The AI industry's obsession with chatbots is fading, replaced by a new frontier: world models, AI systems that don't just describe the world but understand how it works. Companies like World Labs and Advanced Machine Intelligence Labs are raising billions to build AI that can simulate physics, predict consequences, and help robots, architects, and game developers work in virtual environments before touching the real world.

What Are World Models and Why Do They Matter?

World models represent a fundamental shift in how AI systems approach understanding. Rather than simply predicting the next word in a sentence, these systems learn to model how objects occupy space, collide, move, and break. Fei-Fei Li, the Stanford professor who co-founded World Labs in 2024, has spent her career teaching machines to see, and she frames the category in three parts: renderers that create what you can see, simulators that model what happens next, and planners that choose actions toward a goal.

The practical applications span industries. A robot can learn safer behaviors in simulation before operating in a warehouse. An architect can walk through a building that exists only as generated code. A self-driving car can train on rare road events without waiting for them to occur naturally. Game developers can generate interactive 3D worlds from text descriptions. These aren't theoretical benefits; they're the reason investors are writing checks that would have seemed absurd for any AI company two years ago.

How Are Companies Building World Models?

• Generative 3D Creation: World Labs' Marble product lets users generate downloadable 3D worlds from text, image, or video prompts, with integrations for tools like Unity and Unreal Engine, and offers paid plans including a $95-per-month Max tier for commercial use.
• Physics-Aware Simulation: Research projects like WorldAct and HY-World 2.0 are turning static 3D scenes into editable, interaction-ready environments with object-level meshes and collision-aware manipulation capabilities.
• Game Data Training: General Intuition, a New York startup, is training world models from billions of gaming videos, which provide rich examples of physics, object interaction, and environmental dynamics.
• Real-Time Diffusion Models: Companies like Overworld are developing real-time diffusion world models specifically for game engines, enabling faster iteration and more responsive environments.

Who Is Funding This Shift?

The funding numbers tell the story of investor conviction. World Labs raised $230 million in 2024, then secured a much larger 2026 financing round that brought in Nvidia, AMD, Fidelity, Emerson Collective, and Autodesk. Autodesk alone committed $200 million in February 2026 and pledged to collaborate on research and development, a signal that design software companies see world models as essential infrastructure.

Yann LeCun, the former Meta chief AI scientist who left the company in late 2025, launched Advanced Machine Intelligence Labs (AMI Labs) around the conviction that large language models won't be sufficient for general-purpose AI. His company announced a $1.03 billion seed round in March 2026, with backers including Nvidia, Samsung, Toyota, Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Le Monde reported the round valued the Paris-based company at approximately 3 billion euros.

"Machines need internal models of the world, not only systems trained to predict the next token," LeCun has argued, emphasizing that a model that writes about gravity hasn't proved it understands gravity.

Yann LeCun, Founder of Advanced Machine Intelligence Labs

General Intuition, focused on world models trained from game data, announced a $133.7 million seed round. Causal Labs is applying physics-modeling ideas to weather and climate prediction. These companies aren't all chasing the same customer, which is exactly why the category is attracting so much capital.

Why Are Investors Moving Away From Chatbots?

Chatbots have become familiar very quickly. Coding assistants, search assistants, and office copilots are now standard parts of the software stack, and the biggest labs are spending heavily to keep incremental gains coming. World models offer investors a different story: AI that solves problems chatbots can't touch.

The skeptic's case remains fair. Video generation, physics simulation, digital twins, and model-predictive control all existed before investors started using the term "world models" constantly. A new label doesn't magically solve contact physics, long-horizon planning, or data quality. Anyone who has watched a generated video melt a hand into a coffee cup knows the gap between convincing output and usable understanding.

But the category isn't just a rebrand. The money is arriving because several old threads are finally being pulled together: generative 3D technology, self-supervised video learning, cheaper inference costs, simulation data, and growing robotics demand. Google DeepMind has already pushed in this direction with Genie, its line of models for interactive environments. Academic work is accelerating too, with papers like WorldAct (May 2026) and HY-World 2.0 (April 2026) describing methods for turning generated 3D worlds into editable, interaction-ready scenes.

What Does This Mean for the Future of AI?

If you're building or buying AI software, this is the shift to watch. The easy version of AI was asking a model to describe the world. The harder version is asking it to predict what happens when something moves inside it. That second task is where factories, autonomous systems, games, architecture, and robotics live.

World models represent a maturation of AI beyond language. They acknowledge that understanding the world requires more than text; it requires spatial reasoning, physics intuition, and the ability to simulate consequences. The companies and investors betting on this shift are betting that the next wave of AI value won't come from better chatbots, but from systems that can help humans and machines navigate, design, and control complex physical and digital environments.