Latest Public Valuation Anchor 01
4000 USD M [CV004] Diligence report
Megvii Technology
Large historical funding base and real commercial breadth, but sanctions, privacy regulation, and opaque current financials dominate the risk-reward.
Megvii has real technical depth and commercial breadth, but sanctions, biometric-regulation risk, and opaque current financials keep the investment case below threshold for a confident long position.
Cover facts
Company profile
Megvii Technology is a Beijing-founded private AI company that evolved from facial-recognition roots into a broader AIoT stack spanning Brain++, Face++, smart-building software, smart-city management, warehouse automation, and identity products. Public evidence supports substantial historical capital formation and real commercial deployments, but also shows that sanctions, surveillance controversy, and limited current disclosure sharply constrain investability and valuation confidence.
- Website
- megvii.com/en
- Founded
- 2011-10-01
- Founders
- Yin Qi, Tang Wenbin, Yang Mu
- Founding location
- Beijing, China
- Headquarters
- Beijing, China
- Product
- Megvii sells a vertically integrated vision-AI stack: Brain++ for AI productivity, Face++ and related identity APIs, Pangu for access / building workflows, edge hardware such as recognition terminals and cameras, and vertical solutions for smart city, smart building, smart warehouse, and retail environments.
- Customers
- Government and public-space operators, logistics and warehouse users, property and campus operators, telecom-linked ecosystems, retail deployments, and developers or integrators using identity and computer-vision APIs.
- Business model
- Hybrid AIoT model blending software, APIs, hardware-attached deployments, and project or integration revenue across city, building, warehouse, and identity workflows.
- Stage
- Late-Stage Private
- Funding status
- About $1.98B raised across 10 rounds according to public profiles. Last publicly visible financing event is an undisclosed April 2025 Series D; the strongest prior valuation anchor is the 2019 $750M round at slightly above $4B.
Executive summary
Top strengths
- Substantial historical capital raised and continued evidence of domestic financing support
- Broad product stack spanning frameworks, APIs, software, devices, and physical-world AI workflows
- Real commercial proof points across telecom, retail, building access, warehouse, and international smart-access deployments
- Computer-vision research pedigree and internal infrastructure depth via Brain++ and MegEngine-related assets
Top risks
- U.S. Entity List status and surveillance-related human-rights controversy create durable trust, supply-chain, and exit constraints
- Current revenue, margin, burn, and 2025 round economics remain publicly undisclosed, preventing precise valuation work
- Biometric and facial-recognition regulation is tightening in China and Europe, directly affecting core product categories
- Failed Hong Kong and Shanghai listing paths weaken liquidity visibility and increase the private-market discount
- Customer concentration, renewals, and the share of revenue tied to the most policy-sensitive deployments remain opaque
Open gaps
- Audited 2025 or 2026 revenue, gross margin, and cash-flow statements
- Size, valuation, and terms of the April 2025 financing round
- Customer concentration by sector and top-account renewal behavior
- Current litigation, fines, regulator inquiries, and remediation programs related to biometric deployments
- Realistic exit path under continuing sanctions and surveillance stigma
Contents
01Company Overview
1.1 Identity, product scope, and current stage
Megvii Technology sits at the intersection of computer vision, AI infrastructure, and physical-world deployment. The company was founded in Beijing in October 2011 by Yin Qi, Tang Wenbin, and Yang Mu, and public profiles consistently describe the founders as Tsinghua-linked engineers who built the company around large-scale visual-recognition systems. The company is still private as of the 2026 run date, with no completed Hong Kong or Shanghai listing despite multiple filing attempts. On the product side, Megvii still organizes its business around three core verticals — Personal IoT, City IoT, and Supply Chain IoT — and says it commercializes those verticals through a full-stack stack of algorithms, software, hardware, and AI-enabled IoT devices. Face++ remains the best-known front door for outside developers and identity workflows, while Brain++ is described as the internal and commercial AI productivity layer that underpins training, deployment, and operation. Megvii's current-stage story is therefore not that of a pure API vendor or a pure surveillance contractor; it is a late-stage private AIoT company trying to convert computer-vision excellence into multiple physical-world revenue lines while carrying a still-unfinished capital-markets process.[CO001, CO002, CO003, CO004, CO005, CO006]
| Metric | Value/Status | Date | Confidence | Gap/Note |
|---|---|---|---|---|
| Founded | October 2011 in Beijing | 2011 | high | Founder, company, and profile sources align. |
| Stage | Private / late-stage Series D | 2026 | high | No completed IPO found in reviewed sources. |
| Flagship products | Face++ and Brain++ | 2026 | high | Face++ is the external platform; Brain++ is the internal/commercial AI stack. |
| Core verticals | Personal IoT / City IoT / Supply Chain IoT | 2026 | high | Repeated across official pages and Brain++ launch material. |
| Total raised | $1.98B across 10 rounds | 2025-2026 | high | Supported by Tracxn and The Company Check; earlier narrative sources cite lower totals. |
| Last known valuation | Slightly above $4B | 2019 | high | Best-supported public valuation marker remains the 2019 round. |
| Latest disclosed round | Undisclosed Series D with Ant Group, Legend Holdings, and Chongqing Industrial Investment Fund | 2025-04-08 | medium | Comes from Tracxn, not a company press release. |
| Current headcount disclosure | No precise 2026 figure; 2020 press release cited 2,300+ employees | 2020-2026 | medium | Current headcount remains an evidence gap. |
Combines current identity, funding, and stage facts; where 2026 precision is unavailable the gap is stated explicitly.
[CO001, CO004, CO005, CO006, CO008, CO023]Funding, sanctions, and platform-commercialization milestones explain Megvii's path from 2011 founding to a still-private 2025 financing.
Year-only dates are used where the strongest public source did not provide a full day or month.
[CO001, CO002, CO005, CO013, CO014, CO015]Megvii's research core, platforms, and vertical solutions connect directly to both customer proofs and the sanctions overhang.
The figure is a logic map rather than a process diagram; it summarizes relationships described explicitly in the cited sources.
[CO006, CO007, CO008, CO009, CO018, CO023]1.2 Research strengths, founder dependence, and responsible-AI posture
Megvii's public self-description emphasizes research depth and founder-led execution more than formal public-company governance. The company says it operates the world's largest computer-vision research institute and has won 49 international AI competition titles since 2017, including a third consecutive COCO challenge win in 2019. Those claims help explain why the company could build both a consumer-facing developer platform and a deeper internal AI infrastructure layer. Brain++ and MegEngine matter because they indicate Megvii is not merely packaging third-party models; it has invested in its own compute, data, and framework stack. Governance quality, however, is harder to verify from public sources. Megvii does provide some comfort signals: it says it published an AI Ethics Code in 2019 and set up both an AI Ethics Committee and an AI Ethics Research Institute. Even so, the public evidence is much stronger on technical leadership than on current board composition, independent oversight, or 2026 leadership changes. The result is a founder-centric company with a credible research engine and a partially articulated ethics framework, but still an incomplete public governance record by late-stage investor standards.[CO010, CO011, CO012, CO013, CO014, CO015]
| Person / body | Role | Background or remit | Why it matters | Dependency / gap |
|---|---|---|---|---|
| Yin Qi | Co-founder and CEO | Tsinghua-linked founding team; public face of company strategy and fundraising | Central to capital formation, external narrative, and product strategy | High key-person dependence in public narrative. |
| Tang Wenbin | Co-founder | Tsinghua-linked founding team member associated with technical leadership | Anchors technical credibility inside the founding group | Current public remit is less fully disclosed than Yin Qi's. |
| Yang Mu | Co-founder | Third named founder in company and profile sources | Completes founding-control picture and early technical origin story | Lower public visibility than Yin Qi. |
| AI Ethics Committee / Institute | Internal governance bodies | Megvii says both bodies guide responsible AI and its six ethics areas | They are the clearest formal governance artifacts visible publicly | Current board roster and independence details are still not public. |
Enumerates the founders plus the only clearly public governance body pair visible in reviewed sources; the current board roster remains undisclosed.
[CO002, CO003, CO015, CO016, CO042]Key public company-overview metrics show a late-stage, still-private AIoT company with strong funding but a persistent disclosure gap.
The figure mixes hard numbers and categorical status because public disclosure is uneven; unavailable current metrics are stated qualitatively.
[CO001, CO004, CO022, CO023, CO024, CO025]1.3 Funding history, valuation markers, and commercial proof points
Megvii's financing history shows why it remains strategically relevant even after sanctions pressure. In 2019 the company closed a $750 million Series D, with public coverage naming BOCGI, an ADIA subsidiary, Macquarie, ICBC Asset Management (Global), and Alibaba among the syndicate; Reuters said the round valued the company at slightly above $4 billion. More recent database-style sources now converge around $1.98 billion of cumulative funding across 10 rounds, while Tracxn also records an additional undisclosed Series D on April 8, 2025 with Ant Group, Legend Holdings, and Chongqing Industrial Investment Fund participating. Commercial proof is also visible in the company's own releases: Koala reached 191 Beijing supermarkets including Chaoshifa and Wumart, then expanded to Thailand, Brazil, and the UAE; Megvii also cites a smart-access deployment in a 280-meter Singapore mixed-use landmark and intelligent-venue work at major Beijing Winter Olympics sites. These proof points do not substitute for audited revenue or margin disclosure, but they do show the company still has deployment relevance across retail, enterprise access, public venues, telecom ecosystems, and supply-chain workflows.[CO018, CO019, CO020, CO021, CO022, CO023]
| Stakeholder | Role | Economic or strategic importance | Diligence ask |
|---|---|---|---|
| Alibaba | Existing strategic investor | Participated in later funding and is tied to Face++ commercial usage context | Clarify current ownership and commercial revenue dependence. |
| BOCGI | 2019 lead investor | Reuters said it led the 2019 round with $200M | Confirm whether it retains board or preference rights. |
| ADIA subsidiary / Macquarie / ICBC AMG | 2019 round participants | Signal institutional willingness to fund Megvii before Entity List effects fully landed | Map current economics and any step-up expectations. |
| Ant Group / Legend Holdings / Chongqing Industrial Investment Fund | Named 2025 Series D participants | Best visible sign that state-linked and strategic domestic capital still backs the company | Request terms, valuation, and any liquidation preferences. |
| Citi / Goldman Sachs / JPMorgan | Banking and capital-markets relationships | Named in funding and IPO reporting around 2019 | Clarify whether any formal listing mandates remain live. |
| China Telecom | Commercial ecosystem partner | Shows distribution relevance across digital-life and smart-community scenarios | Separate real revenue contribution from strategic signaling. |
Maps the highest-salience capital providers and ecosystem counterparties visible in public sources rather than the full cap table.
