Decart

1.1 Identity and product scope

The clearest current fact pattern is product-centered rather than metrics-centered. On its own website, Decart says it is building "the infrastructure and models that make AI run at the speed of reality," with DOS as the optimization and inference/training layer and Lucy plus Oasis as productized world-model surfaces for live video and physical-AI simulation. The official product pages make the split concrete: Lucy is framed as live video editing for brands, creators, gaming, and commerce, while Oasis is framed as API-accessible, low-latency simulation infrastructure for robotics and autonomous-vehicle workflows. That supports a working view of Decart as a private AI application-infrastructure company monetizing both stack performance and downstream model use cases, not just a single demo product. What remains notably absent from the current public material is the operating baseline that would normally anchor this positioning: no retained source here discloses run-rate revenue, customer count, or headcount, so product ambition is well evidenced while scale is not.[CO001, CO002, CO003, CO004, CO009, CO015]

1.2 Leadership, capital, and strategic stakeholders

Independent May 2026 coverage gives the most usable leadership and capital snapshot. Calcalist identifies Dr. Dean Leitersdorf as CEO and Moshe Shalev as CPO, describing both as Unit 8200 veterans, while AIPressRoom at least corroborates the founder pair even if it contributes less operational detail. The same May 2026 reporting says Decart closed a $300 million round led by Radical Ventures, took total disclosed funding above $450 million, and reached an estimated valuation of roughly $4 billion after a prior August 2025 round at a reported $3.1 billion valuation. Just as important, Calcalist ties the financing to strategic counterparties rather than pure financial sponsorship: Amazon is described as a strategic customer, while NVIDIA is both investor and business partner. That combination matters for later diligence because it suggests Decart's cap table, distribution path, and compute-stack roadmap are already entangled with hyperscaler and chip-platform relationships, yet the retained public sources still do not expose board composition, governance rights, or cap-table control.[CO005, CO006, CO007, CO008]

1.3 Scale signals, milestones, and open questions

The milestone record is directionally strong but unevenly sourced. The official site itself now foregrounds the 2026 financing and associated DOS 2.0, Lucy, and Oasis updates, while Mint adds a current October 2026 signal that Amazon acquisition talks at around $7 billion were reported but remained unconfirmed. Earlier operating history is shakier: AI Chopping Block says the company was registered in September 2023 and had paying customers by December 2023, but that early commercial chronology is only single-source within this pool. There is also a nontrivial identity ambiguity still worth clearing in diligence. Calcalist repeatedly frames Decart as an Israeli company founded in late 2023, whereas Mint describes it as San Francisco-based; that difference could reflect legal domicile versus go-to-market presence, but the current retained record does not let us resolve it. The overarching takeaway is that fundraising momentum and product cadence are easier to verify than the company's present operating scale, governance structure, and exact corporate footprint.[CO009, CO010, CO011, CO012, CO013, CO014]

1.4 Exhibits

2.1 Market boundary and substitutes

Decart's market boundary spans three layers with different budgets and substitutes. DOS addresses model-serving efficiency across hardware, Lucy addresses continuous transformation of live video, and Oasis addresses closed-loop simulation for autonomous vehicles and other physical-AI systems. Included spend is therefore limited to inference optimization, interactive video workflows, and generative simulation that directly use these capabilities. It excludes chip purchases, general cloud capacity, offline post-production, media inventory, robot or vehicle hardware, and broad AI software spend. The practical substitutes are centralized inference, offline video rendering, hand-built simulation environments, and internal optimization work. Combining those pools into one market estimate would conceal materially different adoption clocks and economic buyers.[CM001, CM002, CM003, CM004, CM015]

2.2 Evidence-constrained sizing

No retained source supplies an independently measured, common-unit market-spend estimate for Decart's three wedges. The strongest quantitative signals are instead company-specific and non-additive: American Compute reports 981B open-weight-model tokens served over seven days; Decart says Oasis reached 1M users in 72 hours; and a company-issued release reports sub-35 millisecond performance in a Comcast and NVIDIA edge benchmark. TechTimes separately reports annual revenue in the several tens of millions of dollars, with uneven growth driven largely by custom chip-maker engagements, but that report is neither audited segment revenue nor ARR. These indicators establish activity, technical feasibility, and some commercialization; they do not establish category TAM, addressable spend, market share, or repeatable contract value.[CM007, CM009, CM011, CM013, CM014, CM016]

