As the text-based AI models space had exploded a few years ago, the world models space now seems to be seeing similar momentum.
World Labs, the AI startup co-founded by Stanford professor and AI pioneer Fei-Fei Li, has raised $1 billion in fresh funding. The round drew backing from some of the biggest names in tech, including AMD, NVIDIA, Autodesk, Emerson Collective, Fidelity Management & Research Company, and Sea, among others. Autodesk alone committed $200 million and will serve in an advisory role. The company did not disclose a valuation, though Bloomberg previously reported it had been in funding discussions at approximately $5 billion.
Alongside the funding announcement, World Labs is formally launching Marble — its first commercial product and a clear signal that the company is moving beyond research into real-world deployment.
What Is Marble?
Marble is a multimodal generative world model that lets users create spatially cohesive, high-fidelity, and persistent 3D environments from images, video, or text. Unlike many AI tools that generate flat or transient outputs, Marble produces fully navigable 3D worlds that users can explore, edit, and export.
The product is now open to anyone via freemium and paid tiers — a key differentiator in a space where access remains limited. Google’s Genie 3, for instance, is currently only available to users in the United States, whereas Marble is accessible globally from day one. Users can choose from a curated library of starting images or upload their own, and once a world is generated, they can explore it freely using standard WASD keyboard controls, much like a video game. The result is an experience that feels immediately intuitive and immersive.
Marble also features Chisel, an experimental 3D editor that allows users to sketch out rough spatial layouts — walls, planes, objects — and then apply text prompts to guide the visual style. The approach decouples structure from aesthetics, similar to how HTML defines a web page’s skeleton while CSS governs its appearance. Exports are available as Gaussian splats, traditional meshes, and video files, making Marble practical for integration into existing creative and simulation pipelines.
The Bigger Picture: World Models Are Having Their Moment
World Labs was founded in 2024 by Li alongside co-founders Justin Johnson, Christoph Lassner, and Ben Mildenhall — a team with deep roots in computer vision, generative AI, and graphics research. The company came out of stealth with $230 million in funding, and has now more than quadrupled that figure with this latest raise.
The timing reflects a broader shift in where AI investment and research attention are flowing. For years, the dominant narrative centered on large language models and their ability to process and generate text. Now, a growing cohort of researchers and investors believe the next leap will come from systems that can reason about and interact with the three-dimensional world — so-called “world models.” Competitors like Decart and Odyssey have released public demos, as have some Chinese companies, and Yann LeCun’s world model startup AMI Labs has also attracted significant investor interest, signaling that the category is heating up fast.
Li herself has framed Marble as only a first step. In a recent manifesto, she argued that if large language models taught machines to read and write, world models like Marble could teach them to see and build — with implications stretching from gaming and VFX to robotics, scientific simulation, and eventually medicine.