General Intuition Targets Physical AI With Fresh Funding

General Intuition, a New York startup building foundation models for physical AI, is reportedly in talks to raise at a $6 billion pre-money valuation. The round would bring in Valor Equity Partners, Point72 Ventures, and Seven Seven Six, with existing backers Khosla Ventures and General Catalyst also participating, according to TechCrunch.

From gameplay data to robotic skills

General Intuition began with an unusual training asset: hundreds of millions of hours of video-game footage and the action labels attached to that footage, including which buttons players pressed and when. The company is using those examples to train a model that can learn how agents move through space and time, rather than only predicting text or images.

![Diagram showing how gameplay action data can transfer into physical AI skills]()

The startup plans to use the new capital to improve its general model and focus on robotic embodiments. That puts it in the growing physical AI race, where the central challenge is not just recognizing the world but choosing and executing useful actions inside it. TechCrunch reports that the company has a compute partnership with CoreWeave and expects to spend on infrastructure and hiring.

Why builders should care

For AI builders, the important signal is the company’s attempt to treat action as a first-class training target. Text models learn relationships between words. Physical agents need temporal understanding, spatial prediction, and feedback from the consequences of their decisions. Gameplay offers a large, structured environment for studying those loops before deploying systems into expensive or safety-critical machines.

That does not make a robot-ready model automatic. Game environments are simplified, labels can encode player-specific behavior, and the transfer from a screen to a real body introduces friction from sensors, latency, physics, and safety constraints. The funding story is also still developing, so the reported valuation is not a completed financing.

The broader direction is clear: agent research is moving beyond chat interfaces. Just as smaller AI scientists are challenging frontier models on specialized research tasks, physical AI startups are betting that carefully designed data and training loops can matter as much as raw model scale.

Builder takeaway

Teams experimenting with robotics or browser and GUI agents should watch the same underlying question: can a system learn reliable action policies from rich trajectories, then generalize when the environment changes? General Intuition’s approach suggests that the next competitive edge may come from better action data and evaluation harnesses, not only from larger language models.