The race to build artificial intelligence systems that can understand and interact with the physical world is producing a growing number of candidates. But few companies are pursuing more unconventional strategies than General Intuition, a New York-based startup that believes the key to creating more capable AI agents lies not in warehouses of robots or fleets of self-driving cars, but in hundreds of millions of hours of video game footage.
This vision has attracted some of the biggest names in technology and venture capital. General Intuition announced a $320 million funding round that values the company at $2.3 billion, bringing its total publicly traded funding to $454 million after raising $134 million at its founding last October.
The round was led by Khosla Ventures, with participation from General Catalyst, Amazon founder Jeff Bezos, former Google CEO Eric Schmidt, former F1 champion Nico Rosberg, and researchers from Google DeepMind and MIT.
At the heart of the company is 31-year-old co-founder and CEO Pim de Witte, who argues that today’s AI systems remain fundamentally limited because they lack an intuitive understanding of how their actions affect the world around them.
That philosophy is on display at the company’s research facility, where AI agents reportedly played games like Fortnite for more than 100 hours continuously, according to TechCrunch.
“Our agents are playing 100 hours straight,” said Kent Rollins, chief product officer.
This demonstration was more than just a gaming experiment. The same basic model that controls the virtual character also guided the quadrupedal robot as it moved around the office, the company said.
“The same brain that powers the agents that play the game also powers the robots,” de Witte said.
Relying on a single camera as its primary sensor, the robot roamed around the office in exploration mode, occasionally bumping into chairs and trash cans to understand its surroundings.
According to data analyst Josh Duplantis, only eight minutes of real-world robotics data were needed to fine-tune the robot, and the training data was collected outdoors rather than in an office environment where the robot would move. The ability to transfer knowledge from gaming environments to simulations to physical machines forms the core of General Intuition’s long-term strategy.
Unlike many AI developers who focus on large-scale, text-based language models, the startup is building what researchers call “world models,” or AI systems designed to understand cause and effect, movement, space, and time.
The company’s roots help explain its approach.
General Intuition was born out of de Witte’s previous company, Medal, a platform where gamers could upload and share gameplay clips. Over the years, Medal has amassed hundreds of millions of hours of game footage and created a unique data asset.
But De Witte argues that the real value lies in the metadata, not the video itself. Most gameplay clips include a record of every button the player presses and the exact timing of those actions. That information allows AI systems to observe not just what happened, but why it happened.
“Most competitors are trying to infer behavior from videos alone,” De Witte argues, but General Intuition has access to records of real human decision-making embedded in the data.
“We see this as just the next step in future pre-training,” De Witte said. “We have a single model that can perform actions in response to on-screen Fortnite information, but it can also respond to real-world dynamics in a way that LLM never could.”
The company’s internal simulation platform serves as what executives call a “gym,” a training environment where AI models learn how to interact with a dynamic world.
Exposure to vast amounts of gameplay allows the model to naturally learn physical rules and spatial relationships, according to de Witte. The system demonstrated that it understands that walls are obstacles, ladders can be climbed, and shadows change as the sun moves across the scene.
However, a common intuition is that the simulation itself is not the final product. The company ultimately plans to commercialize the underlying agent AI model, which it believes can be generalized to gaming, robotics, autonomous systems, industrial automation, and other real-world applications.
“We’re not building a self-driving car company,” De Witte said. “We’re making it 10 times easier for the next person to start a self-driving car company.”
Currently, the company already has customers across the gaming, robotics, and simulation markets. The startup also sees opportunities in industrial automation, digital twins, robotic testing and working in hazardous environments.
De Witte says the technology can already operate any system that can be controlled through a familiar interface.
“Anything that can be controlled using a game controller or keyboard and mouse will work,” he said.
Much of the new funding will be used to expand computing infrastructure.
General Intuition has partnered with CoreWeave and plans to invest heavily in training larger versions of its models. A portion of the funding will also support broader deployment of the company’s API later this year.
For investors, the appeal goes beyond the technology itself.
Vinod Khosla, whose company led the round, sees the company’s unique dataset as a potentially decisive advantage.
“If you look at LLM, when inference came along, it was a quantum leap,” Khosla says. “I think the quantum leap in world models is the emergence of intuition in AI, an ability similar to human intuition. The human behavioral and reaction data available in games is an important part of the emergence of intuition.”
This unique data asset has reportedly attracted acquisition interest from major AI research institutes, but General Intuition has said it has rejected multiple offers.
Company executives say the goal is not to become an acquisition target, but to build a foundational AI infrastructure that can support an entire ecosystem of applications.
“Right now, it’s going to be data acquisition, which is kind of uninteresting,” Khosla said.
The company’s ambition comes as tech giants like OpenAI, Anthropic, Google DeepMind, and Meta pursue their own efforts to create more capable AI agents that can operate autonomously in both digital and physical environments.
But general intuition also seeks to differentiate itself through its ethical framework. De Witte, who previously spent several years working on humanitarian efforts such as Médecins Sans Frontières, said the company has no intention of pursuing lethal military applications.
“We don’t want to be an escalating part of the system,” he says.
“Suppose I come forward and say, ‘We have lethal self-government,’ what do you think will happen in other countries?”
He added that he still supports applications such as search and rescue missions.
The company’s culture also reflects its European roots. De Witte, who is Dutch, has hired staff whose views align with his approach to responsible AI development.
“I don’t understand why Silicon Valley would do something like this,” he says. “There’s a reason I’m not there.”
The startup is also thinking about the economic impact of AI. Recognizing concerns about job losses, General Intuition recently launched a platform called Nerve, a marketplace designed to help gamers earn income through data labeling tasks, remote control of robots, and other AI-related tasks.
De Witte believes the gaming community is one of the populations most exposed to future AI disruption and hopes they will benefit from the technology’s growth.
Looking ahead, General Intuition plans to leverage customer adoption to create a self-reinforcing data flywheel. The company plans to select partners not only for commercial reasons, but also for the unique real-world data that those deployments will generate.
“We will select customers who can diversify the implementations for which this generalized foundation model will serve as the backbone,” De Witte said.
“So we will prioritize choosing customers based on whether they can provide us with real-world data that is interesting and useful to advance our research, and whether they have agile in-house teams that allow us to be truly embedded partners and learn from each other.”
However, this strategy remains unproven. Industry researchers broadly agree that transferring capabilities learned in simulation to the physical world remains one of AI’s most difficult challenges.
Khosla acknowledges that it’s an open question whether simulation-to-reality learning will work at scale.
Still, with nearly $500 million in funding, access to one of the world’s largest repositories of human gameplay behavior, and growing interest in AI agents that can understand the physical world, General Intuition has emerged as one of the hottest startups in the next phase of artificial intelligence development.
The bet is that before machines can truly understand reality, they may first need to learn how humans play.