[CO019, CO020, CO021, CO025, CO026, CO037]1.4 Entity List exposure and the still-unfinished IPO path
The central adverse fact in Megvii's company overview is that its capital-markets story has been shaped by sanctions and human-rights scrutiny since 2019. The Federal Register notice that took effect on October 9, 2019 added 28 China-based entities to the U.S. Entity List, and contemporaneous Reuters reporting carried by CNBC and Yahoo Finance identified Megvii as one of the affected AI companies. Human Rights Watch's reporting on Xinjiang surveillance explains why this designation continues to matter: the company sits inside a broader facial-recognition and public-security ecosystem that Western investors and regulators treat as politically sensitive. Reuters also reported that HKEX regulators asked Megvii additional questions during its November 2019 listing hearing rather than approving the deal, leaving a targeted $500 million to $1 billion Hong Kong IPO in limbo. Later profile sources continue to describe Megvii's Hong Kong and Shanghai listing pushes as stalled rather than completed. Taken together, the evidence supports a simple company-overview conclusion: Megvii built a real, heavily funded AIoT business, but one whose public-exit path and international investability remain meaningfully constrained by sanctions, surveillance stigma, and incomplete disclosure.[CO027, CO028, CO029, CO030, CO031, CO032]
| Date | Event | Type | Amount / status | Participants | Implication |
|---|---|---|---|---|---|
| 2011-10 | Megvii founded in Beijing | founding | Company formation | Yin Qi, Tang Wenbin, Yang Mu | Establishes origin point for the company and founding team. |
| 2012 | Face++ launched as cloud visual platform | product | Commercial launch | Megvii | Created the external platform that still anchors the brand. |
| 2019-05-08 | Series D financing closed | financing | $750M; valuation slightly above $4B | BOCGI, ADIA subsidiary, Macquarie, ICBC AMG, Alibaba and others | Marked the last clearly public valuation step-up before sanctions. |
| 2019-07-08 | AI Ethics Code published | governance | Policy launch | Megvii | Public attempt to articulate responsible-AI posture. |
| 2019-08-29 | Named national AI open innovation platform for image perception | partnership | State recognition | Ministry of Science and Technology / Megvii | Strengthened government-aligned positioning in core CV research. |
| 2019-10-09 | Added to U.S. Entity List | regulatory | Effective restriction | U.S. Department of Commerce | Raised technology-access and IPO-execution risk materially. |
| 2019-11-22 | HKEX IPO hearing setback | adverse | $500M-$1B IPO delayed | HKEX Listing Committee / underwriting banks / Megvii | Public listing path stalled after blacklist pressure. |
| 2020-03-25 | MegEngine open sourced | product | Framework release | Megvii / global developers | Showed willingness to externalize core AI infrastructure. |
| 2020-09-21 | Commercial Brain++ launched | product | Commercial release | Megvii enterprise customers | Expanded beyond internal tooling into enterprise AI enablement. |
| 2025-04-08 | Latest Tracxn-listed Series D | financing | Undisclosed amount | Ant Group, Legend Holdings, Chongqing Industrial Investment Fund | Signals continued domestic capital support despite stalled IPO path. |
Single chronology of record for founding, product, financing, governance, regulatory, and adverse milestones used by later chapters.
[CO001, CO005, CO018, CO020, CO022, CO015]02Market Analysis
2.1 Market boundary and adjacencies
Megvii does not participate in a single clean software market. Its own official pages show it selling into three overlapping layers: city-management systems, logistics and warehouse automation, and developer or enterprise identity workflows built on Face++ and Brain++. That means the relevant market boundary is broader than stand-alone facial recognition but narrower than all AI. A useful diligence definition is AI-enabled computer vision for physical-world workflows, where models, edge hardware, deployment services, and operations software are bundled into one buying motion. On that definition, Megvii sits at the intersection of urban AI, security and identity, and supply-chain automation. Official competitor sites reinforce that framing: SenseTime also spans infrastructure plus application layers, YITU still foregrounds smart city AI, and incumbents such as Hikvision, Dahua, Axis, and Hanwha train buyers to expect integrated hardware-software solutions instead of pure SaaS. Megvii's market is therefore best analyzed as a layered AIoT and computer-vision stack rather than as an isolated facial-recognition API niche.[CM020, CM021, CM022, CM023, CM024, CM025]
| Segment / category | Included spend | Excluded spend | Buyer / payer | Relevance to Megvii |
|---|---|---|---|---|
| Core computer vision | Models, software, edge hardware, deployment, analytics | General-purpose AI unrelated to visual workflows | Enterprise ops, municipalities, security buyers | Primary TAM anchor for Megvii. |
| Facial recognition | Identity verification, access control, search, liveness detection | Non-visual authentication methods | Risk, product, security, public-safety teams | Important but too narrow as the only lens. |
| Smart city / urban AI | Traffic, safety, epidemic prevention, city operations | Generic e-government software without CV | Municipal IT and public-security budgets | Matches Megvii's City IoT positioning. |
| Warehouse / logistics AI | Scheduling, safety, forecasting, robotics workflows | Generic ERP or non-vision warehouse software | Logistics and supply-chain leaders | Matches Megvii's Supply Chain IoT positioning. |
| Developer / identity APIs | Face++ and embedded identity modules | Broad consumer software unrelated to CV | Digital-channel, device, and fraud teams | Explains Megvii's developer distribution option. |
Defines the spend layers that matter for Megvii; excludes generic AI that lacks a visual, edge, or physical-world workflow component.
[CM020, CM021, CM022, CM023, CM029, CM030]2.2 Sizing lenses: broad TAM, narrower SAM, and the biometric subsegment
The public market data support a large but highly inconsistent opportunity envelope. IMARC places global computer vision at $21.7 billion in 2025, while Fortune Business Insights puts 2025 at $20.75 billion and Mordor at $27.39 billion. Verified Market Research sits still lower on a 2024 base, underscoring how sensitive the estimates are to market definition and included spend. Growth rates are even more dispersed: IMARC shows a mid-single-digit CAGR, while Fortune, Mordor, and MarketsandMarkets imply mid-teens to low-twenties growth depending on whether the lens is all computer vision or AI-enhanced computer vision. Facial recognition is clearly only a subset of that universe, with TBRC and Mordor placing it around $7.9 billion to $8.6 billion in 2025. That gap matters for Megvii. If investors underwrite the company only as a facial-recognition vendor, the addressable market looks meaningfully smaller; if they underwrite Megvii as a city-plus-logistics-plus-identity platform, the ceiling is much larger. Because sanctions and surveillance stigma narrow Megvii's practical reach, the best diligence posture is to use a range-based TAM and an even more conservative sanctions-adjusted SAM rather than a single heroic market number.[CM001, CM002, CM003, CM004, CM005, CM006]
| Publisher / lens | Year | Geography | Value | Growth | Methodology or caveat | Confidence |
|---|---|---|---|---|---|---|
| IMARC computer vision | 2025 | Global | USD 21.7B | 5.6% CAGR to 2034 | Broad CV lens; includes major verticals and regions | medium |
| Fortune computer vision | 2025 | Global | USD 20.75B | 14.8% CAGR to 2034 | Broader growth framing than IMARC | medium |
| Mordor computer vision | 2025 | Global | USD 27.39B | 15.77% CAGR to 2031 | Higher current base, emphasizes edge and hardware | medium |
| MarketsandMarkets AI in CV | 2025 | Global | USD 23.42B | 22.1% CAGR to 2030 | Focuses on AI-enhanced CV rather than all CV | medium |
| TBRC facial recognition | 2025 | Global | USD 7.88B | 17.5% CAGR to 2030 | Subset market relevant to identity and surveillance | medium |
| Mordor facial recognition | 2025 | Global | USD 8.58B | 15.97% CAGR to 2031 | Subset market with strong privacy-law discussion | medium |
| Sanctions-adjusted Megvii SAM | 2026 | China + politically neutral export markets | Not publicly disclosed | n/a | Requires judgment because Entity List and stigma remove part of the theoretical TAM | low |
Public estimates disagree materially; use these as bookends rather than a single precise TAM. The final row is an analytical caution, not a measured market size.
[CM001, CM002, CM003, CM005, CM007, CM009]Megvii's opportunity should be read as a stack of progressively narrower lenses rather than a single all-AI TAM.
This is a lens stack, not a strict accounting cascade; the underlying sources measure overlapping but non-identical layers of spend.
[CM001, CM005, CM009, CM012, CM014, CM016]Public sources disagree on current size and growth, so range-based underwriting is more honest than a single point estimate.
Midpoints are arithmetic centers used for display legibility; the decision-useful fact is the breadth of the range, not the midpoint.
[CM001, CM005, CM012, CM014, CM016, CM018]2.3 Buyer, user, and payer segmentation
The most useful segmentation for Megvii is by operational workflow rather than by raw industry code. Municipal and public-sector buyers purchase for traffic, public safety, and city-management use cases; enterprise security and facilities teams buy for access, smart-building, and property workflows; logistics operators buy for warehouse efficiency and safety; consumer-device or digital-service teams buy Face++ and identity modules for onboarding, fraud control, and device features. The economic buyer differs across each segment. Municipal IT offices and public-security budgets matter in city projects, while security chiefs, facility managers, and property operators matter in building access. In warehouses the operational buyer is closer to logistics or supply-chain leadership, and in digital identity it is often a product or risk leader. This segmentation also helps explain why Megvii competes against different archetypes in each segment: AI peers in city intelligence, hardware incumbents in physical security, and developer-tool competitors in identity. It also clarifies why public evidence on procurement cycles is still incomplete: Megvii's selling motion is fragmented across multiple buyer classes rather than concentrated in one repeatable SaaS motion.[CM023, CM024, CM025, CM026, CM027, CM028]
| Segment | Buyer | User | Payer | Workflow | Adoption trigger |
|---|---|---|---|---|---|
| Municipal / smart city | Municipal IT, public-safety leads | Traffic and operations staff | City budgets | Traffic optimization, urban governance, epidemic response | Safety, congestion, digital-governance goals |
| Enterprise buildings | Facilities and security managers | Employees, visitors, guards | Property or enterprise budgets | Access control, attendance, visitor management | Throughput, fraud reduction, lower staffing |
| Logistics / warehouse | Operations and supply-chain leaders | Warehouse supervisors, line workers | Operations capex / opex | Scheduling, monitoring, automated decision support | Labor efficiency, safety, forecast accuracy |
| Fintech / identity | Risk or product leaders | End users onboarding or authenticating | Digital product budgets | Liveness, verification, fraud control | KYC efficiency and fraud loss reduction |
| Device makers / OEMs | Product managers, device OEMs | End-device consumers | Device or embedded-software budgets | Face unlock, photography, embedded AI features | Differentiated user experience and security |
Buyer and payer roles vary by workflow, which is why Megvii's GTM is fragmented rather than a single repeatable seat-based software motion.
