A Medal spinout turned gamer button-presses into a $6 billion robotics bet

General Intuition is in talks to raise at a $6 billion pre-money valuation, and the asset underneath is not a product. It is hundreds of millions of hours of recorded player button-presses, spun out of a video game clip platform and now aimed at training general-purpose robots, per TechCrunch reporting.

The round, described as oversubscribed and still finalizing, comes weeks after a $320 million raise at $2.3 billion earlier this summer. A step-up of that size, on a company that has not yet shipped a product, is the round's most market-moving fact.

New investors are Valor Equity Partners, Point72 Ventures, and Seven Seven Six. Existing backers Khosla Ventures and General Catalyst are also participating. Valor is best known for its SpaceX position; if the deal closes, General Intuition would be the first AI lab Valor has backed since SpaceX. Seven Seven Six, the fund Alexis Ohanian founded, adds a consumer/creator-economy investor to a category that has, until now, been funded mostly by deep-tech specialists.

The company's origin is the part the headline number will bury. CEO Pim de Witte spun out General Intuition in October 2025 from Medal, a video game clip-sharing platform. Medal's servers had been logging "action labels," meaning the records of which buttons a player pressed and when, alongside hundreds of millions of hours of gameplay video. The original bet, described at the time of the $134 million seed, was that those action labels could teach AI agents spatial reasoning, because a player's choices encode how a level is actually navigable. Where an unlabeled video shows what a level looks like, an action-labeled clip shows how a human actually moved through it.

In July, the company told TechCrunch it believed robotics was about to have its "ChatGPT moment", a phrase the company itself is using rather than an industry consensus. The new round is intended to fund that expansion: a general model aimed at robotic embodiments, more compute under the company's CoreWeave partnership, and additional hires.

Vinod Khosla, the Khosla Ventures founder, framed the bet in a separate interview as the "emergence of intuition," or the idea that action labels could let a model generalize across tasks it was not explicitly trained on. That framing is one investor's view, not a measured technical result. No independent benchmark has yet shown that button-press-trained policies transfer to physical robots at scale, and "emergence of intuition" is not yet a category term.

Counterforce lives in two places. First, the $6 billion figure is single-sourced: TechCrunch, citing people familiar with the matter, is the only outlet that has reported it. The round is still "in talks" and "still finalizing." Final size, lead investor, and pre-money can all move before close. Second, the asset being capitalized is a corpus, not a deployed product. A bigger training set does not automatically produce a better policy, and a $6 billion pre-money raises the bar for what a v1 product has to prove.

The capital path is the part investors will track. $134 million seed in October 2025, roughly $320 million at $2.3 billion earlier this year, and now in-flight at $6 billion pre-money. On paper, that is roughly a 45x step-up on disclosed capital in ten months, on a pre-product company, into a category whose commercial timeline is still unproven.

The next checkpoint is a close: a definitive term sheet, a second outlet confirming the number, or both. The round's headline number lives in a single outlet's reporting for now.