General Intuition Targets $300M Round at $2B Valuation

Eight months and $134 million ago, General Intuition barely existed as an independent company. Now the New York-based startup is reportedly in talks to raise approximately $300 million at a valuation of roughly $2 billion, according to TechCrunch, with backers including Jeff Bezos, Eric Schmidt, Khosla Ventures, and General Catalyst. The pace of that trajectory is itself the signal.

What makes this interesting is less the investor names and more the product thesis. Most world-model startups are building toward selling the models themselves, like Runway or World Labs. General Intuition is positioning differently: the foundation models it builds are infrastructure, and the agents trained on those models are the actual product. That distinction matters because it aims squarely at where enterprise buyers spend money, on deployable capabilities rather than raw model access.

The feedstock for those agents is Medal's dataset: reportedly 2 billion videos annually from 10 million monthly active users. That archive of spatial-temporal data is what the company uses to teach AI agents how objects behave and how environments evolve over time. It is also apparently what led OpenAI to try to acquire Medal outright before General Intuition spun out.

The honest caveat is that this is a pre-product company raising at a $2 billion valuation, and the product is not due until late summer or early fall. The reporting does not say what the product actually is, and it does not address whether the dataset advantage is defensible against competitors like Runway, Decart, and World Labs who might assemble comparable corpora from other sources. A valuation at this stage is a bet on the team and the thesis, not on demonstrated revenue.

Founder Pim de Witte and co-founders Eloi Alonso, Adam Jelley, and Vincent Micheli have a narrow window to validate that thesis in production. If they do, they will have made a credible case that the path to useful AI agents runs through rich behavioral video data rather than text.