General Intuition Talks $6B Valuation, Up From $2.3B Two Months Ago

General Intuition, a New York AI startup that trains robots and software agents using gameplay footage, is in talks to raise new funding at a $6 billion pre-money valuation, according to TechCrunch, which cited sources familiar with the deal.

That's a 2.6x jump from the $2.3 billion post-money valuation the company set during its Series A round roughly two months ago, according to Crypto Briefing. The Series A closed at $320 million in June 2026. Before that, General Intuition raised a $134 million seed round in October 2025. Total disclosed funding sits at $454 million heading into this new round.

New investors reportedly in talks include Valor Equity Partners, Point72 Ventures, and Alexis Ohanian's Seven Seven Six. Existing backers Khosla Ventures and General Catalyst are also participating, according to TechCrunch. The Series A included Jeff Bezos through Bezos Expeditions and former Google CEO Eric Schmidt via his family office Hillspire, according to Crypto Briefing.

TechCrunch reported the round is still being finalized and is oversubscribed, meaning the company is turning away or capping interest from investors who want in. TechCrunch said it had reached out to Valor to confirm the investment, which would reportedly be Valor's first AI lab bet since SpaceX. That confirmation had not landed as of this writing.

What the company actually does

General Intuition was spun out last October by CEO Pim de Witte from Medal, his video game clip-sharing platform. Medal has roughly 17 million monthly active users who generate billions of gameplay clips a year, according to Crypto Briefing, each one tagged with the specific button presses and actions a player took.

The company uses those tagged clips, what it calls "action labels," as training data for what it terms large action foundation models. The pitch: video games are cheap, scalable physics sandboxes with millions of people generating labeled data around the clock, and that data can teach an AI agent to reason about space and time the same way it would need to for robotics.

Investor Vinod Khosla told TechCrunch he believes action labels are key to building genuine "intuition" in AI models, meaning the ability to generalize to tasks the model was never explicitly trained on. Crypto Briefing reported one General Intuition model demonstrated continuous gameplay for over 100 hours and adapted to a real-world physical navigation task after just 8 minutes of real-world data, though neither outlet cited an independent benchmark or third-party replication of those figures.

The company has a compute partnership with CoreWeave and plans to spend the new funds on more infrastructure, more hiring, and a push into robotic embodiments, according to TechCrunch. Crypto Briefing reported the company also plans to open broader API access later in 2026.

The skeptical read

A 2.6x valuation jump in about two months, with no announced product launch, revenue disclosure, or customer contracts driving it, is the kind of number that should raise an eyebrow. Nothing in either source establishes what changed operationally between June and August beyond investor demand itself.

Physical AI and robotics have become the hottest lane in Silicon Valley funding, and a startup with a genuinely unusual data source—hundreds of millions of hours of labeled human gameplay—is a legitimately different bet than another chatbot wrapper. Valor's history backing SpaceX and Point72's quant-driven diligence process suggest more than pure momentum chasing. Still, a valuation set purely by investor competition to get into an oversubscribed round is a valuation set by scarcity, not by demonstrated commercial traction.

One thing worth keeping in mind: the PressBee item republishing this story added no new reporting and simply confirmed the TechCrunch account was picked up by other outlets. That's republication, not independent verification. Treat the multiple write-ups as one underlying report, not three corroborating ones.

The question remains whether General Intuition's action-label approach produces robots and agents that generalize in the real world the way Khosla predicts, or whether gameplay data turns out to be a poor proxy for physical embodiment. The company's promised late-2026 API opening is the first real test of that claim outside of investor pitch decks.