[CM023, CM029, CM030, CM031, CM032]Megvii faces different buying centers and adoption triggers across each visual-workflow segment.
This matrix uses evidence-backed ordinal scoring to summarize buyer conditions rather than claiming precise numeric survey results.
[CM023, CM030, CM031, CM032, CM033]Physical-world AI deployments narrow sharply from broad interest to scaled, compliant production use.
Funnel values are indexed adoption-density scores, not market-share or customer-count disclosures.
[CM029, CM030, CM031, CM032, CM033]2.4 Growth drivers, adoption constraints, and sanctions-adjusted implications
Across the reviewed sources, five growth drivers recur: automation demand, edge-compute improvements, security and fraud-reduction needs, smart-city digitization, and the expansion of AI into logistics and quality-control workflows. But the same sources show why Megvii cannot capture the full headline upside. Mordor explicitly links future growth in facial recognition and computer vision to stricter consent rules and data-sovereignty design choices, while the Entity List and Xinjiang-linked criticism add a Megvii-specific penalty on top of generic regulatory friction. For a China-based company already on the U.S. Entity List, growth is not constrained only by technology or customer ROI; it is also constrained by procurement optics, cross-border component access, and limits on Western institutional capital. That makes Megvii's practical market opportunity less about total computer-vision spend and more about how much domestic and politically neutral demand can be won in smart city, enterprise access, and logistics use cases. The key diligence implication is straightforward: use public market reports as broad ceiling indicators, but anchor underwriting on narrower buyer segments where Megvii still has visible product fit and less direct geopolitical exclusion.[CM015, CM017, CM019, CM032, CM033, CM034]
| Driver / constraint | Direction | Timing | Implication | Diligence ask |
|---|---|---|---|---|
| Edge compute and hardware acceleration | Positive | Current | Makes CV deployment cheaper and lower latency | How much of Megvii's current stack is edge-optimized vs. cloud-bound? |
| Automation and labor efficiency demand | Positive | Current | Supports logistics, quality, and access-control ROI cases | Which verticals show the fastest payback in Megvii deployments? |
| Smart-city digitalization | Positive | Current but cyclical | Maintains demand for urban-governance solutions in China | How dependent is Megvii on public-sector procurement cycles? |
| Biometric consent and privacy rules | Negative | Current / increasing | Raises deployment friction and pushes privacy-preserving designs | Does Megvii publish enough compliance controls for export markets? |
| U.S. Entity List and export controls | Negative | Persistent | Shrink the effective serviceable market and tech-sourcing flexibility | What markets and suppliers are now out of reach? |
| Surveillance stigma | Negative | Persistent | Reduces Western investor and buyer willingness | Can logistics and enterprise workflows dilute the stigma over time? |
Pairs structural demand drivers with Megvii-specific constraints so the market chapter does not confuse headline TAM with practical capture potential.
[CM015, CM017, CM019, CM032, CM033, CM034]03Competitors
3.1 Landscape: direct peers, incumbents, and specialist substitutes
Megvii competes in a fragmented field that cannot be described by one peer list. The closest China-native platform peers are SenseTime, YITU, and CloudWalk, all of which overlap on vision AI, public-sector workloads, and enterprise deployments. Hardware-heavy incumbents such as Hikvision and Dahua compete from the opposite direction: they start with installed cameras, security channels, and integrated surveillance infrastructure, then layer AI onto that footprint. Global enterprise-trust incumbents such as Axis and Hanwha Vision define buyer expectations outside China around channel quality, long operating history, and trusted physical-security distribution. Clearview AI sits in still another bucket as a narrower facial-recognition specialist built around law-enforcement identification rather than a broad AIoT operating stack. Tracxn and The Company Check both reinforce the point that Megvii faces a very large and diverse peer set, not a neat single market. Diligence should therefore classify competitors by archetype before comparing them on product or economics.[CP001, CP003, CP004, CP005, CP006, CP007]
| Competitor | Category | Scale / status | Target segment | Differentiation | Limitation |
|---|---|---|---|---|---|
| SenseTime | China-native AI platform | Large AI software company; public-market visibility via external sources | Generative AI, vision AI, infrastructure | Platform breadth plus infrastructure ownership | Commercial execution and China AI pressure remain concerns. |
| YITU | China-native AI platform | Private AI company with smart-city and healthcare positioning | Public-sector and healthcare AI | Strong city and healthcare framing | Less visible global distribution than hardware incumbents. |
| CloudWalk | China-native AI platform | Public Chinese facial-recognition company | Government and enterprise vision AI | Direct China peer on public-sector AI | Public disclosure still thinner than large global incumbents. |
| Hikvision | Surveillance incumbent | Global security and video-solutions incumbent | Physical security and surveillance buyers | Installed-base and channel power | Hardware-led posture can be less flexible than software-native AI stacks. |
| Dahua | Surveillance incumbent | Video-centric AIoT incumbent | Security and smart-IoT buyers | Integrated hardware + software distribution | Faces same surveillance-category commoditization risk as Hikvision. |
| Axis | Global trusted vision vendor | ~5,000 employees; $2.1B 2025 sales | Enterprise and network-video buyers | Trust, reliability, and global channel depth | Less associated with frontier AI infrastructure narratives. |
| Hanwha Vision | Global trusted vision vendor | Large smart-visual-intelligence vendor | Industry and security buyers | Global industrial channel and visual-intelligence portfolio | Less visible software-developer surface than Face++. |
| Clearview AI | Specialist substitute | Facial-recognition specialist | Law enforcement and public safety | Focused product for identification workflows | Far narrower product scope than Megvii or SenseTime. |
Profiles competitors by archetype so buyers are not forced into a false one-dimensional peer list.
[CP001, CP003, CP004, CP005, CP006, CP007]Megvii sits between AI-platform breadth and trust-constrained distribution: stronger than specialists on breadth, weaker than incumbents on trust and channel.
Quadrant scores are evidence-backed ordinal judgments synthesizing public positioning, not measured market-share or benchmark data.
[CP015, CP016, CP017, CP018, CP021, CP026]3.2 Capabilities, distribution, and where each competitor wins
Capability comparison breaks along two axes: platform breadth and channel reach. SenseTime is the clearest China-native platform comparator because it explicitly combines infrastructure with multiple application categories, echoing Megvii's Brain++ plus solution-stack logic. YITU is more focused in the public-sector and healthcare narratives visible in reviewed sources. Hikvision and Dahua remain formidable not because they necessarily outperform Megvii on core model research, but because they sell into a hardware footprint and procurement channel that buyers already trust. Axis and Hanwha similarly illustrate how much global enterprise buyers value distribution, reliability, and long-lived device ecosystems. Face++ is one of Megvii's distinctive counters to that structure because it gives the company a developer and identity surface that pure camera incumbents do not obviously match. Clearview, meanwhile, reminds investors that a narrower specialist can still be dangerous in one niche even if it lacks Megvii's broader AIoT ambition. The competitive battle is therefore not accuracy alone; it is breadth, trust, bundling, and route-to-market.[CP002, CP006, CP007, CP008, CP009, CP016]
| Buying criterion | Megvii | SenseTime | Hikvision / Dahua | Axis / Hanwha | Clearview |
|---|---|---|---|---|---|
| AI infrastructure ownership | Strong (Brain++) | Strong (SenseCore) | Medium | Low to medium | Low |
| Developer / API surface | Strong (Face++) | Medium | Low | Low | Low |
| Installed hardware channel | Medium | Medium | Strong | Strong | Low |
| Law-enforcement-specific facial search | Medium | Medium | Medium | Low | Strong |
| Global trust / export posture | Low | Low to medium | Medium | Strong | Medium |
| Software-defined city / enterprise AI breadth | Strong | Strong | Medium | Medium | Low |
Ordinal matrix synthesized from product and company descriptions; it summarizes relative public positioning rather than proprietary benchmark tests.
[CP016, CP017, CP018, CP019, CP021, CP026]Platform peers lead on AI-stack breadth, hardware incumbents lead on channel, and specialists lead only on narrow facial-search depth.
Matrix uses ordinal scoring derived from public positioning rather than a proprietary benchmark.
[CP016, CP017, CP018, CP019, CP021, CP026]3.3 Pricing opacity, switching cost, and multi-homing dynamics
Direct public pricing comparison is weak across almost every relevant competitor. The reviewed sources largely market solutions rather than publish list prices, which is consistent with a market where deployment scope, hardware mix, accuracy thresholds, and compliance needs vary sharply by customer. That opacity matters because it means competitive outcomes often depend more on project design and installed-base economics than on simple price-per-seat comparisons. Once cameras, access control, software rules, visitor flows, warehouse processes, or identity workflows are integrated into a site, switching cost rises meaningfully. Hikvision, Dahua, Axis, and Hanwha benefit from this dynamic through installed devices and channel relationships; Megvii and SenseTime benefit when their platform software and model layers become hard to replace without redesigning the workflow. Multi-homing remains possible at the evaluation stage, but it becomes harder after scaled deployment. For investors, the missing piece is hard proof on win rates, ACVs, and renewals, which public evidence does not disclose cleanly today.[CP022, CP023, CP024, CP025, CP033, CP034]
| Vendor archetype | Price / contract model | Included capabilities | Unknowns | Implication |
|---|---|---|---|---|
| Megvii / SenseTime | Project or solution bundle | Models, software, integration, sometimes hardware | No public apples-to-apples list pricing | Competitive outcomes likely negotiated and scope-specific. |
| Hikvision / Dahua | Hardware-led bundle plus software | Cameras, VMS, analytics, integration | Deal-level discounts and software attach unknown | Installed base can hide true software economics. |
| Axis / Hanwha | Device and solution bundle | Network-video devices, software, services | End-customer package pricing opaque | Global trust may justify premium pricing. |
| Clearview / niche specialists | Subscription or specialist software bundle | Facial search and investigative workflow | Public price detail limited | Specialists can undercut or outfocus broad platforms in narrow niches. |
Public sources reveal positioning but not comparable list prices; packaging opacity is itself a competitive fact.
