- General Intuition has raised $320M at a $2.3B valuation, just three months after a $133.7M seed round, with Khosla Ventures leading both financings.
- The New York lab trains AI on billions of action-labelled gameplay clips from Medal’s 17 million monthly active users, betting that human decision-making in games is the richest available dataset for teaching AI to act in the real world.
- The company is already in conversations for a Series B and has rejected multiple acquisition approaches, with the majority of new funding going toward compute via a deal with CoreWeave.
A New York AI lab that trains its models on billions of gameplay clips from real human players has raised $320M at a $2.3B valuation, and is already in conversations for another round.
General Intuition closed its Series A in January 2026, just three months after a $133.7M seed round that completed around October 2025. Khosla Ventures led both rounds, taking six to eight weeks to close the latest financing. General Catalyst returned alongside new backers, including Jeff Bezos through Bezos Expeditions, former Google chief executive Eric Schmidt, and Formula One world champion Nico Rosberg.
Based on research progress since January, the company says it is now in early conversations for a Series B, making it one of the few labs building foundational AI models to pursue three consecutive rounds inside 18 months.
Pim de Witte, the 31-year-old Dutch co-founder and chief executive, said the investor roster carries value beyond capital. “A lot of doors open when you get to work with these people. I think building a company is in many ways about building optionality on possible futures. And so when you get to work with people that can open doors, it generally is very good for business,” de Witte adds.
The pace of fundraising reflects something investors don’t often say out loud: they backed the research trajectory, not a commercial product. Vinod Khosla, whose firm led both rounds, draws a parallel to the moment large language models developed the ability to reason.
“If you look at LLMs, when reasoning emerged, it was a quantum leap. 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 notes.
The market the company is entering is large and moving fast. Global robotics startups raised over $23 billion in 2026 alone, nearly matching all of 2025, as investors bet that the next wave of AI value will come from models that can act in the physical world rather than simply reason about it.
Most leading AI labs have focused on models that generate text, images, or code. What none has cracked at scale is AI that can reliably act: perceive an environment, predict what will happen next, and choose the best move in real time. That capability lies at the heart of robotics, autonomous systems, and any agent that needs to act in the world rather than just describe it.
From Medal to machine intuition
General Intuition was founded in 2025 by de Witte, Eloi Alonso, Adam Jelley, and Vincent Micheli with backgrounds spanning AI research, gaming infrastructure, and autonomous systems.
The founders also built Medal, a gaming clip platform with more than 17 million monthly active users who upload around 2 billion video clips per year, which is part of the same parent company as General Intuition. Every clip carries something most training datasets lack: precise action metadata, including the mouse movements, button presses, and split-second decisions the player made in the moment. It was that detail that convinced the founding team they were sitting on an asset more valuable than the platform itself.
The startup works by training what it calls large action models on that dataset. Rather than learning from text or static images, its models learn from the moment-to-moment decisions players make: where to move, what to target, when to retreat. The theory is that the cognitive patterns underlying those decisions generalise beyond games into robotics and physical AI.
General Intuition builds two types of models in parallel: world models, which predict how an environment will evolve given a particular action, and action models, which work in reverse, generating the best action given what the model can observe.
At its New York office, de Witte recently demonstrated an AI agent that had been playing a Fortnite-like game for 100 hours straight, while the same underlying model simultaneously powered a quadrupedal robot that navigated the room using a single camera and 8 minutes of real-world data collected on a street outside.
A crowded field and a different data bet
The competitive landscape is well-funded and fast-moving. Physical Intelligence, the San Francisco robotics AI lab, raised $600M at a $5.6B valuation in November 2025, building a general-purpose robot brain trained on real-world manipulation data. Skild AI raised close to $1.4B led by SoftBank at a $14B valuation in January 2026, training its foundation model on human video and physics simulation. G
oogle DeepMind’s Genie 3 world model, publicly released in January 2026, is already being used by Waymo to simulate autonomous driving edge cases. Closer to General Intuition’s approach, Odyssey raised $310M for general world models and Rhoda raised $450M for robot intelligence trained on video, both signalling how quickly capital is concentrating around companies that can simulate or predict physical environments at scale.
What separates General Intuition, the company argues, is the specificity of its training data. Most competitors rely on simulation datasets, hand-collected robotics demonstrations, or human video scraped from the internet — footage that captures what happened, but not what caused it.
Medal’s clips are paired with the precise inputs that generated each on-screen moment, giving the model a direct record of human intention and its consequences. To extend that advantage, it launched Nerve alongside the funding announcement: a platform that pays users for gameplay footage, tele-operated robotics demonstrations, and other real-world action data. It is, in effect, paying people to teach machines how to move. A broader rollout of the company’s commercial API is planned for the end of summer 2026, with the vast majority of new funding going toward scaling compute capacity through a deal with CoreWeave.
Khosla has made physical AI a central theme of its recent portfolio. It backed FieldAI’s $405M raise for general-purpose robot intelligence and co-led Genesis AI’s $105M seed for a robotics foundation model.
General Intuition is a third, distinct bet — this time on action models trained from human gameplay rather than simulation or manipulation data. Khosla has said that General Intuition’s proprietary data position makes it a generational bet rather than an acquisition target. Multiple acquisition approaches from major AI laboratories have reportedly been rejected.
General Intuition operates as a public-benefit corporation, a legal structure that requires the company to consider broader social impact alongside profit, and is legally registered in the Netherlands. De Witte, who spent three years working in the humanitarian sector, including with Doctors Without Borders, has drawn a clear ethical line: the company will not pursue lethal military applications.
“We don’t want to be an escalatory part of the system,” de Witte notes. It has also committed not to develop technology that replaces game developers, designers, or artists.
The funding will go toward research, model development, computing, and hiring AI researchers, infrastructure engineers, and specialists in large-scale video processing and model training. The company has offices in New York, Geneva, London, and Paris. Total raised stands at $454M across the seed and Series A.
