General Intuition raises $320M to scale game-trained AI agents

General Intuition sits at the intersection of three big AGI ideas: world models, agentic behavior, and data flywheels. Training on labeled game play rather than just pixels gives their model richer signal about actions and causality, and the same policy can already control both Fortnite agents and real robots. That’s exactly the kind of generalization people expect from proto‑AGI systems. A $320 million check at a $2.3 billion valuation signals that top-tier investors now see this line of work as a primary frontier, not a side bet.

The focus on proprietary, action-labeled data is also notable. In a world where text and image data are increasingly commoditized and litigated, tying your moat to behavior logs from hundreds of millions of game sessions plus robot teleoperation may be a more durable edge. The hardware side matters too: most of this money is earmarked for compute, effectively turning capital into training runs for ever-larger world models.

If General Intuition can show reliable sim-to-real transfer at scale, it could accelerate progress toward embodied, general-purpose agents that can operate in messy physical environments. That’s a very different path to “general intelligence” than scaling up text-only LLMs, and it raises the stakes for incumbents who’ve mostly treated robotics as an afterthought.