World Labs

1.1 Identity, product surface, and what is actually public

Public sources agree on the broad business description: World Labs is building spatial-intelligence software around persistent, navigable 3D worlds, with Marble as the main product surface and the World API as the programmable interface for developers. The current documentation and launch materials support a concrete product story -- text, image, video, panorama, and coarse 3D inputs become editable 3D worlds that can be rendered on the web or exported into downstream tools. What is less cleanly settled is the company identity itself. HokAI's October 2026 profile says World Labs was founded in January 2024 in San Francisco by Fei-Fei Li, Justin Johnson, Christoph Lassner, and Ben Mildenhall, while a Silicon Valley Invest Club profile lists a 2017 founding year for the same business. That conflict is material because later diligence on team velocity, capital efficiency, and valuation step-ups depends on whether the company should be treated as a 2024 startup or as a longer-running effort whose public emergence simply came later.[CO001, CO002, CO003, CO004, CO005, CO010, CO011, CO014]

1.2 Leadership, funding, and the ownership transition

The core leadership picture is clearer than the founding-date record. Across current third-party profiles, Fei-Fei Li remains the central operating and narrative figure, with Johnson and Mildenhall still positioned as product and technical leaders around the AMD transaction. The September 2026 agreement to sell World Labs to AMD for about $8.2 billion in stock is the defining ownership event in the current refresh cycle, and the transaction matters beyond price because it reframes World Labs from a venture-backed standalone platform into a research-and-product asset inside a semiconductor company if it closes. Funding history is directionally strong but still incomplete in governance detail: the public record supports about $1.23 billion of disclosed capital raised, including a $1 billion round announced on February 18, 2026, but it does not provide a public shareholder register, board composition, control rights, or full commercial terms around strategic investors such as Autodesk and AMD.[CO004, CO006, CO007, CO008, CO009]

1.3 Milestone cadence and the remaining diligence gaps

The milestone cadence is real and unusually compressed. World Labs moved from an early public research moment in late 2024 to RTFM in October 2025, public Marble availability in November 2025, the World API in January 2026, SceniX in July 2026, Atlas in September 2026, and a signed AMD deal later that same month. That pace supports the view that this is not a pre-product story anymore. At the same time, the strongest adverse evidence in the assigned pool is not about product absence; it is about commercial opacity. Independent analysts still say World Labs has disclosed no revenue or ARR despite charging for Marble, offering paid API credits, and attracting both a $5 billion private-round valuation discussion and an $8.2 billion pending exit. Later chapters should therefore treat customer scale, exact headcount, board oversight, and post-close pricing continuity as live diligence items rather than facts established by this chapter.[CO003, CO007, CO010, CO011, CO012, CO013, CO014]

1.4 Exhibits

2.1 Market boundary and substitutes

World Labs' own surfaces define the category as spatial intelligence and world models for virtual and physical worlds, not a generic generative-AI bucket. The included workflows are persistent 3D world creation, editing, export, and simulation across creator, developer, and industrial use cases. The practical substitutes are video-first generation, flat-image tools, and incumbent 3D pipelines that do not deliver editable, explorable worlds from multimodal prompts. That boundary matters because World Labs is trying to win where 3D persistence, camera control, and downstream export are more valuable than raw content novelty.[CM001, CM002, CM003]

2.2 Evidence-constrained sizing lenses

Public evidence supports market lenses, not a clean TAM/SAM/SOM stack. The clearest numeric signals are self-serve packaging, a usage-priced API, and strategic capital from Autodesk and other investors. Those numbers show that World Labs can lower trial friction for creators and developers while attracting large strategic attention from design and physical-world incumbents. They do not show recurring revenue, active customer counts, renewal behavior, or penetration by segment, so any broader market-size claim would be speculative and should remain a diligence gap.[CM006, CM007, CM011, CM012]

2.3 Buyer, user, and payer map

The public packaging already separates at least three buyer paths. Creators and small teams can trial Marble directly via free and paid subscriptions; developers can buy World API credits separately for programmable workflows; and enterprise design, VFX, gaming, digital-twin, or robotics teams are asked to adopt through existing 3D pipelines or early partner access. Export support for game engines, DCC tools, CAD/modeling software, and VR devices lowers switching cost versus closed video outputs, but Atlas itself remains gated, so the broadest enterprise thesis is still ahead of open commercial proof.[CM005, CM006, CM007, CM008, CM009, CM010]