2.3 Buyers, users, payers, and adoption path

Buyer identity follows the product layer. Infrastructure and model-serving teams evaluate DOS against throughput, latency, utilization, and migration effort. Media platforms, retailers, brands, and creator businesses evaluate Lucy against engagement or monetization workflows that must operate while video is live. Autonomous-vehicle and robotics teams evaluate Oasis against the cost, safety, and coverage limits of real-world data collection and fixed simulation. Telecom and edge-infrastructure operators are enabling partners where geographic proximity is required to keep response times interactive. The adoption path therefore runs from technical benchmark or API trial, through workflow integration, to a production contract owned by infrastructure, product, marketing, commerce, or autonomy leaders; public evidence does not disclose conversion rates between those stages.[CM002, CM003, CM004, CM005, CM007, CM008]

2.4 Growth drivers, adoption constraints, and valuation relevance

The principal driver is that low-latency generation creates workflows that offline rendering cannot serve: live editing, responsive worlds, and closed-loop robot training. The principal constraint is proof of durable economics. ToolDirectory notes a fidelity trade-off versus batch renderers, while TechTimes reports uneven growth tied largely to custom engagements; both observations raise questions about product depth, repeatability, and sales concentration. Edge delivery can also add infrastructure dependencies before global deployment. For valuation, technical performance and launch engagement justify treating the category as real, but a revenue multiple cannot be grounded in public market-share data until Decart discloses segment revenue, paying-customer counts, retention, contract duration, gross margin, and customer concentration by DOS, Lucy, and Oasis.[CM006, CM007, CM008, CM013, CM016]

2.5 Exhibits

3.1 Landscape and substitute sets

The assigned evidence does not support a single "AI video" peer group for Decart. Its public surface spans DOS infrastructure, Lucy live video editing, and Oasis interactive world generation, so the buyer's substitute set changes with the workflow. For creator and brand use cases, the retained comparison source frames Runway Gen-4, Google Veo 3, and OpenAI Sora 2 as the relevant alternatives. For robotics and physical-AI teams, Decart itself frames CARLA, Isaac Sim, Genie 3, and Nvidia Cosmos as the more relevant substitutes. That split matters because Decart appears to be competing first on interactivity and deployment shape, not on the same polished offline output the batch video leaders optimize for.[CP001, CP003, CP004, CP005, CP006, CP015]

3.2 Capability and packaging differentiation

The clearest differentiation in the retained evidence is that Decart is selling immediacy. Lucy is described as sub-40ms live editing at up to 100 FPS for enterprises, brands, and creators, while Oasis is described as an API-accessible, action-conditioned world model for physical AI with under 200ms end-to-end latency. The independent comparison article reinforces the same positioning by placing Decart's MirageLSD and Lucy Edit in real-time or near-real-time lanes while labeling Runway, Veo, and Sora as minutes-per-clip tools and Cosmos as offline simulation. Packaging also matters: the same source says Mirage is hosted and free to try, Lucy Edit ships open development weights, and several rivals remain proprietary hosted tools in the assigned evidence.[CP002, CP003, CP006, CP007, CP011]

3.3 Distribution, lock-in, and incumbent response

Decart's moat is not only model behavior; it is also the DOS stack beneath those models. Third-party reporting describes DOS as hardware-aware and usable across Nvidia GPUs, Amazon Trainium, and Google TPUs, which lowers chip-vendor lock-in relative to vendor-bound optimization layers and lets Decart sell infrastructure into existing customer estates. Calcalist and AIPressRoom also say Amazon is already a strategic customer while Nvidia is both investor and business partner. That combination can accelerate enterprise trust and distribution, but it also means the most powerful incumbents can respond through bundled cloud, chip, or frontier-lab relationships instead of only feature-by-feature product copying.[CP008, CP009, CP010, CP014]

3.4 Durability limits and diligence gaps

The assigned evidence supports a real latency wedge, but it does not prove Decart wins every mode of video or simulation demand. The comparison source explicitly leaves polished narrative shots to Runway, Veo, and Sora, which suggests the company is running a different race rather than a universal quality race. Public pricing and contract structure are also thin: the sources show a free demo, open weights, API access, and "Talk to an Expert" motions, but not list prices, discounting, or customer spend. Finally, acquisition reporting introduces ecosystem risk because an Anthropic deal could tighten access to consumer endpoints or repurpose the team into infrastructure work, which would change how third parties rely on Decart's current product surface.[CP011, CP012, CP013]