[CP022, CP023, CP024, CP025, CP034]Megvii's competitive readiness is strongest on stack breadth and weakest on trust-sensitive international distribution.
Scores are analytical summaries based on the evidence reviewed in this chapter; they are not third-party ratings.
[CP018, CP022, CP023, CP024, CP026, CP031]3.4 Moat durability, commoditization risk, and Megvii-specific weakness
Megvii's moat is strongest when it is sold as a full-stack AIoT and vision-infrastructure platform rather than as a stand-alone facial-recognition feature. That is the strategic logic behind comparing it more closely with SenseTime than with pure camera vendors. But the moat is also under pressure from two directions. First, facial recognition and access control can commoditize when buyers view them as one feature inside a broader bundle sold by hardware incumbents. Second, Megvii carries a trust and regulatory overhang that many overseas alternatives do not: the U.S. Entity List, Xinjiang-linked scrutiny, and a stalled IPO path all weaken the company's positioning in sensitive export markets. Those issues do not erase Megvii's technical depth, but they raise the threshold for winning on trust-sensitive deals and make global expansion harder than for Axis, Hanwha, or other non-sanctioned vendors. Competitive diligence should therefore underwrite Megvii as a technically credible but trust-constrained player whose best relative edge remains software-defined, integrated physical-world AI workloads.[CP026, CP027, CP028, CP030, CP031, CP032]
| Moat claim | Threat | Severity | Mitigation / diligence ask |
|---|---|---|---|
| Full-stack AIoT platform | Facial recognition commoditizes into a feature bundle | high | Underwrite Megvii on integrated workflows, not a single biometric feature. |
| Face++ developer channel | Large hardware incumbents bypass developer ecosystems with installed-base reach | medium | Test whether developer distribution converts into durable enterprise revenue. |
| China-native AI leadership | Trust-sensitive export markets discount sanctioned vendors | high | Map which geographies remain realistically open to Megvii. |
| Platform infrastructure ownership | SenseTime and other AI peers match the same stack logic | medium | Differentiate on domain-specific deployments and cost/performance proof. |
| Installed public-sector footprint | Policy backlash or procurement scrutiny reduces future bids | high | Review backlog mix, concentration, and overseas pipeline quality. |
Summarizes how Megvii's moat can erode under both commoditization and trust pressure.
[CP026, CP027, CP030, CP031, CP032, CP033]04Financials
4.1 Revenue model and monetization surfaces
Megvii's public record supports a hybrid monetization model rather than a clean single-product SaaS story. Face++ indicates one software or API-like surface for identity and developer workflows. Device Authentication and related verification products imply another monetization surface around liveness, verification, and fraud-related workflows. Pangu and its open API positioning show the company also sells software layers for access, attendance, visitor management, and regional security operations. At the same time, the Smart City Management and Smart Warehouse solution pages imply project, integration, and operations revenue that depends on deployment scope, hardware mix, and customer workflow redesign. China Telecom and supermarket case studies add evidence that Megvii also monetizes through channel and deployment partnerships. The Smart Identity Verification Device page further suggests dedicated hardware sales attached to identity workflows. This mix is strategically helpful because it widens the addressable revenue pool, but it also means Megvii should be analyzed less like a pure recurring software vendor and more like an AIoT company whose revenue quality differs materially by product line.[CI015, CI016, CI017, CI018, CI019, CI020]
| Stream | Mechanism | Unit | Current value / status | Quality | Diligence ask |
|---|---|---|---|---|---|
| Face++ / identity APIs | Developer or enterprise CV / identity usage | API / software | Visible product surface, no disclosed revenue | Potentially software-like but undisclosed economics | Request API revenue share, major customers, and pricing ladder. |
| Pangu access management | Software plus integration for building workflows | Software / deployment | Visible commercial software surface | Likely mix of software and services | Request software license vs implementation split. |
| Identity hardware devices | Dedicated verification terminals and devices | Hardware / bundled software | Visible hardware surface, no disclosed attach economics | Likely lower-margin than pure software, but useful for solution control | Request hardware GM and software attach rate. |
| Smart city solutions | Project deployment and operations support | Project / solution | Visible official positioning, no disclosed bookings | Potentially large but project-heavy | Request ACV, backlog, and payment-collection profile. |
| Warehouse / logistics solutions | Automation workflow deployment | Project / solution | Visible official positioning, no disclosed revenue | Could be lumpy and implementation-heavy | Request number of live sites and gross-margin profile. |
| Channel partnerships | Partner-led solutions such as China Telecom | Partner revenue / enablement | Visible partnership proof, no revenue detail | Useful channel, unknown direct economics | Request booked revenue and rev-share terms. |
Summarizes the public revenue surfaces visible in product, solution, and partnership pages; no stream has disclosed revenue contribution in the reviewed sources.
[CI015, CI017, CI018, CI019, CI020, CI023]| Price / contract | List vs realized pricing | Included capabilities | Unknowns | Source |
|---|---|---|---|---|
| API / software usage | No public list visible in reviewed sources | Face++, identity, or software layer | Unit pricing, discounting, and volume tiers unknown | Official product pages show surface, not price. |
| Project bundle | Likely negotiated per site / scope | Hardware, deployment, training, support | Gross margin and payment schedule unknown | Smart city and warehouse solutions. |
| Partner-led package | Likely negotiated with channel partner | Telecom / community / video capability stack | Revenue share and attach rate unknown | China Telecom collaboration. |
| Commercial Brain++ enablement | Enterprise platform package | Algorithm lifecycle, compute, deployment support | License basis and recurring mix unknown | Brain++ commercial launch. |
Public evidence shows packages and capabilities, not apples-to-apples list pricing. That opacity is itself a financial risk for diligence.
[CI016, CI021, CI022, CI023, CI024, CI026]Megvii monetizes multiple layers, from developer-facing APIs to project-heavy city and warehouse deployments.
This is a commercial logic map rather than a disclosed accounting bridge.
[CI015, CI016, CI017, CI018, CI019, CI020]4.2 Cost structure and unit-economics visibility
The reviewed sources give much better visibility into product architecture than into financial efficiency. Brain++ is financially relevant because Megvii says the commercial platform can reduce algorithm-production cost by 55% and shorten development time by 80%, suggesting a real internal-efficiency layer rather than pure marketing language. Even so, none of the public sources reviewed here disclose the unit economics investors would normally want for a late-stage private company: no ARR, gross margin, burn, CAC, payback, NRR, or churn metrics appear in the accessible public evidence. Product mix also suggests margin heterogeneity. API or software-led identity products should have structurally different gross-margin behavior from smart-city and warehouse projects that require deployment services, hardware, or customization. The result is a chapter where the business model is visible, but the economic quality of each layer is not. That is a meaningful analytical limitation because Megvii could look attractive on topline deployment activity while still being difficult to scale efficiently.[CI011, CI012, CI021, CI022, CI024, CI025]
| Metric | Value / null | Confidence | Why it matters | Diligence ask |
|---|---|---|---|---|
| Revenue | null | low | No accessible public revenue figure means no topline baseline. | Obtain audited 2025 and trailing-12-month revenue. |
| ARR | null | low | Necessary to separate recurring software from project revenue. | Request recurring revenue by product line. |
| Gross margin | null | low | Required to judge whether API and project layers are economically attractive. | Request gross margin by software, hardware, and services mix. |
| Burn / cash use | null | low | Needed to translate funding history into runway. | Request monthly burn and opex structure. |
| CAC / payback | null | low | Needed to judge go-to-market efficiency. | Request cohort economics or sales-efficiency proxy. |
| Brain++ productivity impact | 80% faster dev / 55% lower algorithm-production cost | medium | Only public efficiency datapoint in reviewed sources. | Validate whether the claim is internal-only or visible in financial outcomes. |
Most unit-economics cells are genuinely unavailable in public sources; Brain++ productivity claims are the only direct efficiency datapoints reviewed.
[CI011, CI012, CI013, CI014, CI021, CI039]Public evidence reveals one efficiency signal — Brain++ productivity — but leaves the rest of the economic bridge blank.
Most links are qualitative because public financial statements are not available.
[CI021, CI022, CI024, CI025, CI036, CI039]Only funding and valuation have defensible public ranges; operating metrics remain mostly undisclosed.
The third row intentionally encodes the absence of a current public valuation rather than inventing one.
[CI001, CI002, CI003, CI006, CI008]4.3 Capital adequacy and financing dependence
Megvii has clearly raised enough capital to build a substantial company, but public evidence is still weak on whether it has enough capital to reach self-sustaining economics. Tracxn and The Company Check converge on $1.98 billion raised across 10 rounds, with an undisclosed Series D recorded in April 2025 and a best-supported last public valuation marker slightly above $4 billion from the 2019 round. The company's own 2019 release says the cash was intended for deep-learning technology, commercialization, talent, and global expansion, which is consistent with a capital-intensive AIoT buildout. The 2021 STAR Market filing also shows that Megvii continued to seek public-market access after the Hong Kong process stalled. The problem is that cash, burn, runway, debt, and working-capital obligations remain opaque. The delayed Hong Kong IPO, later STAR path, and Entity List restrictions matter financially because they reduce financing flexibility even if domestic investors are still supportive. Viewed conservatively, the 2025 round proves capital access is not closed, but it does not prove capital adequacy.[CI001, CI002, CI003, CI004, CI005, CI006]
| Item | Status | Why it matters | Public evidence | Diligence ask |
|---|---|---|---|---|
| Total capital raised | Visible | Shows ability to fund a large AIoT buildout | $1.98B across 10 rounds in profile databases | Reconcile exact cap-table and round chronology. |
| Latest financing | Partially visible | Signals continued domestic investor support | Undisclosed April 2025 Series D in Tracxn | Request amount, valuation, and liquidation terms. |
| Cash on hand | Not visible | Core runway metric | No public disclosure reviewed | Request current cash balance. |
| Monthly burn | Not visible | Determines financing urgency | No public disclosure reviewed | Request burn and opex composition. |
| Debt / project finance | Not visible | Could materially change risk profile | No public disclosure reviewed | Request debt schedule and covenant summary. |
| IPO / liquidity path | Delayed / redirected | Affects financing flexibility | 2019 HKEX setback followed by 2021 STAR filing | Request current listing plan and banker mandates. |
Capital raised is visible, but capital adequacy is not; the table separates what public data shows from what remains essential but missing.
