General Intuition raises $320M Series A to train AI on gameplay video from Medal.tv

General Intuition has raised a $320 million Series A at a $2.3 billion valuation to build foundation models that learn to reason about space and time from gameplay video. Khosla Ventures led the round, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg and researchers at Google DeepMind and MIT. The deal brings the company’s total funding to $454 million, after a $134 million seed in October 2025.

The bet is that video games are a cheap, abundant source of the kind of action-labeled data that physical-world robots and agents are starved for. General Intuition trains its models on data from Medal.tv, which it owns and operates and describes as the world’s largest gameplay clip platform, taking in billions of clips a year. The company says the resulting models are built to “perceive, predict, and act in virtual and physical environments.”

Running both an AI lab and the platform that feeds it gives General Intuition a proprietary data pipeline that rivals would have to license or scrape. The vast majority of the new money is earmarked for compute scaling, building on an existing deal with CoreWeave, the company said. A portion is set aside to open its commercial API broadly by the end of summer 2026.

The approach is unproven at the scale the valuation implies. General Intuition has not published benchmark results or named external customers for the models, and “spatial-temporal reasoning” trained on gameplay has yet to demonstrate transfer to real-world robotics in production. The company’s own homepage is the only first-party record of the round; it has not posted a separate press release, and its chief executive’s name does not appear on the homepage.

Whether gameplay data can teach machines to act in the physical world is the open question the round is funding. The commercial API launch this summer will be the first public test of whether anyone outside the lab finds the models useful.