General Intuition Secures $320M to Train AI on Action Labels

General Intuition has closed a $320 million Series A funding round at a $2.3 billion valuation to train embodied AI agents on video game telemetry. Detailed in the funding announcement, the New York-based lab intends to solve the robotics data bottleneck by utilizing interactive gameplay environments rather than static text or passive video. The company’s upcoming API shifts the physical AI training paradigm from real-world collection to virtual simulation.

Causality Through Action Labels

The technical differentiation of General Intuition centers on action labels. Traditional video models infer intent by observing pixels change over time. General Intuition bypasses visual guesswork by pairing video frames with exact, timestamped hardware inputs. This metadata teaches the model direct causality, demonstrating exactly how a discrete button press alters a spatial environment.

Feature	Passive Video Models	Action-Labeled Telemetry
Intent Inference	Guessed from pixel changes	Explicitly recorded via hardware inputs
Causality Tracking	Low	High
Scale	Unlimited	2 billion clips per year
Hardware Mapping	None	Exact timing of player actions

The lab sources this telemetry through a sister-company relationship with the gaming platform Medal, ingesting roughly 2 billion video clips annually from over 10 million active users. CEO Pim de Witte demonstrated the efficiency of this approach by training an agent for 100 hours in a Fortnite-style world simulation. The resulting model successfully navigated a physical office environment using a quadrupedal robot after just eight minutes of real-world fine-tuning.

Infrastructure and Capital

The leadership team includes researchers Eloi Alonso and Vincent Micheli, the architects behind the DIAMOND diffusion-based world model. The company will direct the majority of the $320 million round toward scaling compute resources. The round was led by Khosla Ventures with participation from General Catalyst, Jeff Bezos, and Eric Schmidt.

A dedicated partnership with CoreWeave handles the high-throughput video pipelines and long-lived multi-node training jobs required to process the massive temporal datasets. The current $2.3 billion valuation follows de Witte’s decision to reject a $500 million acquisition offer for Medal from OpenAI in late 2024, opting instead to spin out the research division into an independent entity.

Market Positioning and Timeline

The lab differentiates itself from pure-generation firms like Decart and World Labs by focusing strictly on “vision in, actions out” systems. General Intuition targets environments where agents must execute multi-step logic and physical interaction, stepping beyond visual synthesis. The company plans to release its first commercial API for game developers and robotics engineers by late summer or early fall 2026.

If your team builds hardware automation or simulation software, the General Intuition API introduces a direct mechanism to transfer virtual telemetry into physical execution. You should map out which segments of your robotic control logic could be pre-trained in synthetic game environments before the public endpoint becomes available.