Analysis
General Intuition has raised at a $6 billion valuation in a round backed by Valor Equity Partners and Point72, TechCrunch reported, as the company extends its spatial-reasoning models from video games into physical robotics.
General Intuition emerged in 2025 out of Medal, the gameplay clip-sharing platform, with an unusual data advantage: an archive of billions of short gameplay videos showing humans navigating three-dimensional environments and reacting to them. The company's thesis is that this corpus teaches something text cannot -- how objects move, what happens next, and how an agent should act in a space it has never seen. It launched with a $134 million seed round co-led by Khosla Ventures and General Catalyst, which was itself one of the largest seed rounds of that year.
The move into robotics is the obvious extension and also the hard part. Agents that play games well operate in worlds with forgiving physics, instant resets and no consequences for failure. Physical robots have none of those properties. The bet is that a world model trained on enough visual sequence data develops general enough intuitions about motion and causality that fine-tuning on real robot data becomes cheap. That is the same wager Physical Intelligence, Skild AI and Google DeepMind's robotics group are making, each from a different data starting point.
“It launched with a $134 million seed round co-led by Khosla Ventures and General Catalyst, which was itself one of the largest seed rounds of that year.”
- Physical Intelligence raised at a multibillion valuation on real-robot teleoperation data
- Skild AI takes a simulation-heavy approach to the same general-purpose control problem
- Nvidia supplies both the training compute and, through its robotics platforms, a competing stack
Valor and Point72 are notable as leads. Valor has a long history in operationally intensive, capital-hungry companies, and Point72's venture arm has been increasingly willing to price early AI rounds at growth-stage numbers. A $6 billion valuation roughly a year after a seed round means every subsequent round has to clear a bar that assumes the robotics transfer works, not merely that it might.
The counterweight worth stating plainly: no one has yet demonstrated that game-video pretraining produces robot policies that outperform models trained on real manipulation data. The transfer hypothesis is well-motivated and unproven, and the gap between a strong benchmark demo and a robot that works reliably in an unstructured warehouse has consumed a decade of capital across this field. At $6 billion pre-product-revenue, the valuation is underwriting a research result that has not been published.
Update (August 26, 2026): Pulse has follow-up coverage — Robotics Startup Generalist Hits $3B Valuation.