World Labs · 34× · vu le 03 sept. 2026
Atlas is a multimodal "Omni-World-Model" from World Labs (co-founder: Fei-Fei Li), unveiled on September 1, 2026. It is a multimodal autoregressive diffusion Transformer trained from scratch on text, images, video, and 3D data, generating videos with pixel-perfect, native camera control from just a few reference images, as well as explicit 3D reconstructions (point clouds, 3D Gaussian Splats). The model is currently in early access for selected partners; an official price, model paper, or general availability date were not published at launch. Atlas is intended to power future products like Marble and is also positioned for robotics simulation applications.
Fonctionnalités
| Fine-Tuning | Not offered/not documented; access only via early access and future cloud APIs, no downloadable weights |
| License | Not open source; proprietary, cloud-accessible model |
| Max Resolution | 1440p |
| Max Video Length | Up to 1 minute |
| Platform | Cloud-based; proprietary model to power Marble and enterprise cloud APIs, no downloadable local model weights |
| Price | Not disclosed (no price at launch, early access only) |
| Release Date | September 1, 2026 (announcement, early access) |
Plus de produits dans cette catégorie: Génération de mondes 3D
Preuves (34)
Company Analysis: World Labs
As of 22/08/2026On balance, the opportunity set for World Labs outweighs the risks for investors who can tolerate high volatility and illiquidity. The company sits at the intersection of several powerful secular trends—spatial computing, robotics, simulation, and generative AI—and has already converted cutting‑edge research into a commercial product (Marble) with growing ecosystem visibility. Its capital base (~$1.2–1.3B raised), blue‑chip strategic investors, and world‑class technical team provide a multi‑year runway to iterate on models and infrastructure. While valuation is likely demanding and financial transparency is limited, the combination of rising mindshare (as reflected in search and media trends), differentiated technology, and potential to become a foundational layer for 3D/physical AI supports a BUY stance for long‑term, risk‑seeking capital. Conservative or valuation‑sensitive investors may prefer a smaller position size or staged entry, but the asymmetric upside justifies inclusion in a high‑conviction frontier‑AI portfolio.
Key Takeaways
- World Labs is a privately held frontier AI company founded by Fei‑Fei Li that focuses on “spatial intelligence” and world models—systems that reconstruct, generate, and simulate 3D environments from text, image, or video prompts, with its flagship product Marble launched commercially in November 2025. (worldlabs.ai)
- The company has raised roughly $1.2–1.3B across multiple rounds, including a $1B round in February 2026 led by investors such as Autodesk, AMD, Nvidia, Fidelity, Emerson Collective, and top-tier VCs, implying multi‑billion‑dollar private valuations and positioning World Labs among the best‑funded AI startups globally. (cbinsights.com)
- Marble is gaining traction as a closed‑API generative 3D world model used for VR/AR, gaming, design, and simulation workflows; it has moved from early demos to public availability with subscription billing, iterative model upgrades (e.g., Marble 1.1), and growing ecosystem integrations, suggesting early but real product–market fit. (docs.worldlabs.ai)
- Competition is intensifying from foundation‑model players (OpenAI, xAI, Mistral, etc.) and specialized 3D/graphics ecosystems (NVIDIA Omniverse, Unity, Runway, Stability AI), but World Labs is differentiated by a deep research pedigree (ImageNet/NeRF co‑authors) and a narrow focus on physical/spatial understanding rather than generic text or image generation. (storage.googleapis.com)
- As a pre‑IPO, loss‑making, infrastructure‑style AI company with heavy compute needs and no disclosed revenue scale, World Labs carries high execution and monetization risk; however, its capital base, strategic investors, and first‑mover advantage in world models create asymmetric upside if spatial AI becomes a core layer of the AI stack over the next 3–7 years. (bloomberg.com)
Action Ideas
For investors with access to private shares or pre‑IPO secondary markets, World Labs offers leveraged exposure to the emerging world‑model / spatial‑AI layer of the AI stack. The company combines (1) a top‑tier founder and research team (Fei‑Fei Li, NeRF co‑authors), (2) a large capital cushion (~$1.2–1.3B raised) that can fund multi‑year R&D, and (3) a differentiated product (Marble) that is already commercially available via API and subscriptions. Strategic investors like Autodesk, AMD, Nvidia, Cisco, and major VCs validate both the technology and its industrial use cases (design, robotics, simulation). If world models become as central to 3D and robotics as LLMs are to text, World Labs could justify valuations well above current private marks over a 3–7 year horizon.
