From Fortnite to robots: General Intuition raises $2.3B on bet that video games can train AI agents for the real world
A quadrupedal robot, its mechanical legs clicking against the polished floor of General Intuition's New York office, moved with the awkward curiosity of a newborn. It had no sense of where it was going, no map, no GPS. Just a single camera and a neural network trained on the same data that let an AI agent play Fortnite for 100 consecutive hours. The robot circled a chair, bumped into a trash can, and kept walking. This was not a demonstration of perfection—it was a glimpse into the future of AI, one where the virtual and the physical are no longer separate.
From virtual worlds to real-world intelligence
General Intuition’s core innovation lies in its ability to train agentic AI models using data from video games. Unlike traditional AI systems, which rely on static datasets or narrow task-specific training, the company’s models learn from the dynamic, multi-layered interactions found in gameplay. This includes not just visual input, but the precise timing and sequence of actions—what buttons were pressed, when, and how they translated into movement or interaction.
The company’s proprietary data comes from Medal, a platform where gamers upload gameplay clips. These clips include action labels—records of player inputs—that provide a level of detail traditional video analysis lacks. This combination of visual and behavioral data allows General Intuition’s models to better understand cause and effect in both simulated and real environments.
A new era of AI pre-training
The startup’s model is designed to be a world model, capable of simulating and responding to complex environments. It's not just about recognizing objects or predicting motion—it's about understanding how to interact with them. This is what makes it particularly valuable in robotics, where the ability to adapt to unforeseen obstacles or environmental changes is critical.
General Intuition has already demonstrated its models in a range of scenarios, from driving simulations to robot navigation. The company’s latest funding round, led by Khosla Ventures, underscores the confidence investors have in this approach. With $454 million raised so far, the startup plans to scale its compute infrastructure and expand access to its API.
The implications are vast. If an AI can learn the rules of a virtual world and apply them to the physical one, it could revolutionize fields like logistics, manufacturing, and even healthcare. But the challenge remains: can this level of generalization hold up in the real world, where environments are unpredictable and data is scarce?
Ethics and the future of work
Pim de Witte, General Intuition’s co-founder and CEO, is acutely aware of the power his company wields. He has drawn a clear line: no applications that could harm humans. This stance is rooted in his past work with humanitarian organizations and a desire to avoid becoming a tool for military escalation.
De Witte also sees an opportunity to create economic value for gamers, who are often the first to face the disruptions of AI. His platform, Nerve, is a jobs marketplace where gamers can label data or operate robots using their existing setups. This approach is a deliberate attempt to ensure that the people who built the data that powers AI systems aren’t left behind.
As the world grapples with the rise of general AI, General Intuition is betting that video games—once seen as a niche form of entertainment—will be the key to unlocking human-like intelligence in machines. Whether or not it wins that bet will depend on its ability to scale its model, prove its effectiveness in the real world, and navigate the ethical and economic challenges that come with it.
The stakes are high, but the path is clear. If General Intuition can bridge the gap between virtual and physical intelligence, it may not just be a breakthrough in AI—it could be the foundation of a new era in technology.
