Turn a prompt or reference image into a 3D scene you can navigate.
World Labs Funding History
World Labs develops spatial AI models and tools for generating and working with three-dimensional environments. Founded by Fei-Fei Li and fellow researchers, it applies world-model research to products such as Marble, which creates persistent 3D worlds from visual and textual inputs.
Company Overview
Founded 2024 · San Francisco, California
Official WebsiteWorld Labs Funding Timeline
World Labs first disclosed its early financing when it emerged from stealth. Its later funding announcement came after it had introduced Marble, a tool for generating persistent 3D worlds.
Growth Round
$1B
World Labs said the financing would advance spatial intelligence. Its first product, Marble, was already generating persistent 3D environments from images, video, or text.
Participation included AMD, Autodesk, Emerson Collective, Fidelity Investments, NVIDIA, Sea.
Source · worldlabs.ai ↗Seed
$230M · $1B valuation
World Labs disclosed its cumulative financing when it emerged from stealth. Contemporary reporting described two earlier rounds, rather than one new investment at launch.
Led by Andreessen Horowitz, NEA, Radical Ventures.
Source · techcrunch.com ↗
World Labs Founding Team
World Labs was founded by researchers working across computer vision and graphics. Their shared focus is spatial intelligence: how AI systems represent and reason about three-dimensional environments. The founding team included Christoph Lassner.
- Fei-Fei Li created ImageNet and was a founding director of Stanford’s Institute for Human-Centered Artificial Intelligence.
- Justin Johnson taught computer science at the University of Michigan before co-founding World Labs.
- Ben Mildenhall researched computer graphics and 3D vision at Google Research after completing his doctorate at UC Berkeley.
World Labs Products
World Labs Company Development
A Public Tool for Spatial Models
World Labs made its spatial research tangible through Marble, a tool for generating explorable 3D scenes. That gives the research a practical output beyond an image of a place: a scene that can be approached from different viewpoints.
World Labs Competitive Landscape
An Explorable Scene or a Directed Shot
Marble and Runway both generate visual material. Marble is built around a world the user can navigate, while Runway's video tools focus on a shot's motion, performance and editing. The desired output drives which workflow is relevant.
FAQ
Why does World Labs count two 2024 tranches as a single US$230M seed?
Announcement packaging: an ~April 2024 stealth tranche (~US$200M valuation, a16z and Radical Ventures) plus a NEA-led mid-2024 tranche (reported above US$1B) were disclosed together at the September stealth exit. The company never confirmed either price — a discretion kept through 2026, when the ~US$5B figure remained Bloomberg-reported talks, never a company statement.
Why did Autodesk — a design-software company — anchor the 2026 round?
Because world models are becoming design infrastructure: Autodesk's US$200M, its largest startup investment ever by its own announcement, bought an advisory seat and a research collaboration to bring world models into professional 3D workflows. It also shifted the base — from 2024's academic-angel roster to strategic capital, with NVIDIA and AMD doubling down and Fidelity arriving.
Why are NVIDIA and AMD investors in World Labs while backing its rivals?
Hedging the platform layer: NVIDIA runs Cosmos and Google DeepMind runs Genie in the same race, so both chipmakers hold positions across contenders — NVIDIA via NVentures in the 2024 seed, both doubling down in 2026. The dual role recurs at SSI (Alphabet and Nvidia invested while Google Cloud supplies TPUs) and Cognition (Nvidia in the Series E): compute vendors cannot pick one winner early.
What re-rated World Labs from ~US$1B to a reported ~US$5B in about 17 months?
Product proof where rivals show theory: Marble's November 2025 general availability, the World API (January 2026) and the July 2026 SceniX acquisition with real-to-sim-to-real results — simulation-trained policies running for hours on robot hardware. Against AMI Labs' US$3.5B on a proof and Ineffable's ~US$5.1B on a promise, that is the product-first price — with no disclosed revenue in the trio.
Is Marble just an AI video generator?
No — persistence is the differentiation: Marble produces downloadable 3D environments exportable to Unreal and Unity, where diffusion-video competitors generate clips that cannot be revisited or edited as geometry. Atlas, unveiled September 2026, extends that persistence into an 'omni' model for generation, reconstruction and simulation — what makes the robotics roadmap coherent.
Why is spatial-world generation different from a funding story about video clips?
A spatial representation concerns a scene that can be explored or manipulated; a video clip is an ordered sequence of frames. World Labs' research context concerns spatial intelligence. Its financing history should not be used as proof that every demonstration supports persistent geometry or reliable physical interaction.