World Labs: Marble World Models & $5B Valuation (2024)
World Labs is Fei-Fei Li's spatial intelligence startup, founded 2024 and valued near $5B in 2026, building Marble, an AI model that creates 3D worlds.
World Labs is a San Francisco AI startup founded in January 2024 by Fei-Fei Li, Justin Johnson, Christoph Lassner, and Ben Mildenhall, with a mission to give AI spatial intelligence in 3D space. It has raised $1.23 billion, including a $1 billion round in February 2026 led by Autodesk and NVIDIA, at a valuation reportedly near $5 billion. Its flagship product Marble turns text, images, or video into explorable 3D worlds, available via a freemium plan and the World API.
World Labs is a spatial intelligence startup founded in January 2024 by Stanford AI pioneer Fei-Fei Li along with Justin Johnson, Christoph Lassner, and Ben Mildenhall. Headquartered in San Francisco, it has raised $1.23 billion total, including a $1 billion round in February 2026 from Autodesk, NVIDIA, and AMD. Its flagship product, Marble, generates persistent, navigable 3D environments from text, images, or video.
Founded: 2024 · HQ: San Francisco, CA, USA · Team: 50-100 · CEO: Fei-Fei Li · Funding: $1.23B total raised: $130M Series A extension (Sept 2024, lead: Andreessen Horowitz, NEA, Radical Ventures) + ~$1B round (Feb 2026, lead: Autodesk, with AMD, NVIDIA, Fidelity, Emerson Collective, Sea Limited) · Valuation: ~$5B (reported in funding talks, Jan-Feb 2026)
About World Labs
World Labs was founded in January 2024 by Fei-Fei Li, Justin Johnson, Christoph Lassner, and Ben Mildenhall, and is headquartered at 640 2nd Street in San Francisco, California. Fei-Fei Li serves as CEO and is widely known as the creator of the ImageNet dataset, the Sequoia Professor of Computer Science at Stanford, and co-director of Stanford's Human-Centered AI Institute. Justin Johnson is a former Stanford researcher specializing in generative models, Christoph Lassner came from Meta Reality Labs and Epic Games where he built the Pulsar renderer, a precursor to Gaussian Splatting, and Ben Mildenhall arrived from Google Research as a co-creator of Neural Radiance Fields (NeRF). The founding thesis was that AI had become strong at language and 2D images but still lacked "spatial intelligence," the ability to perceive and reason about the 3D world, and that this gap represented the next major frontier in AI research. World Labs builds what it calls Large World Models (LWMs), AI systems designed to perceive, generate, and reason about three-dimensional space rather than flat pixels or text. Its first commercial product, Marble, launched in limited beta in November 2025 and moved to a formal commercial launch in early 2026. Marble turns a single image, text prompt, video, set of photos, panorama, or rough 3D layout into a persistent, navigable 3D environment that can be explored, edited, and exported. Generated worlds can be brought into standard game engines such as Unreal Engine and Unity, and viewed on VR headsets including Apple Vision Pro and Meta Quest 3, positioning Marble against both creative tools and emerging "world model" research products from larger labs. In January 2026, World Labs launched the World API, a credit-based developer offering priced at roughly $1.00 per 1,250 credits with a minimum purchase of 6,250 credits ($5), letting businesses embed Marble's 3D world-generation engine directly into games, simulations, robotics pipelines, and other applications. In June 2026 the company described its roadmap as spanning three categories of world models: generation (creating 3D content, Marble's current public face), simulation (testing and training environments), and embodiment (spatial intelligence applied to robotics), framing Marble as the "simulation linchpin" of that broader plan. World Labs has raised approximately $1.23 billion in total disclosed funding. The first round was a $130 million Series A extension in September 2024, led by Andreessen Horowitz, New Enterprise Associates, and Radical Ventures, which valued the company at over $1 billion, about five times its initial valuation at founding. The second was a roughly $1 billion round announced February 18, 2026, anchored by a $200 million strategic investment from Autodesk, with additional participation from AMD, NVIDIA, Fidelity Management & Research Company, Emerson Collective, and Sea Limited. Bloomberg reported in January 2026 that the company was in funding talks at a valuation near $5 billion, roughly five times its 2024 valuation, though the exact valuation of the February 2026 round was not officially confirmed. World Labs' business model combines a freemium consumer product, Marble, with subscription tiers ranging from free to $95 per