Horizon: 60 mo.
For investors already holding private exposure at very high implied valuations (reports of ~$5B+ post‑money), de‑risking or trimming could be prudent. The funding narrative has moved faster than visible commercial traction: despite the $1B round and marquee investors, there is limited evidence of large‑scale recurring revenue, and Marble remains a relatively new product in a rapidly commoditizing generative‑AI landscape. If hyperscalers or open‑source ecosystems deliver competitive world‑model capabilities bundled with broader AI suites, World Labs may struggle to sustain premium pricing or justify unicorn‑plus valuations, especially if capital markets cool for AI infrastructure plays.
Horizon: 24 mo.
For investors with moderate exposure at earlier, lower valuations, a hold stance is reasonable while monitoring execution. The company has cleared key milestones—public launch of Marble, iterative model upgrades, and a large 2026 funding round—yet still needs to prove durable revenue growth, ecosystem lock‑in, and defensibility versus larger AI platforms. Waiting for clearer signals on enterprise adoption (e.g., design, robotics, digital‑twin workflows) and potential IPO timing could improve risk‑reward visibility before adding or exiting. In the meantime, the position offers optionality on spatial AI becoming a core enabling technology across industries.
Horizon: 36 mo.
Google Trends · ↗ rising
Search interest for “World Labs” and its product “Marble” over the last two years appears to follow a classic emerging‑startup curve: low baseline awareness through much of 2024, a noticeable uptick around the company’s emergence from stealth and early funding coverage, and a sharp acceleration from late 2025 onward as Marble launched publicly and major media (TechCrunch, Time, deep‑learning newsletters) and ecosystem partners began highlighting the product. Subsequent spikes in early 2026 coincide with the $1B funding round and strategic investments, after which interest remains elevated but volatile—suggesting growing brand recognition and ongoing newsflow rather than a one‑off hype event. Overall, the pattern is consistent with a rising trend in global Google search interest, albeit from a small base compared with mega‑cap AI brands.
Contrarian Insights
- • : While many view World Labs primarily as a content‑generation tool for VR/gaming, its strongest long‑term moat may lie in industrial and robotics applications—where accurate 3D world understanding and simulation can reduce capex, accelerate design cycles, and enable safer autonomous systems. This could shift the revenue mix from creative tools toward high‑value B2B infrastructure over time, making the business more resilient and less cyclical than typical creative‑software plays. (investor.cisco.com)
- • : Consensus often assumes that hyperscalers will inevitably dominate world models, but regulatory and ecosystem dynamics may favor specialized, independent labs. Alphabet’s own filings highlight a diverse landscape of foundation‑model providers, including World Labs, and enterprises may prefer multi‑vendor strategies for critical simulation and safety‑related workloads. If regulators push against hyperscaler lock‑in, independent players like World Labs could gain bargaining power and strategic value despite their smaller scale. (storage.googleapis.com)
Sources (8)
- https://www.worldlabs.ai/about
- https://techcrunch.com/2026/02/18/world-labs-lands-200m-from-autodesk-to-bring-world-models-into-3d-workflows/
- https://www.bloomberg.com/news/articles/2026-02-18/ai-pioneer-fei-fei-lis-startup-world-labs-raises-1-billion
- https://investor.cisco.com/files/doc_news/Cisco-Invests-in-Spatial-Intelligence-Pioneer-World-Labs-2025.pdf
- https://time.com/7339513/ai-fei-fei-li-virtual-worlds/
- https://techcrunch.com/2025/11/12/fei-fei-lis-world-labs-speeds-up-the-world-model-race-with-marble-its-first-commercial-product-/
- https://docs.worldlabs.ai/
- https://invideo.io/blog/world-labs-marble-3d-worlds/