General Intuition raised $320 million to train robots using video games
English translation from Unibot. Source published 2026-06-30.
General Intuition US Inc. raised $320 million in a Series A round. The company will use the funds to create AI models that can perceive, predict and act in virtual and physical environments.
Game clips instead of real data
Physical AI is now a major topic in robotics. But General Intuition says it has chosen its path. Instead of collecting hundreds or thousands of hours of real-world data or generating simulated data, the company uses billions of game clips uploaded to the Medal platform.
It allows users to share moments from games. By the way, the founder and CEO of General Intuition, Pim de Witt, is the same person who created Medal.
In these game recordings, people perceive the environment and decide how to move through it. This is what makes them valuable, the company explains. Text models provide only descriptions of reality - this is not enough to train physical AI.
Additionally, videos contain built-in action tags. They record which button the player pressed and when. This gives General Intuition more information about how people make decisions.
Company valuation and investors
The New York-based company said its Series A round valued it at $2.3 billion, bringing its total funds raised to $454 million, following $134 million raised in October.
The round was led by General Catalyst. Amazon founder Jeff Bezos and former Google CEO Eric Schmidt also participated.
From words to peace
Many powerful fundamental models are trained on written texts. However, General Intuition states: human intelligence is much broader than language. It has been shaped over millennia through interaction and exploration - through an endless cycle of intentions, actions and consequences in a variety of environments.
Truly smart machines must move from words to the world. They need to learn to perceive, anticipate and improvise. They need to gain a general intuition of reality, say General Intuition.
The company explains that its models learn from unique sets of action-tagged videos collected from countless environments. This diversity, she says, leads to uniquely capable agent systems.
Future plans
Since its founding in 2015, General Intuition has worked to create action models (they decide what to do) and world models (they predict the consequences of actions).
The company plans to use the funding to scale its computing power and focus on pre-training the next version of its model. According to TechCrunch, General Intuition also hopes to make its API more accessible this summer.