3.5 Exhibits

4.1 Multi-line monetization thesis is visible, but realized pricing is not

Decart's public materials consistently describe three commercial surfaces: DOS as the optimization and infrastructure layer, Lucy as the real-time video product, and Oasis as the physical-AI world-model product. That mix matters because it implies the company is not trying to monetize a single consumer app; it is trying to sell infrastructure and higher-level model experiences into enterprise workflows. The product pages and launch coverage point to API access, enterprise deployments, commercial agreements, and expert-led outreach rather than self-serve checkout, which is directionally consistent with sales-led software and infrastructure revenue. What remains missing is the actual commercial contract shape. None of the retained public sources publish list pricing, realized pricing, committed usage minimums, or contract durations for DOS, Lucy, or Oasis, so the monetization thesis is intelligible but the price realization layer still requires private diligence.[CI001, CI002, CI003, CI004, CI011, CI013]

4.2 Public traction points show interest and partner access, not sales efficiency

The strongest public traction signals are proxy metrics rather than financial outputs. Decart says Oasis reached 1M users in 72 hours, Lucy is already deployed with enterprise customers across multiple industries, and AWS has both commercial agreements and joint go-to-market work around Trainium-based deployment. Those data points help establish demand formation, channel access, and product relevance across gaming, live media, robotics, and infrastructure buyers. They do not, however, convert cleanly into underwriting metrics. A large user spike for Oasis does not prove paid production adoption, named enterprise deployments do not reveal ACV or retention, and hyperscaler partnerships do not disclose whether economics accrue as software revenue, services revenue, or strategic co-selling support. The result is a GTM picture that looks enterprise-led and partnership-assisted, but still lacks the CAC, payback, cycle-length, and expansion data needed to judge sales efficiency.[CI007, CI008, CI009, CI010, CI011, CI014]

4.3 Fresh capital reduces immediate pressure, but core underwriting data remains private

Financially, the clearest new fact is the May 2026 financing: Decart publicly announced a $300M round and multiple outlets place total disclosed funding above $450M. That scale of financing likely reduces near-term pressure to raise immediately and gives the company time to continue building DOS, Lucy, and Oasis. Even so, the public valuation anchor is imprecise, with coverage ranging from roughly $3.1B to nearly $4B, and TNW explicitly reported that Decart did not disclose run-rate revenue, the new valuation, or the hyperscaler customers licensing DOS. More importantly, the retained public evidence never closes the basics of underwriting: there is no disclosed revenue by product line, no gross margin, no CAC or payback, no burn, no cash balance, and no runway. The chapter verdict is therefore that Decart may have rare strategic momentum and strong financing support, but revenue quality, margin path, and capital efficiency remain diligence blockers rather than established facts.[CI005, CI006, CI012, CI014, CI015]

4.4 Exhibits

5.1 Product stack and delivered workflows

Decart's public product surface has consolidated into a layered stack rather than a collection of research demos. The company says DOS is the performance layer, with Lucy and Oasis as the exposed model families on top. That matters because the public value proposition splits into two customer workflows: Lucy targets live video transformation for enterprises, brands, and creators, while Oasis 3 targets physical-AI teams that need interactive simulation rather than offline rendering. The product story is differentiated by real-time operation, but the evidence quality differs by surface: Oasis includes API and SDK detail that makes the workflow inspectable, whereas Lucy is still described more through marketing and media coverage than through public technical integration material in this chapter's evidence set.[CE001, CE002, CE008, CE009, CE012]

5.2 Architecture, APIs, and operational model

Oasis 3 is the clearest window into how Decart's stack actually works. The docs describe a hosted, stateful gRPC session where a user prompts a world, streams four-action control chunks, and receives generated camera frames back per stream, which is much closer to an operational simulator than to a one-shot media model. The main integration dependencies are Decart-hosted endpoints, API-key authentication, and the decart-oasis SDK, while W&B describes the internal development workflow as experiment- and artifact-driven with lineage and access controls. Together, those surfaces suggest real engineering depth and reproducibility effort, but they also reinforce how dependent the buyer experience is on Decart's hosted infrastructure and undocumented uptime, security, and support processes.[CE002, CE003, CE004, CE005, CE006, CE007, CE010]

