General Intuition trains AI agents on billions of gameplay videos

Two billion gameplay clips per year, uploaded by ten million active users: that is the raw material on which General Intuition trains its AI agents. The company, which spun out of Dutch video platform Medal in 2025, is building so-called foundation models designed to learn how the physical world works, purely by observing what people do in video games.

General Intuition was founded by four co-founders: CEO Pim de Witte, who helped build Medal, together with Adam Jelley, Brianna K. Martin and Anthony Hu. The company raised €114 million in a seed round, a figure that reflects its ambition: this is not a niche tool for gamers, but an attempt to develop fundamental AI capabilities with broad applicability.

The central question General Intuition sets out to answer is how to teach an AI system to reason about space and time, without overloading it with structured data or explicit rules. The answer the company is pursuing lies in the vast amount of visual information that gamers produce every day.

From clip platform to training infrastructure

Medal started as a platform where gamers could share short videos of their best moments, comparable to what Instagram is for photos. Under the leadership of Pim de Witte, it grew into one of the larger gaming video platforms, with a community of millions of users who daily upload clips from games such as Fortnite, Valorant and Minecraft.

While most AI labs have to purchase, scrape or label their training data, General Intuition has access through Medal to a continuous stream of new visual material. Each clip is a brief window into how a human navigates, makes decisions and responds within a simulated environment. This gives the company a training set that is both large and varied: the games differ considerably in perspective, pace and environment type.

That starting position is precisely why the spin-off is a logical step. The data was already there; what was missing was the research infrastructure to build foundation models from it. General Intuition is the organisation that fills that gap.

Spatial and temporal reasoning as the core objective

What General Intuition aims to develop falls under the category of world models: AI systems that build an internal model of how an environment works, enabling them to reason ahead about what will happen when they take a given action. This is fundamentally different from language models, which primarily recognise patterns in text.

General Intuition's agents process only visual input, similar to how a human experiences a game: via the screen, without access to the underlying game code or a numerical scoreboard. By combining reinforcement learning with that visual stream, the company seeks to train agents that independently learn to move, plan and respond in three-dimensional environments.

The most obvious application is gaming itself: AI-driven characters that behave like human players, or simulations that allow game developers to test new environments. But the broader claim is that spatial reasoning is a skill with value outside games, in robotics, autonomous systems and physical simulations.

€114 million seed financing

The €114 million seed round is large by almost any measure. In the Dutch startup context, such an amount at the earliest stage is rare; it immediately places General Intuition among the most heavily funded deeptech companies the country has produced.

The figure also reflects the nature of the work. Foundation model research requires computational capacity at a scale that smaller rounds cannot finance. Training models on two billion clips per year demands substantial GPU infrastructure, as well as a team of researchers capable of designing and overseeing the architecture and training process.

At the time of writing, no public details are available regarding the investors in this round.

Position in the European AI landscape

General Intuition is a notable presence in a European AI ecosystem that has struggled over recent years to get large foundation model initiatives off the ground. Most major models originate from the United States or China; European labs such as Mistral (France) and Aleph Alpha (Germany) are exceptions that prove the rule.

What sets General Intuition apart from many other European AI startups is the combination of a proprietary dataset at scale and a clear technical focus on a specific area, spatial reasoning, that the mainstream AI community regards as an open problem. This positions it as a research-driven company with its own data foundation, rather than an application built on top of other parties' existing models.

For Dutch investors and policymakers, the company serves as an example of how a consumer platform, in this case Medal, can lay the groundwork for fundamental AI research. Whether that foundation also leads to models that are usable outside the gaming context is a question that will be answered in the years ahead.