World Labs

World Labs is an American artificial intelligence research company founded in 2024 by Fei-Fei Li, the Stanford professor and ImageNet creator. The company is headquartered in San Francisco with research operations linked to Stanford. World Labs is pursuing "spatial intelligence," a research direction focused on AI systems that understand, generate, and reason about three-dimensional environments rather than the two-dimensional data (text, images) that dominates large-language and image-generation models. The company's first commercial product, Marble, generates and edits persistent 3D environments from text, images, video, or 3D layouts.

At a glance

• Founded: 2024 in San Francisco. Public launch September 2024.
• Status: Private. Approximately 50 employees as of early 2026.
• Funding: Approximately $1.23 billion cumulative across two reported rounds. $230 million seed at a $1.3 billion valuation in September 2024 (later expanded). $1 billion at an undisclosed substantially higher valuation in February 2026, with Autodesk contributing $200 million as a strategic investor.
• CEO: Fei-Fei Li (founder)
• Other co-founders: Justin Johnson (University of Michigan computer-science professor), Christoph Lassner (formerly Reality Labs Research), Ben Mildenhall (formerly Google, co-author of NeRF).
• Open weights: Mixed. Some Marble research components have been released; Marble itself is closed.
• Flagship products: Marble (3D world generation, public commercial launch following limited beta).

Origins

World Labs was founded in 2024 by Fei-Fei Li and three co-founders with deep computer-vision and 3D-reconstruction credentials. Li had been the Sequoia Capital Professor of Computer Science at Stanford University, founding co-director of Stanford HAI (Human-Centered AI Institute), and former Chief Scientist of AI and Machine Learning at Google Cloud. She is widely associated with the creation of ImageNet, the 2009 dataset that catalyzed the deep-learning revolution in computer vision, and with the popularization of the term "AI godmother" in technology media.

The founding thesis was articulated in Li's 2023 memoir, "The Worlds I See," and in subsequent public talks: that the field's progress on language and 2D image understanding has produced an AI capability gap on three-dimensional spatial reasoning, and that closing this gap requires a research program oriented around 3D scene representation and world models rather than the LLM and diffusion-model paradigms.

The co-founders brought specific technical credentials matching the spatial-intelligence research direction. Justin Johnson, a computer-science professor at the University of Michigan, had previously been Li's PhD student at Stanford and is a leading researcher in computer vision and 3D scene understanding. Christoph Lassner, formerly of Meta's Reality Labs Research, contributes graphics and 3D-reconstruction expertise. Ben Mildenhall, formerly of Google, is a co-author of the influential NeRF (Neural Radiance Fields) paper that catalyzed the modern 3D-reconstruction research line.

The company's first reported funding round closed in September 2024 with $230 million at a $1 billion-class valuation, led by Andreessen Horowitz, Radical Ventures, and NEA. Marble launched in limited beta in November 2024 and became broadly commercially available in 2025. In February 2026, World Labs raised an additional $1 billion in a strategic round including AMD, Autodesk ($200 million as a strategic anchor), Emerson Collective, Fidelity, NVIDIA, and Sea. The Autodesk investment was paired with a strategic-adviser relationship for integrating Marble and World Labs technology into Autodesk's 3D workflow products.

Mission and strategy

World Labs's stated mission is "to build large world models that achieve spatial intelligence." Li has framed the strategy as a research direction distinct from the dominant LLM and diffusion-model paradigms, focused on 3D scene generation, persistent world representation, and AI agents that can reason about physical environments.

The strategy combines three threads. First, fundamental research on world models and 3D scene representation, scaled to commercial-tier compute. Second, the Marble product line as a commercial application of world-model technology, targeted at gaming, visual effects, virtual reality, robotics, and 3D design workflows. Third, strategic partnerships with established companies in adjacent industries (Autodesk for design, AMD and NVIDIA for compute) that position World Labs as the spatial-AI infrastructure layer rather than as a competitor to existing tools.

The competitive premise is that spatial intelligence is a separable capability domain from language and image generation, that the LLM-flagship labs (OpenAI, Anthropic, Google DeepMind) have not prioritized 3D understanding, and that a focused research program with the right talent depth can produce a meaningful capability lead. The thesis is reinforced by parallel investments at peer labs (Meta AI / FAIR JEPA work, AMI Labs world-model research, Google DeepMind's Genie line) but World Labs is the first dedicated commercial spatial-AI lab at scale.

Models and products

• Marble. World-model platform that generates and edits persistent 3D environments from text, images, video, or 3D layout inputs. Available in free and paid tiers, with export options including 3D meshes and video. Released in limited beta in November 2024 and broadly available since 2025. Targeted at gaming developers, visual-effects studios, virtual-reality content creators, and 3D-design professionals.
• Research artifacts. World Labs has published research papers and released some open-source research components related to spatial reasoning and 3D-scene generation, though the commercial Marble product is closed.

The product strategy beyond Marble has not been publicly detailed. The company's research roadmap, articulated by Fei-Fei Li in public talks, emphasizes scaling world models to support fully interactive AI agents in 3D environments.

Benchmarks and standing

World Labs models are not represented on the standardized LLM benchmarks (Artificial Analysis Intelligence Index, LMArena, GPQA Diamond, SWE-bench) because spatial-intelligence and world-model capability evaluations require different benchmark approaches. The field has not yet converged on standardized 3D-generation benchmarks comparable to the LLM leaderboards, which constrains direct comparison with peers.

Industry coverage of Marble has been generally positive on the technical achievement of generating persistent 3D environments from natural-language inputs, with 3D-design and gaming-industry publications treating the product as a meaningful new tool. Whether the capability translates to durable commercial adoption depends on factors (output quality, integration with existing workflows, pricing) that are still being established in the market.