month based on generation quotas and commercial usage rights, and the credit-based World API for developers and enterprises. No public revenue or ARR figures have been disclosed as of mid-2026. The strategic investors are notable for their industry ties: Autodesk's investment signals plans to connect Marble-generated 3D content with Autodesk's design and creative software, while NVIDIA and AMD's participation reflects deep dependencies on GPU compute for training and running world models at scale. Team size has grown quickly from around 42 employees in December 2025 to roughly 60 to 70 by mid-2026, with hiring concentrated at the San Francisco headquarters and focused on research scientists and engineers in 3D graphics, neural rendering, computer vision, and robotics-oriented spatial AI. Open roles are listed on a Greenhouse job board and have also been promoted through Andreessen Horowitz's portfolio jobs page. The founding leadership team, led by CEO Fei-Fei Li alongside co-founders Johnson, Lassner, and Mildenhall, has remained unchanged since the company's January 2024 founding, with no reported layoffs or restructuring. World Labs' research agenda centers on spatial intelligence, described internally as transforming "seeing into doing, perceiving into reasoning, and imagining into creating." The company publishes blog posts and research notes on 3D world generation, the role of generative AI in storytelling, and the technical foundations of spatial reasoning, drawing on the founders' prior work on ImageNet, NeRF, and Gaussian Splatting-adjacent rendering. As of mid-2026, World Labs has not published a formal responsible scaling policy, AI safety framework, or third-party red-teaming program comparable to those at larger frontier-model labs; its safety posture remains research-stage and product-quality focused. In the world-model category, World Labs' closest rivals include Google DeepMind, whose Genie and Veo models touch on interactive and video-based world generation, OpenAI's Sora and related video-to-3D research, Runway's generative video and 3D creative tools, and Tencent's Hunyuan World models, which offer an open alternative for 3D world generation. World Labs differentiates itself through its founders' deep research pedigree in 3D vision and graphics and through strategic distribution ties to Autodesk and NVIDIA, though it remains far smaller than the large frontier labs and has no broad conversational AI product. World Labs is a private, Delaware-incorporated company based in the United States with no disclosed EU AI Act classification, GDPR-specific data residency commitments, or compliance certifications as of mid-2026. There is no indication of an imminent IPO. The company's near-term priorities, following its $1 billion February 2026 round, are to scale Marble and the World API, grow its San Francisco research and engineering team, and advance toward the simulation and embodiment stages of its world-model roadmap.
Mission
Advance spatial intelligence so AI can perceive, generate, reason about, and interact with the 3D world, not just 2D pixels and text.
Products
- Marble (Generative 3D world model (consumer/creator product)): https://marble.worldlabs.ai
- World API (Developer API for 3D world generation): https://docs.worldlabs.ai
Links
Website · GitHub · Twitter · LinkedIn · Blog · Docs
Frequently Asked Questions
What is World Labs and what do they build?
World Labs is a spatial intelligence company founded in January 2024 by Stanford AI pioneer Fei-Fei Li, along with computer vision researchers Justin Johnson, Christoph Lassner, and Ben Mildenhall, headquartered in San Francisco. The company builds Large World Models (LWMs), AI systems designed to perceive, generate, and reason about three-dimensional space rather than flat 2D pixels or text. Its flagship product, Marble, launched in limited beta in November 2025 and formally in early 2026, turns a single image, text prompt, video, or rough 3D layout into a persistent, navigable 3D environment. Marble targets creators, game developers, and visual-effects studios, while the World API, launched in January 2026, lets developers and businesses embed Marble's world-generation engine directly into their own applications and pipelines. Generated worlds can be exported into standard 3D formats for use in engines like Unreal Engine and Unity, and Marble supports viewing on Apple Vision Pro and Meta Quest 3 headsets. One anchor figure of the company's scale is its $1.23 billion in total funding as of February 2026, reflecting strong investor confidence in the world-model category. World Labs positions itself as a research-driven leader in moving AI's understanding from 2D images into full 3D spatial reasoning.
