A startup called General Intuition just raised $320 million at a $2.3 billion valuation by betting that the hundreds of millions of hours of video game footage uploaded to the internet hold the key to training AI agents that can navigate and manipulate the physical world. The company's approach sidesteps one of the biggest bottlenecks in robotics: the need to collect enormous amounts of expensive, real-world training data. Instead, it extracts action labels from gaming clips, teaching AI models not just what the world looks like, but how to move through it.

Why Are Tech Companies Abandoning Chatbots for World Models?

After years of pouring resources into large language models (LLMs), the AI systems behind ChatGPT and Claude, a growing number of researchers and entrepreneurs are pivoting toward what they call "world models." These are AI systems designed to understand and predict how the physical world works, not just how to predict the next word in a sentence.

The shift reflects a fundamental limitation of text-based AI. A chatbot trained on all of humanity's books, articles, and online content can write coherent essays and answer questions, but it cannot pick up a coffee mug or navigate a room.

"There's all the geometry of the world, the dynamic of how I move my hand, the physical interaction of the contact with the cup. This is much more complex than just predicting the next word in a sentence," explained Martial Hebert, dean of computer science at Carnegie Mellon University.

Martial Hebert, Dean of Computer Science at Carnegie Mellon University

For roboticists and AI researchers, world models represent the bridge between pure language understanding and what the industry calls "physical AI" or "embodied AI," the ability for machines to perceive, reason about, and act in three-dimensional space.

How Is General Intuition Using Gaming Data Differently?

General Intuition's breakthrough lies not just in using video game footage, but in leveraging the action data embedded within it. Most competitors try to infer what a player did by analyzing video alone, but General Intuition extracts the precise button presses and timing from gaming clips. This distinction matters enormously for training AI to understand causality, the relationship between an action and its consequence.

The company spun out of Medal, a platform where gamers upload and share video game clips. That existing repository of hundreds of millions of hours of gameplay, paired with the action labels, gave General Intuition a proprietary dataset that would be prohibitively expensive to recreate manually. During a demonstration at the company's New York office, a quadrupedal robot navigated an office space using the same AI model that had been playing Fortnite for 100 hours straight. Remarkably, the robot required only eight minutes of real-world robotics data to fine-tune the model for its specific hardware, and that data was collected on the street, not in the office where it was deployed.

"We view this as just the next stage of future pre-training. 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," said Pim de Witte, General Intuition's 31-year-old co-founder and CEO.

Pim de Witte, Co-founder and CEO at General Intuition

What Makes This Funding Round Significant?

General Intuition's $320 million Series B round, announced on June 25, 2026, brings the company's total disclosed funding to $454 million since its launch in October 2025. The round was led by Khosla Ventures and included participation from General Catalyst, Jeff Bezos, Eric Schmidt, Nico Rosberg, and researchers at Google DeepMind and MIT.

Vinod Khosla, whose firm led the investment, framed the bet in terms of a quantum leap in AI capability.

"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 stated.

Vinod Khosla, Founder of Khosla Ventures

The vast majority of the funding will go toward scaling compute capacity through a partnership with CoreWeave, with the company planning to focus on pre-training the next version of its model. A portion has been earmarked for making its API more broadly available by the end of summer 2026.

How Are Other Companies Approaching World Models?

General Intuition is not alone in recognizing the potential of world models. Several prominent AI researchers have launched competing ventures, each with a slightly different vision for how these systems should work.

• Overworld: Founded by computer scientist Louis Castricato, who left his doctoral studies at Brown University, this startup is building video game worlds where scenes can adapt dynamically as virtual characters interact with them. Castricato emphasized that his approach optimizes for interaction above all else.
• World Labs: Led by Fei-Fei Li, known as the "Godmother of AI," this San Francisco-based startup is exploring how AI can learn the statistical structure of space and time, understanding how light falls on surfaces and how objects respond to physical forces.
• Advanced Machine Intelligence Labs: Founded by Yann LeCun, who stepped down as Meta's chief AI scientist, this Paris-based company is investigating how world models enable AI agents to predict the consequences of their own actions.

Fei-Fei Li has attempted to create a taxonomy of world models to clarify the confusion around competing definitions. She identified three categories: "renderers" that prioritize visual quality but cannot reliably teach robots, "simulators" that faithfully represent physical structure, and "planners" that predict what an AI agent should do in unstructured environments.

"A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first," Li wrote.

Fei-Fei Li, Founder of World Labs

Steps to Understanding How World Models Differ from Language Models

• Training Data Source: Language models learn from text, images, and video scraped from the internet. World models additionally incorporate action labels, physics simulations, and spatial reasoning to understand how the physical world responds to intervention.
• Generalization Capability: Language models excel at predicting the next word or pixel. World models aim to generalize from simulated environments to real-world robotics tasks with minimal additional training data.
• Practical Application: Language models power chatbots and coding assistants. World models are designed to power robots, autonomous systems, and interactive simulations that must respond realistically to physical constraints.

What Are the Ethical Boundaries General Intuition Is Setting?

De Witte, who spent three years working in the humanitarian space including with Doctors Without Borders, has drawn a clear ethical line for how General Intuition's technology will be used. The company will not develop agents for lethal autonomy or military applications, though de Witte said he is comfortable with the models being used for search and rescue missions.

De Witte's values also extend to addressing AI-driven job displacement. The company recently launched Nerve, a jobs marketplace that lets gamers earn money using their existing setups. Workers can start with data labeling tasks and eventually move toward robot teleoperation and other roles. De Witte noted that Medal's user base is precisely the generation most exposed to AI-driven displacement, and he wants them to have a financial stake in what's coming next.

Rather than building end-user products like self-driving cars, General Intuition positions itself as an ecosystem enabler, similar to Anthropic or OpenAI. The company plans to license its agentic model to other developers, making it easier for others to build applications on top of the technology. Today, the startup has a handful of customers in gaming, simulation, and robotics sectors.

Why Is Venture Capital So Bullish on World Models Right Now?

Venture capitalists are betting heavily on world models because they see them as the next frontier after large language models have matured. Steve Jang, co-founder and managing partner at Kindred Ventures, which is investing in Overworld and other world model companies, believes the future will feature many different types of models with different philosophies and architectures, not a single dominant approach.

The market opportunity is substantial. If world models can reliably train robots to perform complex physical tasks with minimal real-world data collection, the cost and timeline for deploying robotic systems across manufacturing, logistics, and other industries could drop dramatically. That potential has attracted backing from some of the most prominent names in technology and venture capital, signaling confidence that this is where AI development is heading next.