite as a key reason for their confidence in the company’s long-term potential.

The company’s leadership has also set ethical boundaries, explicitly ruling out military applications that could harm humans. Pim de Witte, the CEO and co-founder, emphasized this stance, stating that the company does not want to be part of systems that enable lethal autonomy. He suggested that such a stance is particularly relevant given broader industry trends in defense and military AI.

The new funding will primarily be directed toward scaling compute capacity, with a portion allocated to making the company’s API more broadly available by the end of summer. General Intuition also has a deal with CoreWeave to support its compute needs as it pre-trains the next version of its model.