2.4 Growth drivers, constraints, and unresolved sizing gaps

Adoption drivers are strongest where teams already care about 3D persistence, camera control, and interoperability: design and digital twin workflows, game and VFX asset creation, developer tooling, and robotics simulation. Constraints are equally visible. TechCrunch reports skepticism in gaming around IP, energy use, and quality, while independent analysis flags unresolved post-AMD pricing continuity and no public Atlas pricing or named early-access partners. Because public evidence still lacks named production customers, revenue, or independent demand totals, this chapter can support a market thesis but not a defensible market-share model.[CM011, CM012, CM013, CM014, CM015, CM016]

2.5 Exhibits

3.1 Competitive landscape and substitute set

The direct startup peer set in retained evidence is narrower than the broader “AI 3D” label suggests. Runway, Decart, and Odyssey are the most relevant venture-backed comparables because they are all being financed around world-model or adjacent generative-media infrastructure, while Google DeepMind and Nvidia represent incumbent research and platform pressure rather than clean startup comps. Autodesk and the established DCC/game-engine stack matter as status-quo substitutes because many buyers can still solve the same creation or simulation job by staying inside incumbent tools and only adding generative layers selectively. That framing matters because World Labs is not competing only against other startup demos. It is competing against a mature workflow stack that already owns designers, animators, and simulation teams, plus well-funded model builders that can afford to move fast. Public evidence therefore supports a landscape where World Labs is clearly in the top tier of funded spatial-intelligence startups, but not obviously alone in product direction or capitalization, and less proven commercially than the best-disclosed peer in the set.[CP005, CP009, CP010, CP011, CP012, CP013]

3.2 Product scope and buyer tradeoffs

World Labs’ strongest public differentiation is that Marble is already framed as a commercial product rather than just a research preview. Across the official docs, launch blog, and TechCrunch review, Marble is presented as a persistent, downloadable, editable 3D world generator that accepts text, image, video, panorama, and coarse 3D inputs, then exports assets back into downstream creative workflows. That makes it easier to understand where the product fits relative to manual 3D authoring, game engines, or pure research demos: it is positioned as an acceleration layer for world creation, not a full replacement for the rest of the production stack. The tradeoff is that public capability breadth is ahead of public commercial proof. Buyers can see list pricing, visible community/gallery surfaces, and anecdotal use cases, but they still cannot see disclosed customer counts, renewal data, or realized enterprise economics. In other words, World Labs has made product evaluation easier than business-model evaluation. That leaves feature breadth as a real wedge, while pushing the hardest underwriting questions into diligence.[CP001, CP002, CP003, CP004, CP005, CP007, CP008, CP015]

3.3 Distribution leverage, switching cost, and lock-in

The Autodesk relationship is strategically important less because it proves revenue today and more because it validates where World Labs wants to sit in the stack. TechCrunch describes the partnership as exploratory, with media and entertainment first and no data-sharing commitment, which means it is best read as distribution optionality and incumbent validation rather than a locked-in reseller channel. Combined with the official emphasis on mesh export, DCC integration, browser sharing, and VR delivery, the evidence points to World Labs trying to slide into existing pipelines instead of forcing users to abandon them. That lowers adoption friction, but it also constrains moat claims. If customers still finish their work inside Autodesk tools, Blender/Maya/3ds Max, Unity, Unreal, or robotics simulators, then World Labs benefits from incumbent distribution while remaining exposed to incumbent imitation or bundling. Switching costs may rise if teams build repeatable workflows around Marble-generated worlds, but the retained public record still does not show the contract, retention, or workflow-embeddedness evidence needed to call that durable lock-in.[CP006, CP008, CP014, CP016]

3.4 Moat durability verdict

On the current public record, World Labs looks differentiated enough to win attention, partnerships, and experimentation, but not yet differentiated enough to claim an unassailable moat. The company has enormous capital, a strong founding team, and a product surface that is broader than the demo-stage peers cited in retained sources. At the same time, Runway already has the clearest disclosed commercialization in the peer set, and Google DeepMind plus Nvidia can pressure the category from much larger platform positions. The result is a market structure where World Labs may be ahead on product packaging and workflow fit, but still competes in a category that is rapidly attracting adjacent incumbents, direct startup rivals, and internal-build substitutes. The most important disconfirming fact is not technical parity; it is the absence of public revenue, customer-count, renewal, and enterprise-pricing disclosure. Until those metrics are visible, moat durability remains a plausible thesis rather than a demonstrated outcome.[CP009, CP011, CP012, CP013, CP014, CP016]

