General Intuition has drawn major attention with a fresh funding round that values the startup at $2.3 billion. The company raised $320 million to advance an AI approach built on a bold idea: gameplay data can help train agents that operate in the physical world.
At the center of the system is a model that learns from video game clips, especially the action signals hidden inside them. Those records show not just what appears on screen, but how players respond moment by moment. According to the company, that extra layer gives the model a stronger sense of spatial-temporal reasoning and cause-and-effect.
General Intuition says its technology can move from gaming environments to simulation and then into robotics. In demonstrations, the same model that navigates a game like Fortnite also powers a quadruped robot using a single camera. The startup says only minutes of real-world robotics data can help refine the system for physical tasks.
The company emerged from Medal, the clip-sharing platform that supplied a large archive of gameplay and action data. That dataset became the foundation for General Intuition's training pipeline, which the team describes as a kind of internal "gym" for building general-purpose agents.
The new round was led by Khosla Ventures, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers connected to Google DeepMind and MIT. Most of the capital will go toward compute expansion, while part of it will support broader API access later this year.
Beyond the technology, General Intuition is also shaping a wider ecosystem. It recently introduced Nerve, a marketplace designed to connect gamers with paid digital work such as data labeling and robot teleoperation. The company says it wants to create pathways for people already familiar with game-based interfaces.
With its mix of proprietary data, robotics ambition, and platform strategy, General Intuition is positioning itself as a foundation layer for the next generation of intelligent systems. If the model scales successfully, it could help define how AI learns to act in both virtual and real environments.