5.3 Maturity signals, dependencies, and technical risk

The maturity signal is strongest where independent or partner surfaces overlap with the official story. World-models.io independently describes OASIS as a real-time neural game engine, W&B frames Oasis 3 as production-grade tooling for robotics partners, and Forbes reports Lucy 2 running live at 1080p and 30 frames per second while being tested in live creator and Omniverse-adjacent workflows. Even so, the evidence remains uneven. One assigned third-party WebRTC page resolves only to a newsletter-signup stub, and the assigned pool does not provide a public status page, third-party security certification, or explicit privacy/compliance documentation for Lucy or Oasis. That leaves Decart with a credible real-time architecture story, but a thinner public proof set around reliability, controls, and independent deployment depth than the headline performance claims imply.[CE006, CE007, CE011, CE013, CE014]

5.4 Exhibits

6.1 Customer mix spans creators, enterprises, and physical-AI teams

Decart's public go-to-market is bifurcated between immersive-video use cases and physical-AI simulation. On the Lucy side, the company positions the product for enterprises, brands, creators, and viewers across retail and e-commerce, advertising, social media, live streaming, and gaming. On the Oasis side, Decart frames the initial buyer set as robotics teams, AV companies, and research groups that need closed-loop simulated environments rather than consumer entertainment alone. Calcalist adds a higher-value infrastructure layer on top of those verticals by reporting contracts with cloud providers, AI labs, and hyperscalers. The result is a customer map with multiple user surfaces, but the largest disclosed economic value likely sits in a still-thin set of platform relationships rather than in disclosed long-tail self-serve accounts.[CU002, CU003, CU006, CU007, CU014]

6.2 Adoption proof is real, but it mixes consumer reach, benchmarks, and unnamed contracts

The strongest adoption evidence is directionally positive but heterogeneous. Decart's about page says Oasis reached 1 million users in 72 hours, while an ACCESS Newswire release says the company's models power millions of users. Those signals matter because they move Decart beyond lab-demo status, but they are not equivalent to paid enterprise retention. Named deployment proof is clearest around technical validation: Comcast benchmarked Lucy on distributed NVIDIA-powered edge infrastructure at sub-35 millisecond latency, and AIPressRoom says Lucy is already deployed in virtual try-on, live streaming, and in-video advertising. Calcalist separately reports revenue-generating contracts with several cloud providers, AI labs, and hyperscale companies. Taken together, the evidence supports real market pull, but the public record still blends user scale, benchmark scale, and contracted scale into one narrative.[CU001, CU004, CU006, CU007, CU009, CU012]

6.3 Named customer proof is concentrated in strategic-platform references

The clearest named commercial proof in the assigned pool clusters around large platform counterparties rather than a diversified list of end customers. Calcalist says Amazon joined as a strategic customer and describes AWS collaboration for media, commerce, advertising, and physical-AI workloads. ACCESS Newswire ties Decart to Comcast and NVIDIA in a benchmarked edge-inference setting. That is useful proof that sophisticated partners are willing to work with Decart, but it is still more design-partner and infrastructure-validation evidence than a mature roster of disclosed production accounts with contract values, deployment breadth, or renewal history. The public chapter, therefore, supports platform relevance but not broad customer-base diversification.[CU005, CU006, CU010, CU013]

6.4 Durability, expansion, and concentration remain the main diligence gap

The core customer risk is not whether Decart has any adoption; it is whether the company can convert a handful of strategic references into repeatable, diversified, and durable revenue. Venture Capital Tracker's caution that a multibillion-dollar valuation still requires enterprise or platform distribution rather than demo reels captures the right burden of proof. Within the assigned evidence set, none of the retained sources disclose customer count, NRR, GRR, churn, renewal rates, contract length, or top-customer mix. That leaves investors unable to distinguish pilot enthusiasm from durable production usage, or to quantify how dependent Decart may be on a few hyperscalers, cloud providers, or flagship partnership channels for its current revenue trajectory.[CU008, CU010, CU011, CU013]

6.5 Exhibits

7.1 The top risks are valuation whiplash, product sprawl, and proof gaps

Public evidence supports a high-level risk ranking even without a disclosed lawsuit or enforcement action. First, Decart is being discussed in rapidly moving strategic-transaction narratives. One independent source frames Anthropic at around $6 billion, while Mint cites Amazon talks at around $7 billion and explicitly says those talks could still fall apart. Second, the company is trying to commercialize a full stack at once—DOS infrastructure, Lucy for live video, and Oasis for physical-AI simulation—which expands execution surface area before public revenue, retention, or margin disclosure catches up. Third, the loudest operating proof in the current source set is product-speed and usage marketing, not durable customer economics or audited financials.[CR001, CR006, CR007, CR008, CR009, CR010, CR013]

