In one of the splashier AI funding rounds of the year, investors are betting that the path to robots that can navigate the real world runs through video games.
The raise
General Intuition has closed a $320 million Series B at a $2.3 billion post-money valuation, TechCrunch reported. (The $2.3 billion figure is the valuation; the cash raised is $320 million.) Khosla Ventures and General Catalyst led, as they did in the company's $133.7 million seed round last October, with new backers including Jeff Bezos, former Google chief Eric Schmidt and the former Formula One driver Nico Rosberg. Total disclosed funding now stands at roughly $450 million.
What it does
General Intuition builds AI models for what the industry calls embodied AI — software "agents" that must understand and move through three-dimensional space, whether in a game or in a robot. Its bet is that video games are an unusually rich source of the training data such agents need.
The company spun out of Medal, a platform where gamers upload clips of their play. Crucially, Medal captures not just the video but the player's inputs — button presses, joystick moves — synced to each frame, giving the AI "action labels" baked into the footage rather than guessed at afterward. Medal's library spans tens of thousands of games and draws billions of clip uploads a year. The company says models trained on that data have driven an agent playing Fortnite for more than 100 hours, and, with the same underlying model fine-tuned on a few minutes of real footage, helped a four-legged robot cross unfamiliar terrain.
The plan
General Intuition does not intend to sell robots or games. Like an OpenAI or Anthropic for physical-world AI, it plans to be a model provider, opening access to its models to outside developers and scaling computing power through a partnership with CoreWeave. Chief executive Pim de Witte told TechCrunch the company views game data as "just the next stage of future pre-training." He has also said the firm will avoid lethal-autonomy military uses while allowing applications such as search and rescue.
The bigger bet
The round lands amid a surge of investor money into "world models" and embodied AI — the idea that machines need a learned sense of physical space, not just language, to be useful in the real world. The wager that game-derived data can substitute for the slow, costly work of gathering real robotics data remains contested among researchers; plenty of approaches compete, and it is far from settled that gameplay generalizes cleanly to physical environments. What gives General Intuition a specific claim is the scale of Medal's footage and the action labels embedded in it — an advantage rivals relying on generic video cannot easily copy. Whether that translates into models good enough to be worth $2.3 billion is the question the company's new backers are paying to answer.
