General Intuition, a startup training AI agents to understand space and time through video game footage, is raising around $300 million（約480億円） at a $2 billion（約3200億円） valuation—signaling heavyweight investor confidence in a new category of AI.

What happened: General Intuition, which spun out of Medal eight months ago with a $134 million（約210億円） seed round, is in talks to raise around $300 million（約480億円） from backers including Jeff Bezos, Eric Schmidt, Khosla Ventures, and General Catalyst. The fresh funds would value the startup at just over $2 billion（約3200億円）.

Why it matters: The startup trains embodied AI using Medal's dataset of 2 billion videos per year from 10 million monthly active users—a unique resource of interactive, first-person gameplay footage that competitors cannot easily replicate. This dataset has reportedly attracted acquisition interest from OpenAI and other major AI labs, suggesting the video game data itself is a defensible moat in the race to build world models (AI that learns to simulate and reason about physical space and time).

What to watch: General Intuition will use the funds to scale compute capacity and release a new product by the end of summer or early fall. Unlike rivals (Runway, Decart, World Labs, Google's Genie 3) that sell world models directly, General Intuition builds them to train agents—the agents themselves are the product.