7.2 Physical-AI and safety exposure is credible, but public compliance detail is thin

The official Oasis materials make the regulatory and safety question unavoidable because Decart is not positioning the product as a toy benchmark; it says Oasis is for physical AI and is starting with autonomous vehicles. That creates real diligence burden around simulation validity, model misuse, customer safety workflows, and any sector-specific compliance overlays. Yet the assigned public evidence does not confirm licenses, certifications, disclosed incident history, or a litigation record tied to those use cases. That does not prove a hidden problem, but it does mean investors should treat regulatory, legal, and safety diligence as an open gate rather than as de-risked simply because the product is technically impressive.[CR002, CR011, CR012, CR014, CR016]

7.3 Operational scaling risk is amplified by ambitious latency claims and sparse customer proof

Decart's official pages market extreme responsiveness—sub-40ms Lucy edits and Oasis response under 200ms—across video, simulation, and physical-AI workflows. Those promises are strategically attractive, but they also raise the bar for uptime, model quality, deterministic behavior, and supportability in enterprise deployments. The same public set still leaves customer proof thin: the About page highlights Oasis reaching 1M users in 72 hours, and Lucy references partners and customers plus broad verticals, but the assigned evidence does not identify named paying customers, renewals, concentration, or conversion from viral usage into recurring revenue. That combination makes commercialization and service reliability a top underwriting risk.[CR002, CR003, CR010, CR011, CR013]

7.4 Partner leverage and financing opacity could magnify downside quickly

The partner map is promising but not yet obviously diversified enough to neutralize dependency risk. Independent reporting says DOS runs across Nvidia GPUs, Google TPUs, and Amazon Trainium, and Mint separately notes AWS gave Decart early access to Trainium3. That is helpful for platform flexibility, but it also means supplier relationships and privileged compute access are central to the product story. Meanwhile, AIPressRoom says the company did not disclose current valuation or run-rate revenue even while raising another $300 million, and Venture Capital Tracker warns that multibillion-dollar valuations require enterprise or platform distribution rather than demo-driven excitement. If transaction rumors fade or financing expectations reset before operating proof catches up, the downside could compress quickly.[CR004, CR005, CR006, CR008, CR009, CR012, CR015, CR016]

7.5 Exhibits

8.1 Recommendation and price discipline

The price-sensitive call remains track with medium confidence, high risk, and a stretched valuation stance. Public evidence shows real strategic validation: Calcalist reports a $300 million round at an estimated $4 billion valuation with Amazon as a strategic customer and NVIDIA as both investor and partner, while The Wall Street Journal reports Amazon discussed an acquisition around $7 billion. But those anchors still bracket a wide range rather than underwriting a precise entry price. The reviewed sources do not disclose audited revenue, gross margin, retention, or the preference stack, so the market can observe technical proof and partner interest without seeing the commercial quality that should justify paying above the last mark.[CV004, CV005, CV006, CV007, CV009, CV012, CV013, CV014]

8.2 Upside proof versus underwriting gaps

The upside case is not imaginary. Decart presents a coherent stack of DOS plus Lucy and Oasis, with Lucy positioned for live video editing and Oasis 3 aimed first at autonomous-vehicle simulation and then broader physical-AI use cases. That technical narrative has attracted heavyweight validation: Calcalist says Amazon joined as a strategic customer and NVIDIA invested, while the company says it is already generating revenue through contracts with large cloud providers, AI laboratories, and hyperscale companies. The underwriting problem is that the public package stops short of the numbers an investor needs. AIPressRoom explicitly notes that Decart did not disclose current valuation or run-rate revenue, and the reviewed public sources still do not disclose gross margin, retention, customer concentration, board control, or liquidation preferences.[CV001, CV002, CV003, CV005, CV006, CV008, CV009, CV013]

8.3 Scenario envelope and diligence triggers

The public record supports an outcome envelope, not a clean fair-value point estimate. The best hard price anchor remains the reported $4 billion May 2026 financing. The best upside signal is the rumored ~$7 billion Amazon process, which could prove strategic scarcity but is not equivalent to a signed transaction or a standalone public comparable set. Meanwhile, lower-quality directories still surface stale or partial funding snapshots, which is another sign that the external record is thin outside a few key reports. That makes diligence on current revenue quality, customer concentration, preference terms, and exit process discipline the final gating work: if those items clear, the range can re-rate upward; if they do not, the downside can fall back toward or below the last private mark.[CV004, CV007, CV010, CV011, CV012, CV013, CV014]

8.4 Exhibits