A Startup That Trains AI Using Video Game Footage Just Tripled Its Valuation

The quick version

A startup called General Intuition is reportedly in talks to raise a new round of funding that would value the company at $6 billion — before the new money even comes in. That's nearly three times what it was worth just a few weeks earlier, when it raised $320 million at a $2.3 billion valuation. New backers said to be joining include Valor Equity Partners, Point72 Ventures, and Seven Seven Six, on top of existing investors Khosla Ventures and General Catalyst.

What does General Intuition actually build?

The company builds "foundation models" — big, general-purpose AI systems — that are meant to teach AI agents and robots how to move through space and time the way humans and animals intuitively do. Think of it as trying to give a robot a sense of physical common sense: how to navigate a room, react to a moving object, or understand cause and effect in the physical world.

The company's approach to training this kind of AI is unusual: it uses hundreds of millions of hours of video game footage. That's not a coincidence — General Intuition's founder, Pim de Witte, previously built Medal, a platform where gamers share clips of their gameplay. That gave the company a massive head start on training data. Specifically, they use "action labels," which are records of exactly which buttons a player pressed and precisely when — essentially a detailed log of cause-and-effect decisions made by a real person in a simulated environment.

Why gameplay data, and why does it matter?

Investor Vinod Khosla, whose firm backed the company, argues these action labels are key to something AI researchers call "generalization" — the ability for a model to apply what it learned in one situation to a completely different one it's never seen before. The idea is that if an AI learns enough about how actions lead to outcomes in a huge variety of game scenarios, it can start applying that same instinct to new tasks, including ones in the real, physical world — like controlling a robot.

Where the money is going

The startup says the new funding will go toward three things: improving its core AI model with a specific focus on robotics ("embodiment," in industry jargon, meaning giving AI a physical body to control), buying more computing power (it has a partnership with cloud company CoreWeave for this), and hiring more people.

One detail that stands out: this deal reportedly marks Valor Equity Partners' first investment in an AI lab since it backed SpaceX. That's a signal of how seriously some investors are taking the bet that "AI that understands physical space" could be as big a deal as AI that understands language.