General Intuition Secures $320M to Advance AI Training Through Video Game Data

General Intuition is using gameplay action data to train agentic models that generalize from virtual environments to physical robotics, recently raising $320 million to scale its compute infrastructure.

Startups, Policy & World Desk
Updated Nov 8, 2026 3 min read
General Intuition Secures $320M to Advance AI Training Through Video Game Data

General Intuition is attempting to bridge the gap between virtual gameplay and physical robotics. The startup, led by CEO Pim de Witte, is developing an agentic model trained on vast amounts of video game data, specifically utilizing the action labels—the precise button presses—recorded from millions of hours of gameplay via the platform Medal. According to TechCrunch, this approach aims to teach AI models spatial-temporal reasoning, or how to navigate through time and space, in a way that allows the same brain to control both a virtual agent in a game and a physical quadrupedal robot.

The company’s world model serves as an internal training environment, or the gym, rather than the final product. Unlike traditional AI training that relies on video alone, General Intuition argues that incorporating action data helps the model distinguish between the self and the environment, fostering a deeper understanding of causality. In demonstrations, the model has exhibited an ability to learn physical constraints, such as the solidity of walls or the function of ladders, without being explicitly programmed. The company claims this method offers a scalable alternative to the expensive, slow data collection processes typically required for robotics.

Evidence of this capability was showcased in the company’s New York office, where an agent navigated a virtual environment for 100 hours straight. Simultaneously, a quadrupedal robot powered by the same model demonstrated basic navigation, clipping chair legs and bumping into obstacles—behavior reminiscent of a toddler learning spatial awareness. According to data analyst Josh Duplantis, the robot required only eight minutes of real-world data to fine-tune its model, a process that occurred outside the office environment. This suggests a potential path toward generalization that avoids the need for massive, specialized real-world datasets.

To support this development, General Intuition recently closed a $320 million funding round, bringing its total disclosed capital to $454 million, as reported by TechCrunch. The round, led by Khosla Ventures with participation from investors including Jeff Bezos and Eric Schmidt, values the company at $2.3 billion. The majority of these funds are earmarked for scaling compute capacity through a deal with CoreWeave, which will support the pre-training of the next model version. Additionally, the company plans to broaden access to its API by the end of summer, marking a shift toward commercial availability.

Despite interest from potential acquirers, including a major lab that approached the startup after it spun out of Medal, the founders remain focused on independent growth. Vinod Khosla, whose firm led the latest round, characterized the startup as a generational bet rather than an acquisition target, citing the proprietary nature of its gameplay data as a key competitive advantage. The company is positioning itself as a provider of generalized agents for both simulation and physical applications, though it faces the challenge of proving these models can function reliably at scale in the real world.

The company has also established strict ethical boundaries regarding its technology. CEO Pim de Witte, who previously worked with Doctors Without Borders, has stated that General Intuition will not develop agents for lethal autonomy or military use, preferring to focus on applications like search and rescue. This stance contrasts with broader trends in Silicon Valley, where interest in defense-related AI is increasing. De Witte’s approach is influenced by his background and the European composition of his team, which he suggests informs the company’s distinct operational culture.

Sources

  1. 01 General Intuition’s $2.3B bet that video games can train AI agents for the real world — TechCrunch — Startups