Leadership

As of April 2026, World Labs's senior leadership includes:

• Fei-Fei Li, Chief Executive Officer and founder. Sequoia Capital Professor of Computer Science (on leave) and founding co-director of Stanford HAI. Creator of ImageNet (2009). Former Chief Scientist of AI and Machine Learning at Google Cloud (2017 to 2018). Author of "The Worlds I See" (2023). Public face for World Labs and a widely cited figure in the spatial-intelligence research thesis.
• Justin Johnson, co-founder. Computer-science professor at the University of Michigan (on leave). PhD from Stanford under Fei-Fei Li. Computer-vision and 3D-scene understanding research lead.
• Christoph Lassner, co-founder. Formerly senior researcher at Meta's Reality Labs Research; graphics and 3D-reconstruction expertise.
• Ben Mildenhall, co-founder. Formerly Google research scientist; co-author of the NeRF (Neural Radiance Fields) paper that catalyzed modern 3D-reconstruction research.

The team is reported to be approximately 50 employees as of early 2026, with hiring concentrated in computer-vision, 3D-graphics, and machine-learning research roles. The Stanford and Michigan academic connections continue to support recruitment of senior PhD-level researchers.

Funding and backers

World Labs's funding history through April 2026 includes the September 2024 $230 million round at a $1 billion-class valuation, with Andreessen Horowitz, Radical Ventures, and NEA as lead investors, and the February 2026 $1 billion round at an undisclosed substantially higher valuation. The Marble launch in late 2024 was an interim milestone between the two rounds.

The February 2026 round added strategic-corporate investors well-aligned to the spatial-AI thesis. AMD's investment supports compute-infrastructure access. NVIDIA's investment is paired with the company's research and product use of NVIDIA silicon. Autodesk's $200 million strategic investment positions World Labs as Autodesk's preferred AI partner for 3D-design workflows, complementing Autodesk's existing CAD and entertainment-industry product lines. Sea (the Singapore-based technology conglomerate) and Emerson Collective add geographic and strategic-foundation diversity. Fidelity's participation provides public-market-style capital ahead of any potential IPO.

The cumulative funding of approximately $1.23 billion places World Labs among the better-funded Insurgent labs, behind only the largest cohort (SSI, Thinking Machines Lab, AMI Labs) by capital scale.

Industry position

World Labs occupies a structurally distinctive position among Insurgent labs. The combination of Fei-Fei Li's research credentials, the four-co-founder team's specific focus on spatial intelligence, the early commercial product (Marble) that distinguishes World Labs from pre-product peers, and the strategic Autodesk partnership produces a profile differentiated from the more LLM-focused Insurgent cohort.

Strategic risks include the open question of whether spatial intelligence becomes a separable commercial market large enough to support a $2 billion-valuation company, the competitive pressure from Frontier and Incumbent labs as they invest more in 3D and world-model research (notably Meta's JEPA work, Google DeepMind's Genie line, AMI Labs's research direction), and the integration complexity of selling into established 3D-design, gaming, and VFX industries with mature workflow expectations.

Strategic strengths include Li's unmatched research credibility in computer vision, the early product launch and commercial traction with Marble, the strategic Autodesk partnership that aligns with a major existing 3D-software vendor, and the focused research direction that does not require winning on LLM benchmarks to succeed.

Competitive landscape

World Labs competes with several Frontier and Insurgent labs:

• Meta AI / FAIR. The JEPA-family research from LeCun's group at Meta is the closest published parallel to spatial-intelligence research, though Meta's commercial direction is now closed-source frontier (Muse Spark) rather than spatial-AI products.
• AMI Labs. LeCun's new venture continues the JEPA research line as a deliberate alternative to LLMs and overlaps with World Labs's spatial-AI thesis, though AMI is pre-product and World Labs has Marble in market.
• Google DeepMind. Genie multimodal foundation models address overlapping spatial-AI use cases. DeepMind's robotics and embodied-AI research also overlaps.
• Physical Intelligence. Robotics-foundation-model startup with substantial 3D-world capabilities; positioned as a Specialized lab in the broader taxonomy.
• Wayve. Robotics and autonomous-driving foundation-model company; another Specialized peer.
• Runway. Video and 3D content-generation startup; commercial overlap with Marble in some VFX and creative workflows.
• Other Insurgent labs. SSI, Thinking Machines Lab, and the broader 2024 to 2025 founder cohort compete for senior AI talent and capital.

Outlook

Several open questions affect World Labs's trajectory in 2026 and 2027:

• Marble adoption metrics across gaming, VFX, VR, and 3D-design industries, which determine whether the spatial-AI commercial market scales.
• Continued Marble product development, including expanded export formats, integration with 3D-design tools beyond Autodesk, and potential consumer-facing variants.
• Research and product development from the AMI Labs JEPA program and Google DeepMind's world-model line, which will indicate whether spatial intelligence remains a defensible market for a focused player.
• The strategic implications of the Autodesk partnership, including any deeper product integration or possible acquisition interest.
• Continued senior-talent recruitment from computer-vision and graphics academia, which the founding team's Stanford and Michigan connections continue to support.
• Future funding rounds or potential IPO candidacy, given the public-market interest signaled by Fidelity's February 2026 participation.

Sources

• TechCrunch: World Labs lands $1B, with $200M from Autodesk, to bring world models into 3D workflows. February 2026 funding round.
• PYMNTS: World Labs Raises $1 Billion to Scale Spatial AI. Strategic partnership context.
• Financial Content: Fei-Fei Li's World Labs Unveils 'Marble' to Conquer the 3D Frontier. Marble product launch context.
• The Decoder: Fei-Fei Li's World Labs raises one billion dollars for spatial intelligence. Strategic analysis.
• World Labs official site. Marble product page.
• Wikipedia: Fei-Fei Li. Founder background.