Who founded World Labs and who is the CEO?
World Labs was founded in January 2024 by four co-founders: Fei-Fei Li, Justin Johnson, Christoph Lassner, and Ben Mildenhall. Fei-Fei Li serves as CEO and is one of the most recognized names in AI, known as the creator of the ImageNet dataset that helped trigger the deep learning boom, and she remains the Sequoia Professor of Computer Science at Stanford and co-director of Stanford's Human-Centered AI Institute. Justin Johnson is a former Stanford researcher with a background in generative models and computer vision. Christoph Lassner joined from Meta Reality Labs and Epic Games, where he developed Pulsar, an early differentiable renderer that helped lay the groundwork for techniques like Gaussian Splatting. Ben Mildenhall came from Google Research and is a co-creator of Neural Radiance Fields (NeRF), a foundational technique for generating photorealistic 3D scenes from images. The founding thesis was that AI had mastered language and 2D images but still lacked spatial intelligence, the ability to understand and reason about 3D space, and that this gap was the next major frontier for AI. World Labs emerged from stealth in September 2024 with a $130 million Series A extension. As of mid-2026 there have been no reported changes to the founding leadership team.
How much funding has World Labs raised?
World Labs has raised approximately $1.23 billion in total disclosed funding across two main rounds. The first was a $130 million Series A extension announced in September 2024, led by Andreessen Horowitz, New Enterprise Associates, and Radical Ventures, which valued the company at over $1 billion, about five times its initial valuation when it was founded earlier that year. The second was a roughly $1 billion round announced in February 2026, anchored by a $200 million strategic investment from Autodesk, with additional participation from AMD, NVIDIA, Fidelity Management & Research Company, Emerson Collective, and Sea Limited. Bloomberg reported in January 2026 that World Labs was in funding talks at a valuation near $5 billion, roughly five times its 2024 valuation, though the company has not officially confirmed the exact valuation of the February 2026 round. The strategic investors are notable: Autodesk's stake signals plans to integrate Marble-generated 3D content into Autodesk's design and creative software, while NVIDIA and AMD's involvement points to deep ties around GPU compute for training and running world models. No public revenue or ARR figures have been disclosed as of mid-2026. There is no indication of an imminent IPO; World Labs remains a private, venture- and strategic-backed company focused on scaling Marble and the World API.
What products does World Labs make?
World Labs' primary product is Marble, a generative world model that creates persistent, explorable 3D environments from a single image, a text prompt, a video, multiple photos, a panorama, or a rough 3D layout. Marble launched in limited beta in November 2025 and moved to a formal commercial launch in early 2026, with a freemium pricing structure ranging from a free tier to $95 per month based on generation quotas and commercial usage rights. In January 2026, World Labs launched the World API, a credit-based developer offering priced at roughly $1.00 per 1,250 credits with a minimum purchase of 6,250 credits ($5), letting businesses integrate Marble's 3D-world generation directly into games, simulations, robotics pipelines, and creative tools. Generated 3D worlds can be exported to standard formats for use in engines like Unreal Engine and Unity, and can be explored on VR headsets including Apple Vision Pro and Meta Quest 3. The consumer Marble product targets game developers, filmmakers, and VFX artists, while the World API targets enterprise and developer customers building on top of the underlying world model. As of mid-2026, World Labs has described its roadmap as spanning three categories of world models: generation, simulation (Marble's current focus), and embodiment for robotics. The company has not announced an open-source model release, and no products have been discontinued since its 2024 founding.
Where is World Labs headquartered and how big is the team?
World Labs is headquartered in San Francisco, California, at 640 2nd Street. The company has not publicly disclosed additional international offices as of mid-2026, and its operations appear concentrated at the San Francisco hub. Team size has grown quickly: reports put headcount at around 42 employees in December 2025, rising to roughly 60 to 70 employees by mid-2026. This growth coincided with the company's $1 billion February 2026 funding round, which is expected to fund further hiring in research and engineering roles. Open positions are listed on a Greenhouse job board and have also been promoted via Andreessen Horowitz's portfolio jobs page, reflecting a16z's early investment in the company. Hiring has focused heavily on research scientists and engineers with backgrounds in 3D graphics, neural rendering, computer vision, and robotics-oriented spatial AI. There have been no reported layoffs or restructuring since the company's founding in January 2024. As a young, well-funded startup, World Labs is expected to keep expanding its San Francisco team through 2026 as Marble and the World API scale.
