A startup backed by investor Vinod Khosla raised $320 million in June 2026, reaching a $2.3 billion valuation, on an unconventional bet: training artificial intelligence on video games and then using it to control real robots. Here is how the model works, and why it could speed up physical robotics.

From game controller inputs to robot sensors

The company, General Intuition, built its AI model from millions of hours of recorded gameplay, capturing not just what happened on screen but also the controller signals behind it: which button was pressed, at what moment, and with what intensity. That training gave the system something close to a spatial and temporal intuition for movement, a capability its creators say is not tied to any single game but transfers to other environments, including physical ones.

A four-legged robot learned to move in minutes, not months

The demo that caught the industry’s attention involved a quadrupedal robot. Instead of the hundreds of thousands of training hours typically required in robotics, the team fine-tuned the model with just eight minutes of real-world data. The robot, equipped with nothing more than a front-facing camera and no additional sensors, successfully navigated an office with people walking and objects moving around, without being specifically prepared for that scenario.

Why this could change the rules of the game

«The generalization of the model itself is the product,» said Pim de Witte, CEO of General Intuition, explaining that the goal is not to build robots but to offer the base model so other companies can build on top of it. The bet is that physical robotics won’t need millions of hours of task-specific training data, and can instead rely on a general-purpose model that already understands movement, drastically cutting the time and cost of getting a robot to work in the real world. If the promise holds, building robots could become as modular as building software on top of existing platforms is today.

At Webmusica we build web solutions with cutting-edge tech. Let’s talk about your project.

Source: TechCrunch.