3.5 Exhibits

4.1 Public monetization surfaces exist, but realized pricing does not

World Labs now has two clearly public commercial surfaces: the Marble application and the World API. Official product material shows Marble as a persistent 3D-world creation workflow rather than a demo viewer, and World Labs' November 2025 launch post moved the product from beta to general availability. Third-party pricing coverage adds useful packaging detail: HokAI says Marble runs from a free plan capped at four world generations through a Max plan capped at 75, with separate enterprise packaging, while the API is sold on a fixed credit schedule. What remains missing is the reader's core underwriting bridge: no retained source discloses actual subscription price points for each Marble tier, realized discounts, revenue mix between app and API, or conversion from free usage into paid revenue.[CI002, CI003, CI008, CI009, CI016]

4.2 Unit economics are mostly inferential because public cost and volume data are thin

The strongest public cost-structure proxy is operational rather than financial. World Labs' own documentation lists generation times from roughly 20 seconds for a draft to around one hour for a high-quality mesh export, implying compute-heavy and likely variable service-delivery cost by workflow. Public traction evidence is also directional, not underwritable: StartupHub says World Labs already has paying customers across gaming, visual effects, robotics training, and architectural design, while HokAI describes self-serve free-plan onboarding for Marble and separate account funding for the API. Even the burn proxy is unstable because public headcount signals conflict, with StartupHub citing 150 employees while HokAI lists a 50-100 team range. That combination supports commercial activity, but not a defendable CAC, payback, or gross-margin model.[CI004, CI010, CI014, CI015, CI016]

4.3 Capital availability looks strong on paper, but runway cannot be verified

Public financing visibility is much better than operating visibility. World Labs' February 2026 funding announcement says the company raised $1 billion from a syndicate that included AMD, Autodesk, Emerson Collective, Fidelity Management & Research Company, NVIDIA, and Sea. Forbes adds that Autodesk contributed $200 million and took a strategic advisory role, while Bloomberg previously reported an approximately $5 billion round valuation that the company declined to confirm. The September 2026 AMD transaction announcement creates a second capital support vector: Ars and HokAI both describe an about-$8.2 billion all-stock acquisition agreement targeted to close by year-end, subject to approval. That backdrop materially lowers the odds of a near-term emergency capital raise, but it still does not reveal cash on hand, monthly burn, working-capital needs, or debt obligations.[CI005, CI006, CI007, CI011, CI012]

4.4 Financial verdict: compelling strategic funding story, insufficient underwriting data

The retained evidence supports a credible revenue architecture and a much stronger balance-sheet narrative than the June 2026 report had, but not a full underwriting case. Marble and the API give World Labs visible monetization rails, and the company appears to have reached at least some paid usage across multiple verticals. However, the most important investor-grade inputs remain absent from the public record: revenue, ARR, gross margin, burn, cash, customer concentration, renewal behavior, and any clear statement of post-close contract continuity. The right financial conclusion is therefore not that the business lacks traction, but that public evidence still cannot separate product promise from durable economics. Treat capital dependency as buffered in the short term and underwriting quality as still diligence-blocked.[CI008, CI009, CI010, CI011, CI012, CI013]

4.5 Exhibits

5.1 Product surface and user workflow

The clearest delivered product is Marble, not Atlas. Across the product site, public docs, and the launch post, World Labs consistently describes Marble as a creator workflow for turning text, images, video, panoramas, and coarse 3D structure into persistent 3D environments that can then be edited, expanded, composed, recorded, and exported. That matters because it anchors the company in a practical workflow—scene creation and manipulation—rather than an abstract “world model” claim. The evidence also shows a split between what is usable today and what is still being staged for later rollout: Marble is broadly exposed through the web product and documentation, while Atlas is presented as early access and “coming to Marble soon,” which suggests World Labs is commercializing with a staged product ladder rather than a single fully generalized platform release.[CE001, CE002, CE003, CE008, CE009, CE014]

5.2 Architecture, operating model, and roadmap linkage

Public technical detail is strongest at the system-description level. Atlas is described as a multimodal autoregressive diffusion transformer that grounds text, images, video, and 3D inputs into a shared spatial context; Marble is the product layer that packages that world-model capability into creation, editing, and export workflows. Together, the sources imply a stack with three layers: multimodal capture and prompting, spatial-context model inference, and downstream rendering/export surfaces such as splats, meshes, and recorded flythroughs. The product roadmap is also explicit that Atlas is intended to power future Marble versions, so current commercial surface area should be treated as a productization wrapper around a still-scaling model family rather than a frozen architecture.[CE004, CE005, CE006, CE007, CE010]