[CI001, CI002, CI003, CI004, CI030, CI031]Megvii's capital story is clear on funding volume and unclear on the operating metrics needed to judge adequacy.
Ordinal scoring summarizes what the chapter can and cannot see in public financial evidence.
[CI023, CI024, CI025, CI030, CI031, CI032]4.4 Financial verdict: visible commercialization, opaque economics
The clearest conclusion from public sources is not that Megvii lacks commercial activity, but that its economic quality is under-disclosed. Official and partner releases show real deployments in retail, telecom-linked digital-life channels, smart-city settings, and warehouse or building workflows. Database-style profiles also show that the company has continued to attract capital. But the most basic investor questions remain unanswered in public: current revenue, profit, gross margin, burn, runway, and financing terms are all absent or incomplete. That forces a cautious verdict. Megvii looks financially more like a well-funded but still opaque AIoT integrator than a transparently scalable software company. In practical diligence terms, that means even bullish product evidence should be haircut until management opens the books on segment mix, cash, and margin trajectory. Until audited revenue-quality and capital-adequacy metrics are available, any underwriting case should assume higher execution and financing risk than the product footprint alone would suggest.[CI009, CI010, CI013, CI014, CI029, CI035]
| Missing private metric | Impact | Exact diligence path |
|---|---|---|
| Audited 2025 revenue and gross profit | Prevents any reliable valuation multiple or margin bridge | Obtain audited statements or draft prospectus financials. |
| Cash, burn, and runway | Prevents judgment on financing urgency | Request board or investor materials covering liquidity. |
| Product-line revenue mix | Prevents separation of software economics from project economics | Request bookings and revenue by major business line. |
| Customer concentration and payment terms | Prevents working-capital and collection-risk analysis | Request AR aging, top-customer exposure, and contract terms. |
| Debt and off-balance-sheet obligations | Prevents full capital-structure analysis | Request debt schedule, guarantees, and project liabilities. |
These gaps are the minimum financial asks needed before any confident investment view.
[CI009, CI010, CI011, CI012, CI013, CI014]05Product & Technology
5.1 Core platform: Brain++, MegEngine, and research depth
Megvii’s technical identity is broader than facial recognition alone. At the core of the stack is Brain++, which the company describes as a proprietary AI productivity platform spanning algorithm production, model training, deployment, and support workflow. The commercial Brain++ launch matters because it is framed not just as infrastructure but as a measurable efficiency layer, with claims of 80% faster algorithm development and 55% lower algorithm-production cost. Megvii also ties its face-recognition technology directly to the MegEngine deep-learning framework, showing that the company owns a meaningful part of its model-development substrate. Public GitHub repositories for MegEngine and MegFlow strengthen that point by demonstrating that parts of the stack are open to outside inspection and not purely black-box marketing. Academic papers such as the 2015 LFW work and RepVGG further support the view that Megvii has genuine computer-vision research pedigree, and the CVF publication page confirms that RepVGG reached a major conference venue. The technical takeaway is that Megvii is best understood as a vertically integrated vision-AI stack with both internal infrastructure and outward-facing research outputs.[CE001, CE002, CE003, CE010, CE011, CE012]
| Layer | Evidence | Why it matters | Public limit |
|---|---|---|---|
| Brain++ | Proprietary AI productivity platform | Shows internal tooling and AI lifecycle ownership | Performance and adoption claims are company-led. |
| MegEngine | Open-source deep learning framework | Shows framework ownership below the application layer | Public repo does not prove commercial deployment scale. |
| MegFlow | Open-source ML workflow project | Suggests orchestration tooling around long-tailed demands | Commercial usage is not publicly quantified. |
| Research outputs | LFW paper, RepVGG paper | Shows external research pedigree | Current benchmark leadership remains unproven here. |
Summarizes the evidence for Megvii’s underlying technical substrate.
[CE001, CE010, CE012, CE013, CE014, CE015]Megvii’s public technical story runs from core frameworks through tooling into products and vertical solutions.
The figure organizes the company’s disclosed stack; it is not a direct architecture diagram from Megvii.
[CE001, CE010, CE012, CE015, CE016, CE018]5.2 Product stack: from APIs and SDKs to hardware and edge deployment
Megvii’s public product surfaces show a company trying to control multiple layers of the computer-vision stack. Face++ and the enterprise FaceID assets show clear developer and identity-product distribution, while the Face++ China site suggests broad API and SDK reach. Pangu adds a higher software layer for attendance, access, visitors, and regional security operations, and the open API positioning implies Megvii wants third parties to embed its algorithms into their own systems. Hardware pages then extend the same stack to devices and the edge: access-control terminals support large on-device recognition libraries, identity devices advertise fast verification, and network cameras tie cloud-edge-device algorithms directly to sensor products. This architecture matters strategically because it lets Megvii monetize the same underlying models through APIs, enterprise software, and purpose-built hardware. It also makes the company more than a pure model vendor, but less asset-light than one. External developer platforms such as Papers With Code, Hugging Face, and Replicate also show at least some of this stack leaking into broader developer discovery channels. The product story is therefore one of stack control rather than one flagship application.[CE004, CE005, CE006, CE007, CE008, CE012]
| Surface | Public evidence | Customer / partner role | Tech implication |
|---|---|---|---|
| Face++ | Web API and SDK platform | Developer and enterprise integration | Extends distribution beyond direct solution sales. |
| Enterprise FaceID | Dedicated enterprise asset | Identity and verification workflows | Shows packaged productization of core CV. |
| Pangu | Access / attendance / visitor software | Building and campus operators | Shows workflow software above the model layer. |
| Pangu Open API | Embedded SDK / API positioning | Partner product builders | Shows algorithm embedding into third-party stacks. |
These surfaces show Megvii is not just shipping devices; it also exposes reusable software interfaces.
[CE004, CE005, CE016, CE017, CE040, CE041]| Device / modality | Public claim | Role in stack | Diligence question |
|---|---|---|---|
| Access-control terminal | Offline recognition and 100k-entry library | Edge identification and access control | How often does hardware pull through software revenue? |
| Identity verification device | >99% accuracy, <400ms verification | Fast identity checks at edge | Are these claims independently benchmarked? |
| Smart network camera | Cloud-edge-device deep learning | Video analytics and sensing | What is third-party accuracy in live deployments? |
The hardware layer matters because it can anchor Megvii inside physical workflows, not just cloud inference.
[CE006, CE007, CE008, CE034, CE037]Megvii spans APIs, software, hardware, and embedded partner surfaces instead of one narrow product form.
Ordinal judgments synthesize public positioning rather than internal revenue mix.
[CE004, CE005, CE006, CE008, CE016, CE017]5.3 Verticalization: how Megvii translates core vision tech into workflows
Megvii does not present its technology as one general-purpose API alone; it repeatedly translates the core stack into workflow-specific solutions. The reviewed pages cover smart campuses, marketing interaction, building access, temperature measurement, SMB attendance, warehouse automation, and urban governance. That breadth indicates the company’s technical advantage is not just perception accuracy, but packaging — combining models, software logic, devices, and operational flows into repeatable deployment blueprints. Case studies reinforce that claim. Singapore access infrastructure, China Telecom ecosystem cooperation, supermarket rollouts, and property-management partnerships all suggest that Megvii can operationalize the tech beyond lab benchmarks. For diligence, this is important because it shows the product/tech chapter has real commercialization bridges. It also suggests the strongest lock-in appears where Megvii’s devices, identity layers, and operations software are installed together, while the heaviest implementation burden appears in warehouse, city, and large-building environments.[CE018, CE019, CE020, CE021, CE022, CE023]
| Vertical | Public product page / case | Tech packaging | Why it matters |
|---|---|---|---|
| Campus / building | FaceID, Smart Park, SMB attendance | Identity, access, workflow logic, devices | Shows repeatable AIoT packaging. |
| Retail / health screening | Supermarket deployment | Recognition plus temperature / access workflow | Shows rapid productization under real operational conditions. |
| Warehouse | Smart warehouse solution | Perception plus process automation | Shows industrial operations relevance. |
| Smart city | Urban governance solution | Vision plus public-space workflow orchestration | Shows public-sector-grade systems integration. |
| Telecom ecosystem | China Telecom cooperation | Partner-led service embedding | Shows stack portability into another platform. |
Maps the company’s verticalization strategy from one technical core into multiple operational settings.
[CE018, CE020, CE022, CE023, CE024, CE026]Public evidence shows Megvii extending core vision technology into multiple vertical deployments over time.
The timeline mixes research and commercialization milestones to show translation from core CV into workflows.
[CE003, CE023, CE024, CE026, CE027, CE029]5.4 Technical verdict: broad stack, real research, incomplete independent validation
The strongest public argument for Megvii’s product and technical position is breadth. The company appears to own meaningful infrastructure, expose external developer surfaces, ship edge hardware, and wrap computer-vision capabilities into multiple vertical solution packages. Open-source repositories and academic papers make the technology story more credible than a normal marketing-only vendor narrative. External hubs such as Papers With Code, Hugging Face, Replicate, and ModelScope add another useful signal: Megvii-related assets are visible in ecosystems where developers actually discover and run models. But public evidence also has a clear limit: much of the most favorable performance framing still comes from Megvii itself. There is not enough current independent evidence in the reviewed source set to settle questions about present benchmark leadership, patent depth, inference economics, or product-line deployment scale. That means the technical case should be treated as credible but not fully independently audited. Investors can be confident Megvii is a real stack builder; they should be less confident about exactly how defensible each layer is today without deeper customer, benchmark, and engineering diligence.[CE029, CE030, CE031, CE032, CE033, CE036]
| Missing external proof | Why it matters | Next diligence step |
|---|---|---|
| Current benchmark leadership | Technical moat claims require third-party comparison | Run live bake-offs against peer systems. |
| Patent depth and ownership map | Needed to judge IP defensibility | Collect patent-family list and assign to product lines. |
| Inference economics by product line | Needed to judge cost advantage at scale | Request deployment-level compute and hardware cost data. |
| Commercial adoption by product | Needed to separate flagship products from edge cases | Request live deployments, ARR, and customer counts by product. |
| Independent accuracy audits | Needed to validate fast/accurate marketing claims | Obtain customer validation reports or partner test data. |
These are the main reasons the chapter stops short of calling Megvii’s tech stack fully verified.
