General Intuition raises $2.3B betting video game tech can train AI for real-world applications

Walking onto General Intuition’s R&D floor in New York, co-founder and CEO Pim de Witte immediately pointed to a monitor on a standing desk. It looked like someone was playing Fortnite, but it wasn’t a person.

“Our agent has been playing for 100 hours straight,” said Kent Rollins, the company’s chief product officer, with a smile.

Before I could fully process the sight of an AI navigating a virtual game world, the electronic hum of a large quadrupedal robot approached.

“The same brain powering the Fortnite agent is driving this robot,” de Witte explained. Josh Duplantis, a data analyst streaming the robot’s live camera feed, added that the bot was in “exploration” mode.

Using its single camera as its only eye, the large, bug-like robot walked up to me, circled around, and continued into the office. It occasionally clipped chair legs or bumped into a trash bin, moving much like a toddler learning to navigate its surroundings. Duplantis noted it took just eight minutes of real-world robotics data to fine-tune the AI model for the quadruped. Remarkably, that data was collected on the street, not in the office where the bot was currently operating.

General Intuition’s core mission is an agentic model capable of generalizing from gameplay to simulation to physical embodiment. This ability to understand its place in the world has attracted significant investment.

On Thursday, General Intuition announced a $320 million raise at a $2.3 billion valuation, confirming prior TechCrunch reporting. This brings the startup’s total disclosed funding to $454 million, following a $134 million launch round last October.

The company spun out of de Witte’s previous venture, Medal, a platform for uploading and sharing video game clips. The hundreds of millions of hours of gameplay footage provided the initial dataset to train General Intuition’s model in spatial-temporal reasoning — understanding how to move through space and time.

However, the crucial ingredient wasn’t the video itself, but the action labels embedded within: records of exactly which buttons a player pressed and when. De Witte argues that most competitors, who try to infer actions from video alone, are missing a key piece.

“We view this as the next stage of pre-training,” de Witte said. “We have a single model that can respond to Fortnite visuals and take action, as well as handle real-world dynamics in a way an LLM never could.”

De Witte then had me test General Intuition’s world model on a laptop. This simulated environment is generated frame-by-frame rather than rendered by a traditional game engine. As I often do when testing such models, I walked straight into walls. In other demos, agents might pass through, but this one did not. From millions of hours of gameplay, it learned that walls are solid, ladders are for climbing, and shadows lengthen as the sun moves.

For General Intuition, this world model isn’t the final product; it’s the training environment (internally called “the gym”). The company aims to sell the agentic model itself. De Witte argues that action data from gameplay helps the model distinguish the “self” from the “environment,” providing a richer understanding of causality.

While General Intuition’s technology looks impressive in demos, it isn’t the only player in this space. Moreover, scaling such a model to work reliably in the physical world remains unproven. Most similar approaches require vast amounts of real-world data, which is slow and expensive to gather. General Intuition’s strategy is that gameplay offers a scalable shortcut.

Investors seem to agree. The latest round was led by Khosla Ventures, with participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers from Google DeepMind and MIT.

Most of the funding will go toward scaling compute capacity. General Intuition has partnered with CoreWeave and plans to focus on pre-training the next version of the model. A portion is also allocated to making its API more widely available by the end of summer.

Vinod Khosla, whose firm led the round, cited de Witte’s vision and the company’s proprietary data position as key factors.

“With LLMs, reasoning was a quantum leap,” Khosla said in an interview. “With world models, I believe the leap is the emergence of intuition — a human-like capability. The action and reaction data from games is key to developing that intuition.”

The vision is a generational company

General Intuition leverages data from Medal’s video game clips. Image Credits:Medal.TV

General Intuition isn’t the only company recognizing that Medal’s human action data is vital for building dynamic world models and general agents. Brianna Martin, the startup’s chief of staff, noted that the company was partly born after Medal rejected an acquisition offer from a major lab. Other offers have followed since.

De Witte and his co-founders — Eloi Alonso, Adam Jelley, and Vincent Micheli — have no interest in being acquired, nor do investors seek an early exit. The volume and quality of proprietary data General Intuition accesses via Medal are why Khosla views this as a generational bet rather than an M&A target. The company aims to become the backbone for generalized agents and world models in both simulation and reality.

“At this point, it would just be a data acquisition, which is uninteresting,” Khosla said.

This bet also involves trusting de Witte’s personal values.

Having spent seven years in humanitarian work, including with Doctors Without Borders, de Witte has drawn a clear ethical line: no agents will be used to harm humans.

“We don’t want to fuel an arms race,” de Witte said. “If I announced, ‘We’re building lethal autonomy,’ what do you think would happen in other countries?”

This restriction on military applications comes as Silicon Valley grows increasingly focused on defense. De Witte is comfortable with his models being used for search and rescue, but he feels Silicon Valley’s recent obsession with defense “infects the ecosystem.”

De Witte is Dutch, and much of his team is European, shaping the company’s culture. He brought on Martin partly because of her public resignation from Palantir over its work with U.S. Immigration and Customs Enforcement.

“I don’t understand why Silicon Valley operates the way it does,” he said. “There’s a reason I’m not there.”

De Witte’s ethics extend beyond limiting what the models won’t do. As a gamer who made $1.5 million building and hosting a private RuneScape server as a teenager, he is also concerned about the people left behind by AI advancements.

General Intuition recently launched Nerve, a jobs marketplace allowing gamers to earn money using their existing hardware. Sign-ups start with data labeling and can progress to robot teleoperation and other tasks. De Witte noted that Medal’s user base is precisely the generation most vulnerable to AI-driven displacement, and he wants them to have a stake in the future.

A data flywheel

De Witte envisions General Intuition as an ecosystem enabler, similar to Anthropic or OpenAI — a model provider that allows others to build upon its technology. Currently, the startup serves a handful of customers in gaming, simulation, and robotics.

“We’re not building a self-driving car company,” de Witte said. “We’re making it 10 times easier for the next person to build one.”

The company says that once its API reaches more customers, it can test various use cases — such as testing a robot in a digital twin of a factory floor, powering a human-like bot in a gaming studio, or sending a quadruped to navigate hazardous environments.

While a quadruped is the first physical embodiment General Intuition has tested in the real world, it has also experimented with drones and other devices, including testing the model in driving games.

“It works on anything you can control with a game controller or keyboard and mouse,” de Witte said.

Building a data flywheel is one of the primary goals.

“We’ll select customers who can diversify the embodiments this generalized foundation model serves as the backbone for,” de Witte said. “We prioritize customers who can provide real-world data that advances research, and who have agile internal teams for deep partnership and mutual learning.”

Khosla emphasized that General Intuition’s proprietary data is what has brought it this far, and its ability to continue collecting unique data will be essential. Despite impressive demos, whether simulation-to-real-world transfer can scale remains an open question that no one has fully answered yet.