5.3 Maturity, differentiation, and technical risks

The differentiation story is credible but not yet fully de-risked. Independent coverage supports that Marble’s key distinction is persistent, downloadable 3D worlds with AI-native editing and structure control, which is a more operationally useful output than ephemeral 2D world demos. At the same time, independent observers also note that public technical disclosure is thin: there is no public Marble paper, Atlas benchmark discussion is company-authored rather than independently reproduced, and evidence on control, planning, or reliability under production conditions is limited. World Labs documents creator-facing generation times and desktop/mobile constraints, but the assigned source pool does not surface public security, privacy, certification, or uptime materials. That leaves the chapter’s core judgment unchanged: the product is specific enough to believe, but the technical diligence burden remains high before underwriting broad enterprise, robotics, or platform claims.[CE006, CE010, CE011, CE012, CE013, CE015]

5.4 Exhibits

6.1 Customer evidence starts with creators, technical teams, and developers

The public record in this pool points first to creator and technical users rather than classic procurement-led enterprise accounts. World Labs' own Marble Labs hub says it is for "artists, engineers, and designers" and frames public case studies around gaming, VFX, design, robotics, and health systems. The docs reinforce that positioning by organizing the product around Gallery, Create, and Studio, while Technology Magazine says the World API lets developers integrate generation into programmable workflows. Taken together, the visible customer base looks like a mix of self-serve creators, studios, developer teams, and experimental technical users, with the clearest proof concentrated in creative production and developer-adjacent use cases.[CU001, CU003, CU006, CU007, CU008, CU009, CU010]

6.2 Adoption signals are real, but the disclosed metrics remain thin

World Labs has more customer-proof than a stealth research lab, but less than a mature software vendor. The company blog says Marble progressed from an early beta cohort to general availability for anyone, and both Stanford Tech Review and TechCrunch describe a priced commercial product rather than a waitlist-only demo. Docs and the gallery show browsing, sharing, VR viewing, and export paths that make repeat project usage plausible. Even so, the clearest numeric adoption datapoint in the assigned pool is secondhand: an ICMR case study says Marble had 3,000+ users as of December 2025. Public materials do not disclose current active users, paid conversion, API usage, or segment mix, so visible momentum is easier to prove than present scale.[CU002, CU004, CU005, CU006, CU007, CU011]

6.3 Named proof favors creative collaborations, while durability is still unproven

The named references World Labs publishes are useful but should be scored carefully. Marble Labs names Promise, Magnific, OpenArt, HTC VIVE, Escape.ai, WithLore, Rosebud AI, VIVERSE, James C. Kane, and a joint Lightcraft/Beeble example, which is enough to show real external usage conversations and workflow experiments. However, those headlines do not confirm production status, contract value, renewal behavior, or expansion. Independent sources in this pool also stay cautious: ICMR says there were still questions about product-market fit and scaled industry adoption, while the broader external coverage emphasizes packaging, category creation, and early creative use cases more than customer durability. Public evidence therefore supports adoption interest and workflow relevance, but not customer concentration, retention, or enterprise depth.[CU009, CU012, CU013, CU014, CU015]

6.4 Exhibits

7.1 Ranked risk stack

The most material near-term risk is not demand; it is continuity. World Labs has signed a definitive agreement to join AMD for an all-stock transaction that still requires regulatory approval and is only expected to close before the end of 2026. At the same time, an independent profile notes that neither side has publicly said what happens to Marble or World API terms after closing, while the same source says Atlas is early access and intended to power future Marble versions. That creates a live transition risk for anyone underwriting multi-quarter product usage, pricing, or roadmap stability. The second risk cluster is enterprise readiness: public materials reviewed by HokAI say World Labs has not published model cards or a responsible scaling policy, lists no SOC 2 or ISO 27001 certification, and describes training use of user-submitted content with only an opt-out mechanism. The third cluster is operational maturity: Marble's own docs still quote multi-minute generation times, a one-hour high-quality mesh export, and desktop-only dependence for some advanced workflows, while TechCrunch's launch review still described edge morphing and prompt-quality inconsistency.[CR001, CR002, CR004, CR005, CR006, CR007, CR008, CR009, CR012]

7.2 Regulatory, legal, and governance exposure

The legal and governance picture is not one of disclosed litigation or enforcement, but of material public unknowns. The AMD transaction itself cannot be treated as a closed fact until regulators clear it, so diligence has to test closing timing, remedies, and any product or customer-contract implications of the ownership change. Separately, the public privacy-and-security posture leaves unanswered questions for enterprise use cases that could involve proprietary footage, design files, or simulation data. HokAI's review says uploads may be used for model training, benchmarking, and product development unless customers opt out, and it found no public model cards, responsible scaling policy, SOC 2, ISO 27001, or bug bounty. Those are not proof of a hidden defect, but they are real diligence gaps because the business is selling a tool that asks users to upload valuable content into a frontier-model workflow.[CR001, CR004, CR005, CR006]