[CE003, CE007, CE030, CE036, CE037, CE042]Research and open-source evidence are reasonably credible; performance and deployment economics remain less independently validated.
The confidence map rates what the reviewed sources can actually support, not what Megvii may know internally.
[CE003, CE012, CE015, CE029, CE030, CE031]06Customers
6.1 Customer footprint: what is visible publicly
Megvii’s public customer evidence is meaningful but uneven. The company clearly operates across multiple buyer types: government and city operators, logistics and warehouse users, building and campus operators, telecom-linked ecosystems, retail sites, and developer or integration channels exposed through Face++. Named public references include the Singapore smart-access deployment, China Telecom cooperation, a Beijing supermarket rollout, and Jinyu property-management cooperation. These references matter because they prove the company is not just selling abstract AI models; it is landing real physical-world deployments. At the same time, the evidence is not organized like a conventional SaaS customer page with logos, case studies, and quantifiable ROI by account. Much of what is visible is deployment proof or partner proof rather than investor-grade customer analytics. The right reading is therefore that Megvii has real customer traction across several sectors, but public evidence only partially reveals who the most economically important customers actually are. Even the developer-facing layer is visible more through platform surfaces than through named public accounts.[CU001, CU002, CU003, CU004, CU005, CU012]
| Reference | Type | What it proves | Limit |
|---|---|---|---|
| Singapore development | Customer proof | International smart-access deployment | No contract value disclosed. |
| China Telecom | Partner proof | Large channel / ecosystem distribution | Partner cooperation does not disclose end-customer revenue. |
| Koala supermarkets | Customer proof | Multi-site retail deployment | Pandemic use case may not generalize. |
| Jinyu property management | Partner proof | Property-tech / building distribution | Revenue scale undisclosed. |
Named references establish commercial traction but provide little investor-grade revenue detail.
[CU002, CU003, CU004, CU005, CU017, CU036]| Evidence type | Strength | Weakness | What investors still need |
|---|---|---|---|
| Company deployment release | Proves activity and product fit | Often omits economics and customer ROI | ACV, term length, renewal, and scope. |
| Partner release | Shows channel access and ecosystem trust | Can overstate commercial depth | Booked revenue and attach rates. |
| Solution page | Shows target customer segment | Does not prove live adoption | Named live accounts and volume. |
| Third-party profile | Adds sector framing | Often generic or lagged | Primary customer disclosures. |
Separates the existence of evidence from the quality of that evidence.
[CU017, CU018, CU031, CU036]Public evidence points to several distinct customer layers, from large institutional buyers to developers and integrators.
Layer sizes are ordinal, not customer counts.
[CU001, CU006, CU007, CU010, CU013]6.2 Where the customers come from: verticals, channels, and buyer archetypes
The reviewed source set suggests Megvii’s customer motion is strongly verticalized. Smart city pages point toward government and public-space operators; warehouse materials point toward large industrial or logistics accounts; Smart Park, Pangu, and hardware pages point toward property operators, campuses, and enterprise-entry workflows. China Telecom demonstrates that distribution can also run through large platform or channel partners, while Face++ and related identity assets show a separate developer or integrator pathway. This means Megvii should not be understood as having one unified go-to-market motion. Instead, it appears to combine direct enterprise selling, public-sector solution selling, hardware-attached building deployments, and partner-enabled distribution. That diversity is strategically positive because it widens the reachable customer base, but it also complicates any attempt to generalize one average contract type, average sales cycle, or average retention pattern across the whole company.[CU006, CU007, CU008, CU009, CU010, CU013]
| Vertical | Likely buyer | Megvii surface | Buying driver |
|---|---|---|---|
| Smart city | Municipal or public-space operator | City management + network camera | Security, traffic, and governance workflows. |
| Warehouse / logistics | Factory, warehouse, or 3PL operator | Smart warehouse + robotics case study | Efficiency and throughput. |
| Buildings / property | Property manager, campus operator, enterprise admin | Pangu, Smart Park, access devices | Access, attendance, tenant experience. |
| Telecom ecosystem | Large platform / operator partner | China Telecom cooperation | Channel distribution and digital-life services. |
| Developers / integrators | Enterprise developers, solution integrators | Face++, API, Device Authentication | Embed identity and CV functions. |
Shows why Megvii’s customer base should be analyzed by buying center, not just sector label.
[CU006, CU007, CU008, CU010, CU013, CU014]Megvii reaches customers through multiple motions rather than one simple sales channel.
This map synthesizes the distinct customer motions implied by the source set.
[CU006, CU007, CU010, CU013, CU020, CU023]6.3 Deployment quality, repeatability, and what the named references really prove
The named references that matter most are the ones that show repeatability or deployment complexity. The Beijing supermarket rollout is useful because nearly 200 sites imply operational replication rather than a one-off pilot. The Singapore project is useful because it demonstrates exportability and international deployment credibility. China Telecom and Jinyu matter because they suggest Megvii can be embedded in larger property, community, or ecosystem channels rather than always having to win every site one by one. The warehouse case-study material adds another important point: some of Megvii’s customer value is tied to workflow redesign and automation, not just recognition accuracy. Still, these references do not answer the core investor questions. They do not reveal ACVs, renewals, contract durations, revenue share by sector, or whether the most visible deployments are typical or exceptional. Publicly, the chapter can prove traction and technical fit far more easily than it can prove durable customer economics.[CU019, CU020, CU021, CU022, CU027, CU030]
| Reference | Repeatability signal | Implementation depth | Investor takeaway |
|---|---|---|---|
| Koala supermarket rollout | High: nearly 200 sites | Medium | Best proof of multi-site repeatability in reviewed set. |
| Singapore access project | Medium: one named landmark development | High | Best proof of international deployment credibility. |
| China Telecom cooperation | Medium: partner ecosystem breadth | Medium to high | Suggests scalable channel motion if commercialized deeply. |
| Warehouse automation cases | Medium: industrial case set | High | Shows workflow redesign, not just device install. |
Focuses on what the best references actually prove commercially.
[CU019, CU020, CU021, CU022, CU027, CU030]The most useful public customer references show deployment spread across retail, international access, and partner ecosystems.
Timeline uses the clearest dated references from the reviewed customer source set.
[CU003, CU004, CU006, CU007, CU021]6.4 Customer risks and gaps: concentration, trust sensitivity, and missing metrics
The main analytical weakness in Megvii’s customer chapter is not lack of activity but lack of disclosure quality. Public sources do not identify top customers by revenue, do not break down concentration by sector, and do not disclose churn, expansion, or renewal behavior. That matters because Megvii’s visible sectors — especially smart city, property, and industrial projects — could be lumpy or policy-sensitive even if they look impressive in case-study form. Trust and privacy scrutiny also matter. A company can have clear technical fit for identity or access workflows while still facing resistance in overseas or politically sensitive accounts because of sanctions or surveillance concerns. Developer-facing distribution through public hubs may widen reach, but it still does not solve the underlying transparency gap on who pays, renews, and scales. The prudent conclusion is that Megvii’s customer base looks real, diverse, and operationally serious, but not transparent enough to underwrite customer quality with the confidence expected for a late-stage investment case.[CU024, CU025, CU028, CU029, CU034, CU035]
| Missing metric | Why it matters | Current public status |
|---|---|---|
| Top-customer concentration | Needed to judge dependency risk | Not disclosed in reviewed sources. |
| Sector revenue split | Needed to separate government from enterprise exposure | Not disclosed in reviewed sources. |
| Renewal / churn / NRR | Needed to judge durability | Not disclosed in reviewed sources. |
| Average contract size | Needed to understand deployment economics | Not disclosed in reviewed sources. |
| Named reference revenue contribution | Needed to judge whether lighthouse accounts are material or cosmetic | Not disclosed in reviewed sources. |
These are the core customer metrics missing from public evidence.
[CU024, CU028, CU034, CU035, CU036]Customer activity is visible, but customer-quality disclosure remains weak.
Ratings reflect the evidence available in public sources, not internal company knowledge.
[CU021, CU024, CU025, CU028, CU029, CU035]07Risks
7.1 Sanctions and human-rights overhang
Megvii’s highest-confidence risk is the one already formalized by governments and repeated in major reporting: the company sits on the U.S. Entity List and carries a surveillance-related human-rights overhang. That matters in two separate ways. First, it creates technology-supply friction by limiting access to certain U.S. items and components. Second, it creates trust friction with banks, partners, and customers that are sensitive to sanctions, export controls, or reputational scrutiny. Human Rights Watch’s Xinjiang reporting gives that overhang moral and policy weight beyond normal geopolitical noise. Even where the exact technical role of Megvii in any one surveillance workflow may be debated, the diligence consequence is straightforward: the company’s brand is linked to a politically charged category of AI that many institutions would rather avoid. Commercial sanctions databases now echo that status as well, showing that the risk is embedded not only in policy but in day-to-day compliance workflows. This is why sanctions and rights scrutiny should be treated as structural constraints on Megvii’s upside, not one-time press noise.[CR001, CR002, CR005, CR006, CR016, CR017]
| Risk vector | Evidence | Commercial effect | Why it matters |
|---|---|---|---|
| Entity List | Federal Register, Reuters/CNBC | Technology-supply and trust friction | Formal, durable regulatory status. |
| Human-rights overhang | HRW, profiles, controversy summaries | Brand and reputational damage | Can deter customers, partners, and bankers. |
| Surveillance branding | Ongoing third-party summaries | Long-tail stigma | Makes upside more policy-sensitive. |
Summarizes the highest-confidence structural risk vectors.
[CR001, CR002, CR005, CR006, CR016, CR035]Sanctions, rights scrutiny, and trust-sensitive demand reinforce each other rather than acting as isolated issues.
The figure captures causal interaction across disclosed risk factors.
