General Intuition Raises $300M for Embodied AI Training

General Intuition Closes $300M Raise at $2B Valuation

According to TechCrunch, General Intuition, a New York-based startup focused on spatial-temporal reasoning for AI agents, is in advanced discussions to raise approximately $300 million. The round would value the company just above $2 billion, sources told the publication. The financing round includes backing from Jeff Bezos, Eric Schmidt, along with returning investors Khosla Ventures and General Catalyst.

This funding announcement arrives eight months after the company separated from Medal, a video-clip platform serving the gaming community, when General Intuition completed a seed round worth $134 million.

The Medal Dataset as Competitive Moat

General Intuition’s differentiation rests on privileged access to Medal’s proprietary video archive: 2 billion clips annually generated by 10 million monthly active users. Unlike static video corpora used by competitors, this dataset captures interactive, first-person gameplay—a training signal that encodes human spatial reasoning and real-time decision-making within dynamic environments.

According to the TechCrunch report, this dataset’s uniqueness has caught the eye of larger players. OpenAI previously pursued acquiring Medal outright, and multiple other AI laboratories have expressed interest. The company argues that this interactive training signal enables machines to develop perception, anticipation, and real-time interaction capabilities in both simulation and physical robotics contexts.

A Different Path Than Generative World Models

The emerging world-model sector includes Runway, Decart, and World Labs, all of which recently released products designed for video synthesis and dynamic scene generation. Google’s Genie 3 recently integrated mapping data to enhance real-world simulation fidelity. These companies target gaming and robotics training as near-term applications.

General Intuition’s strategy diverges fundamentally. Rather than licensing world models as video generators—the commercial model rivals pursue—the company positions trained AI agents themselves as the primary product. The world model functions as training infrastructure, not revenue-generating output.

Scaling and Near-Term Roadmap

According to sources cited by TechCrunch, General Intuition will deploy the new capital to expand computational capacity, with a target product release planned for late summer or early autumn 2026. The team includes founder Pim de Witte, who co-founded Medal, alongside researchers Eloi Alonso, Adam Jelley, and Vincent Micheli—specialists in world modeling and simulation architecture.

Why This Matters

This funding round reflects a sectoral bet that embodied AI training—rather than visual generation—represents the durable value-capture opportunity in world models. For teams evaluating robotics simulation platforms or embodied-AI infrastructure, General Intuition’s dataset advantage and agent-first positioning offer a differentiated path compared to generative competitors. The backing from Bezos and Schmidt signals conviction in the thesis that interactive video datasets unlock qualitatively different reasoning than passive visual corpora. The late-summer product launch will be the first public test of whether this technical differentiation translates to commercially viable agent performance.