7.3 Operational execution and dependency risk

Product readiness still looks early relative to the ambition of serving gaming, VFX, design, robotics, and developer workflows at once. World Labs' own materials stress that product feedback is a core part of improving Marble, which is consistent with a platform still learning from usage rather than one already stabilized around a narrow enterprise workflow. The docs also show that the heaviest export step remains slow and some advanced features are still desktop-only, limiting adoption in teams that need predictable turnaround or mobile review paths. TechCrunch's hands-on launch review adds another execution signal: Marble looked strong enough to impress, but it still showed edge morphing and prompt-result inconsistency. Finally, the company is becoming more dependent on AMD as owner, investor, and hardware partner at the same time that outside commentary flags uncertainty on future openness of formats and product terms.[CR002, CR007, CR008, CR009, CR010, CR011, CR012, CR014, CR015]

7.4 Mitigations, monitoring indicators, and kill criteria

The mitigation playbook is straightforward but evidence-heavy. Before treating the risk stack as manageable, investors or customers should require explicit post-close commitments on Marble and World API pricing, support, export rights, and roadmap ownership; evidence of independent security assurance or formal customer controls; and product-level metrics showing that generation quality and turnaround are production-usable rather than demo-usable. The strongest public positive is that Marble already exists as a commercial product with exportable outputs and a broad set of use cases, which means these risks are attached to a real platform instead of a research-only preview. The cleanest thesis-break signals would be a delayed or conditional acquisition that disrupts product continuity, terms changes that reduce portability or customer rights, a meaningful security or privacy incident, or continued inability to show commercial metrics and enterprise controls after the AMD transition narrative is complete.[CR001, CR002, CR003, CR006, CR007, CR009, CR011, CR012, CR013]

7.5 Exhibits

8.1 Public price support improved, but standalone underwriting is still thin

Public financing context is directionally better than it was in the prior report: multiple independent profiles now converge on a February 2026 round of roughly $1 billion and on a reported valuation around $5 billion, while HokAI says the later signed AMD transaction at about $8.2 billion is the only valuation either side has put in writing. That matters because it shows real strategic demand for the asset, not just venture enthusiasm. The underwriting problem is that the February round's exact post-money, preference stack, and dilution consequences remain undisclosed, and Value Add VC explicitly notes that the company never confirmed the ~$5 billion figure in public. For a minority investor, a signed strategic M&A price is also not the same thing as a clean contemporaneous venture entry mark with known rights, timing, and downside protections.[CV001, CV002, CV003, CV010, CV011]

8.2 Product proof is visible, but commercial proof is still the missing bridge

World Labs has crossed an important threshold from research story to shippable product: official materials and TechCrunch show Marble as a public commercial offering with paid tiers, downloadable outputs, multiple input modes, and a separate developer API. That supports the thesis that the company has turned world-model research into something buyers can try today. It does not yet support a high-conviction valuation case. Public sources still do not disclose revenue, customer counts, retention, gross margin, or named scaled deployments, and both HokAI and Silicon Valley Invest Club state that no revenue figure is public. Value Add VC's comparison is the sharpest external benchmark: Runway reportedly reached roughly $300 million ARR before or around a similar valuation band, while World Labs remains valued primarily on team quality, product potential, investor signal, and strategic optionality.[CV004, CV005, CV006, CV007, CV008, CV009, CV012, CV013]

8.3 Recommendation stays track until revenue quality and deal economics are visible

The decision implication is not that World Labs lacks upside; rather, the public record still leaves too much unpriced uncertainty for a buy recommendation at strategic-takeout or late private-market optics. The bull case is that AMD closes near the announced value and the Autodesk relationship plus World API / Marble adoption eventually prove a category-defining infrastructure position. The base case is that the signed deal validates technology and team, but outside investors still cannot underwrite minority-entry returns because core economics and rights remain private. The bear case is that if the AMD close slips or strategic enthusiasm cools, the market re-anchors on the lack of disclosed revenue and customer proof. That keeps the right call at track, with diligence focused on revenue quality, named production customers, February 2026 round terms, and any pricing or contractual changes ahead of transaction close.[CV010, CV011, CV013, CV014, CV015]

8.4 Exhibits