What is World Labs' mission or research focus?
World Labs' stated mission is to advance spatial intelligence, giving AI systems the ability to perceive, generate, reason about, and interact with the three-dimensional world, rather than being limited to 2D pixels and text. The company describes its goal as transforming seeing into doing, perceiving into reasoning, and imagining into creating, through what it calls Large World Models, or LWMs. This mission draws directly on the founders' research lineage: Fei-Fei Li created the ImageNet dataset that helped launch the deep learning era, Ben Mildenhall co-created Neural Radiance Fields (NeRF) for photorealistic 3D scene generation, and Christoph Lassner built Pulsar, an early renderer that influenced techniques like Gaussian Splatting. World Labs publishes research and product blog posts covering topics like 3D world generation, the role of generative AI in storytelling, and the technical foundations of spatial reasoning. The company has framed its long-term roadmap around three categories of world models: generation (creating 3D content), simulation (Marble's current focus, used for testing and training), and embodiment (applying spatial intelligence to robots). As of mid-2026, World Labs has not published a formal AI safety framework, responsible scaling policy, or red-teaming program comparable to those at large frontier-model labs. Its safety posture is best described as research-stage and focused primarily on content and product quality rather than frontier-model risk management. The company's broader bet is that spatial intelligence, not just language, will be a defining capability of next-generation AI systems.
Is World Labs compliant with SOC 2, GDPR, HIPAA?
As of mid-2026, World Labs has not published any confirmed SOC 2, ISO 27001, or HIPAA certifications, and the company does not appear to operate a public trust center. No dedicated privacy policy or terms-of-service URL could be independently verified on the company's main site at the time of writing, which is typical for an early-stage startup that has not yet built out a formal enterprise compliance program. There is no public information about a default data-retention policy for Marble or World API usage, nor about an enterprise zero-retention or opt-out tier. For GDPR, no data processing agreement, standard contractual clauses, or EU data residency commitments have been disclosed. World Labs has not announced a HIPAA-eligible tier or business associate agreement option. Under the EU AI Act, World Labs has not published a formal classification, though as a generative AI provider it would likely need to meet general-purpose AI provider obligations if Marble or the World API scale into EU markets. Enterprise customers evaluating World Labs for regulated workloads should treat its compliance posture as unverified and request documentation directly from the company. This is an area likely to develop as the company matures from its current early-stage, consumer- and developer-focused footing.
Who are World Labs' main competitors?
World Labs competes in the fast-growing world model category, where its closest rivals include Google DeepMind, whose Genie and Veo models also touch on interactive and video-based 3D-like environments, and OpenAI, whose Sora video model and related research push toward video-to-3D generation. World Labs' key differentiator is its founder team's deep research pedigree in 3D vision and graphics, including the creators of NeRF, ImageNet, and an early Gaussian Splatting precursor, giving it a strong technical foundation specifically for spatial reasoning rather than general multimodal chat. Against Google DeepMind and OpenAI, World Labs is far smaller and lacks a broad consumer LLM product, but it benefits from being focused on 3D world generation as its core product rather than a side feature. Runway is another competitor, offering generative video and 3D creative tools aimed at filmmakers and designers, an audience that overlaps with Marble's. Tencent's Hunyuan World models are also emerging as open competitors in 3D world generation, particularly for developers who want self-hostable alternatives. World Labs' strategic backing from Autodesk and NVIDIA gives it a potential distribution edge into professional 3D design and GPU-compute ecosystems that smaller rivals lack. One category where World Labs currently has no real answer is general-purpose conversational AI or coding assistants, areas dominated by large frontier-model labs. Tencent's Hunyuan World initiative is a competitor that has emerged more prominently in the last 12 months, with no comparable public offering in 2024.