[CR001, CR002, CR005, CR016, CR028]7.2 Capital-markets and liquidity risk
Megvii’s financing history proves that capital access exists, but its listing history proves that exit certainty does not. Reuters and related coverage show that the 2019 Hong Kong process was disrupted after the blacklist, while the later STAR Market filing shows the company kept pursuing public-market optionality without completing the journey. That combination raises three risks. The first is pure liquidity risk: investors do not know when a clean public exit becomes feasible. The second is financing risk: a private company with policy controversy may remain dependent on supportive domestic capital rather than broad global capital. The third is valuation risk: without a public listing or current disclosed round terms, outside investors are left triangulating value from stale markers. The wider 2026 export-control environment also remains fluid, which increases the difficulty of betting that policy pressure will simply fade. Viewed conservatively, Megvii is funded enough to continue operating, but not transparent enough to dismiss capital-markets risk.[CR003, CR004, CR007, CR024, CR025, CR031]
| Issue | Public evidence | Risk implication | Missing evidence |
|---|---|---|---|
| HK IPO setback | Reuters / Yahoo coverage | Delayed liquidity and banker sensitivity | No current timing visibility. |
| STAR filing without listing | SSE filing | Repeated but unfinished listing path | No clear present listing roadmap. |
| Private funding dependence | Tracxn and company funding history | Capital access may skew domestic and relationship-based | No 2025 terms or valuation disclosed. |
| Stale valuation markers | WOWLS, 2019 coverage | Exit pricing uncertain | No updated market-clearing price. |
Separates what is visible about liquidity risk from what remains opaque.
[CR003, CR004, CR007, CR024, CR025, CR031]Megvii’s risk profile is shaped by a sequence of sanctions and unfinished listing attempts.
Dates mark the clearest publicly supported capital-markets milestones in the reviewed set.
[CR003, CR004, CR007, CR024, CR025]7.3 Privacy, biometric regulation, and market-access risk
Megvii’s product surface sits in exactly the categories regulators are tightening around the world: biometrics, public-space imaging, identity verification, and AI-driven decision support. China’s PIPL already treats biometrics as sensitive personal information, and the 2025 face-recognition measures raise the compliance bar further for public-place and face-recognition deployments. Europe moves from another direction, but the practical implication is similar: biometric AI will face heavier scrutiny, documentation, and deployment constraints. These rules matter even if Megvii remains China-centric, because they influence how products must be designed, sold, and explained to customers. They also make export expansion harder for a company whose brand is already controversial. Translated legal references and European political debate point the same way: policymakers are converging on tighter oversight rather than a permissive reset. In risk terms, privacy regulation is not merely a compliance detail for Megvii; it is a strategic limiter on market access and customer trust.[CR009, CR010, CR011, CR012, CR013, CR014]
| Regime | Relevant rule | Why Megvii cares | Operational implication |
|---|---|---|---|
| China PIPL | Biometrics are sensitive personal information | Core products rely on face and identity data | Consent, notice, and governance burden rise. |
| China 2025 face-recognition measures | More detailed rules for application security | Public-place and face-recognition deployments get stricter | Product design and operations must adapt. |
| EU AI Act | Tighter control over biometric AI | Future expansion faces tougher scrutiny | Sales and compliance burden increase abroad. |
The same product strengths that help Megvii sell also place it in heavily regulated categories.
[CR009, CR010, CR011, CR012, CR013, CR015]| Constraint | Mechanism | Near-term effect | Longer-term effect |
|---|---|---|---|
| Sanctions | Export controls and stigma | Harder procurement and partnership in sensitive contexts | Persistent trust handicap. |
| Privacy regulation | Consent and public-place restrictions | Higher compliance cost | Lower expansion flexibility. |
| Foreign civil-liberties pressure | Political opposition to facial recognition | Harder public-sector sales abroad | Narrower international TAM. |
| Private-company opacity | Missing public metrics | Harder investor underwriting | Higher valuation discount. |
Shows why Megvii’s risk stack compounds instead of acting independently.
[CR015, CR016, CR017, CR028, CR033, CR037]Megvii’s main product surfaces sit in categories that face tighter regulation in China and abroad.
Heat-map levels reflect regulatory exposure implied by the chapter’s sources, not legal opinions.
[CR009, CR011, CR013, CR015, CR019, CR020]7.4 Operating risk, mitigants, and the unknowns public evidence cannot close
Public sources also show why the case is not simply uninvestable. Megvii’s products extend beyond the most controversial smart-city narratives into warehouse automation, building software, developer tools, telecom channels, and retail deployments. That breadth mitigates the risk that the company is a one-product or one-customer story. Still, it does not close the most important unknowns. Public evidence does not provide current customer concentration, audited operating metrics, fine history, or a clear map of how much revenue comes from the most policy-sensitive categories. Project-heavy AIoT deployments also imply execution and working-capital risk that are hard to price without internal data. The appropriate synthesis is that Megvii has real mitigation vectors, but the downside remains difficult to quantify — which is itself a risk. The chapter therefore ends in a familiar place for controversial infrastructure AI: enough commercial proof to stay interested, but not enough disclosure to relax. The investment case therefore depends on whether an investor can gain private diligence access good enough to price those unknowns.[CR018, CR019, CR020, CR023, CR026, CR027]
| Category | Mitigant visible publicly | Unknown still open |
|---|---|---|
| Product diversification | Warehouse, building, API, telecom, retail activity | How much revenue each segment contributes. |
| Capital access | Funding history and 2025 round visibility | Current terms, runway, and liquidation preferences. |
| Commercial activity | Named deployments and channels | Renewals, customer concentration, unit economics. |
| Compliance posture | Awareness of regulated categories | Current litigation, fines, audits, or remediation status. |
The public record gives partial comfort on breadth but limited comfort on quantification.
[CR020, CR023, CR026, CR029, CR030, CR032]Megvii has visible mitigation vectors, but its hardest risks remain policy and disclosure-driven.
This matrix synthesizes the chapter’s final judgment on severity and mitigants.
[CR020, CR023, CR026, CR028, CR029, CR032]08Valuation
8.1 Valuation anchors: what the public record actually gives us
Megvii’s public valuation record is anchored far more by funding history than by operating disclosure. The strongest historical marker remains the 2019 financing round that Reuters reported valued the company at slightly above $4 billion. Tracxn and The Company Check still show a very large cumulative funding base — about $1.98 billion across 10 rounds — and Tracxn records an additional undisclosed domestic round in 2025. These facts matter because they establish that Megvii once commanded strong late-stage private-market support and has not been cut off from capital entirely. But they do not solve the central valuation problem: public investors still do not know the company’s current revenue, margin, growth quality, or 2025 round pricing. As a result, the 2019 mark is useful as an anchor, but dangerous as an estimate. It is an old datapoint, not a current clearing price.[CV001, CV002, CV003, CV004, CV005, CV006]
| Anchor | Value / status | What it tells us | Why it is insufficient |
|---|---|---|---|
| 2019 private round | ~$4.0B+ | Last strong public valuation marker | Too stale and pre-blacklist shock. |
| Total capital raised | $1.98B | Shows meaningful historical investor support | Funding total is not enterprise value. |
| 2025 domestic round | Undisclosed | Shows access to capital remains open | No size or price disclosed. |
| Public revenue / margin | Not disclosed | Would normally support multiple selection | Missing today. |
These are the main hard anchors available in public evidence; none is sufficient for a precise current mark.
[CV001, CV003, CV004, CV005, CV006, CV031]The last strong private mark weakens as new risk layers accumulate and current disclosure stays absent.
This is a valuation-logic bridge, not an accounting waterfall.
[CV004, CV008, CV009, CV019, CV032]8.2 Public comparables — and why none are clean
Megvii needs public comparables, but every comparator answers a different question. SenseTime is the closest public China AI platform analogue because it combines large-scale AI ambitions with still-visible losses and disclosure from a public listing. Hikvision shows what a mature trusted AIoT and security incumbent looks like at scale, but its valuation framework is shaped by profitability and entrenched hardware channels that Megvii does not match. Palantir is useful from the opposite direction: it demonstrates how much valuation support audited growth, profitability, and U.S. market access can create for an AI platform, even though the product set is very different. Axis is less about multiple math and more about trust and route-to-market quality. The lesson is that Megvii cannot be valued by blindly porting one peer multiple; investors need a cross-check basket and a large haircut for differences in disclosure, trust, business mix, and policy burden.[CV012, CV013, CV014, CV015, CV023, CV024]
| Comparable | Why it helps | Why it misleads |
|---|---|---|
| SenseTime | Closest China AI platform disclosure analogue | Still not Megvii; different business and public-market context. |
| Hikvision | Shows mature trusted AIoT/security economics | Too incumbent and profitable to map cleanly. |
| Palantir | Shows premium for transparency and public access | Very different geography, trust profile, and software mix. |
| Axis | Shows trust and channel quality benchmark | Not a growth-AI platform analogue. |
No one peer solves the problem; the set is only useful as a bracket.
[CV012, CV013, CV014, CV015, CV023, CV024]| Discount driver | Why it exists | Likely direction |
|---|---|---|
| Entity List | Adds supply and trust risk | Negative valuation impact |
| Human-rights overhang | Narrows investor and customer comfort | Negative valuation impact |
| Private-company opacity | Blocks precise multiple work | Negative valuation impact |
| Failed IPO path | Raises liquidity discount | Negative valuation impact |
| Commercial breadth | Supports floor value | Partially offsets negatives |
Summarizes why Megvii trades as a discounted optionality story, not a clean comparable-multiple story.
[CV008, CV009, CV010, CV016, CV018, CV032]Public comps help explain Megvii only when adjusted for trust, disclosure, and business mix differences.
Ordinal placement reflects chapter evidence rather than live market multiples.
[CV012, CV013, CV014, CV015, CV023, CV024]8.3 Discounts, upside drivers, and scenario framing
The biggest analytical mistake in valuing Megvii would be to treat the company either as a pure surveillance asset or as a clean software platform. Public evidence supports a more complicated picture. Brain++, Face++, warehouse automation, smart-building workflows, telecom channels, and retail or international deployment proof all suggest real commercial breadth. That breadth supports a valuation floor above zero and argues against treating the business as politically untouchable. At the same time, the Entity List, surveillance controversy, failed IPO path, and absence of audited public operating metrics all justify heavy discounting to the stale 2019 anchor. In practical terms, the valuation case is a tug-of-war between product breadth and policy drag. That is why the right public-evidence output is a scenario range: a downside case where the policy and liquidity discount dominates, a middle case where commercial breadth offsets part of it, and an upside case that only works if future disclosure and exit optionality improve materially.[CV008, CV009, CV010, CV016, CV017, CV018]
| Scenario | Indicative range | What must be true |
|---|---|---|
| Downside | $1.5B-$2.5B | Policy discount, opaque economics, and exit risk dominate. |
| Base | $2.0B-$4.5B | Commercial breadth offsets part of the policy and liquidity discount. |
| Upside | $4.0B-$6.0B | Disclosure improves, growth quality holds, and future exit routes reopen. |
Ranges are public-evidence estimates, not market-clearing prices.
