General Intuition is building around a simple but ambitious idea: the actions players take in video games may help train AI agents for the physical world. The company is using gameplay data, simulation and robotics work to develop a general agentic model that can understand movement through space and time.
The startup has now raised $320 million at a $2.3B valuation. That puts fresh weight behind its claim that video games can become a training ground for AI systems that need to act, adapt and eventually operate beyond screens.
A model trained on action, not just video
General Intuition was spun out of Medal, Pim de Witte's other company, which lets gamers upload and share video game clips. Those uploads provided hundreds of millions of hours of gameplay that became the startup's initial dataset.
The company is not relying only on what appears on screen. Its key input is the action labels embedded in gameplay clips: records of exactly what buttons a player pressed and when. De Witte argues that this matters because many competitors are trying to infer actions from video alone.
That distinction shapes the company's view of AI training. A gameplay clip can show a character moving, but the action labels connect that movement to a human decision. For General Intuition, that link between observation and control is what makes the data useful for spatial-temporal reasoning.
De Witte described the approach this way:
"We view this as just the next stage of future pre-training," de Witte said. "We have a single model that can respond to Fortnite information on the screen and take action, but also to real-world dynamics in a way that an LLM could never."
From Fortnite-like gameplay to a quadrupedal robot
At General Intuition's New York office, the same underlying system was shown powering an AI agent playing something like Fortnite and a large quadrupedal robot moving through the office. Kent Rollins, the company's chief product officer, said the agent had been playing for 100 hours straight.
The robot relied on a single camera and had a default mode described as "exploration." It walked around the office, circled a visitor and sometimes clipped chair legs or bumped into a trash bin. The behavior was not presented as fully polished autonomy, but as evidence that the same model could be adapted across different forms of control.
Josh Duplantis, a data analyst, said it took just eight minutes of real-world robotics data to fine-tune an AI model for the quadruped. That data was collected on the street, not in the office where the robot was navigating.
This is the core promise General Intuition wants to prove: a model that learns from gameplay can transfer to simulation and then to embodiment. The company has also tried drones and other devices, including testing the model in driving games. De Witte said, "It works on anything that you can control using a game controller or a keyboard mouse."
The world model is a gym
General Intuition has also built a world model, a simulated environment generated frame by frame rather than rendered by a traditional game engine. In a demo, it behaved like an environment that had learned basic expectations from gameplay, such as walls blocking movement, ladders being used for climbing and shadows changing as the sun moves.
For the company, that world model is not the product. Internally, it is treated as a training environment, or "the gym." The product General Intuition ultimately wants to sell is the agentic model itself.
The reasoning is that an agent needs to understand more than images. It needs a sense of what it controls, what it does not control and how actions change an environment. De Witte argues that action data from games helps the model distinguish the "self" from the "environment," giving it a stronger grasp of causality.
That is still an open challenge. The source article notes that getting a model like this to work at physical-world scale has not yet been done. Many approaches require large amounts of real-world data that are slow and expensive to gather. General Intuition's bet is that gameplay provides a scalable shortcut.
Investors are backing the data advantage
The latest funding round was led by Khosla Ventures, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers at Google DeepMind and MIT. The round brings General Intuition's total disclosed funding to $454 million, following a $134 million round it raised at launch last October.
Most of the new round will go toward scaling compute capacity. General Intuition has a deal with CoreWeave and plans to focus on pre-training the next version of the model. A smaller portion is set aside for making its API more broadly available by the end of summer.
Vinod Khosla said he was drawn to de Witte's vision and the company's proprietary data position. He framed the opportunity around the emergence of intuition in world models:
"If you look at LLMs, when reasoning emerged, it was a quantum leap," Khosla told me in a phone interview. "In world models, I think the quantum leap is the emergence of intuition in the AI, a human intuition-like capability. The human action data and reaction data you have in games is the key part to the emergence of intuition."
Khosla also said that, at this stage, an acquisition would be less compelling because it would mainly be a data acquisition. General Intuition has already received acquisition interest, including an offer from a major lab that Medal turned down, according to Brianna Martin, the startup's chief of staff.
An AI platform with boundaries
General Intuition wants to be a model provider that others build on top of, similar in role to Anthropic or OpenAI. Today, it has a handful of customers in gaming, simulation and robotics.
De Witte does not describe the company as a builder of every downstream application. He said, "We're not gonna build a self-driving car company. We're gonna make it 10 times easier for the next person to build a self-driving car company."
The company expects customer deployments to help create a data flywheel. It plans to prioritize customers that can provide useful real-world data and have internal teams capable of working closely with General Intuition. Potential uses include testing a robot in a digital twin of a factory floor, powering a humanlike bot inside a gaming studio or sending a quadruped into hazardous environments.
There are also stated limits. De Witte, who spent three years working in the humanitarian space, including with Doctors Without Borders, said no agents will be employed to harm humans. He said he is open to search and rescue missions, but drew a line around lethal autonomy.
General Intuition has also launched Nerve, a jobs marketplace where gamers can earn money using their existing setups. People who sign up begin with data labeling and can eventually move toward robot teleoperation and other tasks. De Witte said Medal's user base is the generation most exposed to AI-driven displacement, and he wants them to have a stake in what comes next.