[CV034, CV035, CV036, CV038, CV042]A range is more defensible than a point estimate because the key variables are policy and disclosure, not just growth.
Scenario bands are estimation outputs from public evidence, not observed market prices.
[CV034, CV035, CV036, CV038, CV039, CV042]8.4 Valuation verdict: a range around stale anchors, not a precision mark
From public evidence alone, Megvii looks like a company whose valuation should be bracketed rather than pinpointed. The last clean funding anchor is old, the private-company disclosure gap is large, and the sanctions overhang is real. Those factors make a public-like premium multiple difficult to defend. Still, the company’s capital history, continued commercial activity, and exposure to large AI markets mean it also cannot be dismissed as a stranded asset. The most defensible interpretation is a risk-adjusted valuation range centered below the 2019 mark, with upside only if investors gain confidence on revenue quality, compliance posture, and realistic exit routes. In other words, Megvii’s value is not primarily a function of TAM; it is a function of whether private diligence can narrow the trust and disclosure discount enough to let the market believe the rest of the story. Until that happens, scenario discipline matters more than spreadsheet precision, and patience matters more than theoretical upside today.[CV034, CV035, CV036, CV037, CV038, CV039]
| Potential path | What would need to improve first |
|---|---|
| Domestic strategic sale | Buyer confidence on compliance, product fit, and integration value. |
| Future listing attempt | Audited disclosure, policy stability, and better investor trust. |
| Longer private hold | Supportive domestic capital and clearer unit economics. |
Valuation upside is partly a question of which exit path becomes realistic.
[CV021, CV037, CV041]| Constraint | Why it blocks precision |
|---|---|
| Stale last round | The only clean public valuation marker is from 2019. |
| Opaque 2025 round | The newest financing exists without disclosed pricing or terms. |
| Missing audited operations | Revenue, margin, and mix are not public enough for tight multiple work. |
| Policy discount uncertainty | Sanctions and trust effects are real but hard to quantify precisely. |
Summarizes why this chapter uses ranges and scenarios rather than a single valuation mark.
[CV019, CV031, CV032, CV041, CV042]Valuation depends heavily on whether Megvii can move from stale funding anchors to a credible future exit path.
The timeline tracks the public milestones that most affect exit credibility.
[CV004, CV005, CV008, CV021, CV037, CV041]Disclaimer
This report is based on publicly available information as of 2026-08-14 and does not constitute investment advice. Megvii is a private company with limited public disclosure, so valuation and operating conclusions should be treated as scenario-based rather than precise.
Evidence index
| ID | Statement | Confidence | Sources |
|---|---|---|---|
| CO001 | Megvii was founded in October 2011 in Beijing. | High | SO002, SO003, SO005 |
| CO002 | Megvii's founders are Yin Qi, Tang Wenbin, and Yang Mu. | High | SO002, SO003, SO019 |
| CO003 | The founding team is consistently described as Tsinghua University alumni associated with Andrew Yao's Yao Class network. | Medium | SO002, SO003, SO019 |
| CO004 | Megvii remains a private company rather than a listed issuer as of the 2026 run date. | Medium | SO003, SO005, SO007 |
| CO005 | Megvii markets Face++ as its flagship facial-recognition and computer-vision platform. | High | SO001, SO024 |
| CO006 | Megvii describes Brain++ as its proprietary AI productivity platform. | High | SO001, SO015 |
| CO007 | Brain++ is built around MegEngine, MegData, and MegCompute. | High | SO015, SO001 |
| CO008 | Megvii still frames its commercialization around Personal IoT, City IoT, and Supply Chain IoT. | High | SO001, SO015, SO019 |
| CO009 | Megvii sells full-stack solutions combining algorithms, software, hardware, and AI-enabled IoT devices. | High | SO001, SO015, SO025 |
| CO010 | Megvii says it operates the world's largest computer-vision research institute. | Medium | SO002 |
| CO011 | Megvii says it has won 49 world championships in leading international AI competitions since 2017. | Medium | SO002 |
| CO012 | Megvii says it won the ICCV COCO challenge for a third consecutive year in 2019. | Medium | SO002 |
| CO013 | MegEngine was open sourced in March 2020. | High | SO016, SO015 |
| CO014 | Megvii launched the commercial version of Brain++ in September 2020. | Medium | SO015 |
| CO015 | Megvii published an AI Ethics Code of Conduct in 2019. | High | SO017, SO002 |
| CO016 | Megvii says it established both an AI Ethics Committee and an AI Ethics Research Institute. | Medium | SO002, SO017 |
| CO017 | China's Ministry of Science and Technology recognized Megvii as a national next-generation AI open innovation platform for image perception in 2019. | Medium | SO018 |
| CO018 | Megvii announced that its 2019 Series D financing totaled approximately $750 million. | High | SO014, SO009, SO008 |
| CO019 | Megvii named BOCGI, an ADIA subsidiary, Macquarie, and ICBC Asset Management (Global) among the participants in the 2019 Series D. | High | SO014, SO009 |
| CO020 | Reuters reported that BOCGI led the 2019 financing with a $200 million commitment. | Medium | SO009 |
| CO021 | Reuters reported that existing investor Alibaba also participated in the 2019 round. | Medium | SO009 |
| CO022 | Reuters reported that the 2019 financing valued Megvii at slightly above $4 billion. | High | SO009, SO005, SO007 |
| CO023 | Tracxn says Megvii has raised $1.98 billion across 10 funding rounds. | Medium | SO006, SO005, SO004 |
| CO024 | The Company Check also reports total funding of $1.98 billion across 10 rounds. | Medium | SO004, SO006 |
| CO025 | Tracxn says Megvii's latest funding round was an undisclosed Series D on April 8, 2025. | Medium | SO006, SO005 |
| CO026 | Tracxn attributes the April 2025 round to Ant Group, Legend Holdings, and Chongqing Industrial Investment Fund. | Medium | SO006 |
| CO027 | GovInfo's October 9, 2019 Federal Register notice says the U.S. government added 28 China-based entities to the Entity List for acting contrary to U.S. foreign policy interests. | Medium | SO013 |
| CO028 | CNBC and Reuters reported that Megvii was one of the Chinese AI firms blacklisted in October 2019 over allegations tied to human-rights abuses in Xinjiang. | High | SO010, SO011 |
| CO029 | Human Rights Watch documented Xinjiang's Integrated Joint Operations Platform as a mass-surveillance system used against Uyghurs and other Turkic Muslims. | Medium | SO012 |
| CO030 | Reuters reported in November 2019 that HKEX regulators asked Megvii additional questions and did not approve its IPO at the committee hearing. | Medium | SO011 |
| CO031 | Reuters reported the company was targeting a Hong Kong IPO sized at roughly $500 million to $1 billion before the regulatory setback. | Medium | SO011 |
| CO032 | Nextomoro says Megvii's Hong Kong and Shanghai listing attempts from 2019 through 2024 remained suspended rather than completed. | Medium | SO003, SO007 |
| CO033 | Megvii said its Koala temperature-screening solution was deployed in 191 Beijing supermarkets including Chaoshifa and Wumart. | Medium | SO019 |
| CO034 | Megvii said Koala deployments also reached enterprise customers in Thailand, Brazil, and the UAE. | Medium | SO020 |
| CO035 | Megvii said its Singapore smart-access project served a 280-meter integrated development with over 60,000 square meters of lease area. | Medium | SO021 |
| CO036 | Megvii said it helped digitalize major Beijing Winter Olympics venues including the Bird's Nest and Ice Ribbon. | Medium | SO022 |
| CO037 | Megvii said its China Telecom cooperation covers smart community, Tianyi home-security, Tianyi cloud-eye, and open video-capability scenarios. | Medium | SO023 |
| CO038 | Megvii's city-solution page positions smart traffic, AI-enabled epidemic prevention, and urban-governance workflows as core applications. | Medium | SO025 |
| CO039 | Megvii's warehouse-solution page positions logistics-center scheduling, forecasting, and automated decision-making as core supply-chain use cases. | Medium | SO026 |
| CO040 | The 2020 Koala deployment release described Megvii as having more than 2,300 employees and four R&D centers in China at that time. | Medium | SO019 |
| CO041 | WOWLS characterizes Megvii as politically radioactive in Western markets because its surveillance use cases and Entity List status constrain investor appetite. | Low | SO007 |
| CO042 | Neither the public sources reviewed here nor the company's English about page provide a current 2026 board roster or current headcount figure. | Medium | SO002, SO003, SO004 |
| CM001 | IMARC estimates the global computer-vision market at $21.7 billion in 2025. | Medium | SM001 |
| CM002 | IMARC projects the global computer-vision market will reach $35.4 billion by 2034. | Medium | SM001 |
| CM003 | IMARC projects a 2026-2034 CAGR of 5.6% for computer vision. | Medium | SM001 |
| CM004 | IMARC says Asia Pacific held more than 41% of the computer-vision market in 2025. | Medium | SM001 |
| CM005 | Fortune Business Insights values the global computer-vision market at $20.75 billion in 2025. | Medium | SM002 |
| CM006 | Fortune Business Insights projects the market will grow to $24.14 billion in 2026. | Medium | SM002 |
| CM007 | Fortune Business Insights projects the market to reach $72.8 billion by 2034. | Medium | SM002 |
| CM008 | Fortune Business Insights implies a 2026-2034 CAGR of 14.8% for the computer-vision market. | Medium | SM002 |
| CM009 | MarketsandMarkets estimates the AI-in-computer-vision market at $23.42 billion in 2025. | Medium | SM003 |
| CM010 | MarketsandMarkets projects the AI-in-computer-vision market to reach $63.48 billion by 2030. | Medium | SM003 |
| CM011 | MarketsandMarkets estimates a 2025-2030 CAGR of 22.1% for AI in computer vision. | Medium | SM003 |
| CM012 | The Business Research Company says the facial-recognition market reached $7.88 billion in 2025. | Medium | SM004 |
| CM013 | The Business Research Company projects facial recognition will reach $17.63 billion by 2030. | Medium | SM004 |
| CM014 | Mordor Intelligence estimates the facial-recognition market will grow from $8.58 billion in 2025 to $20.88 billion by 2031. | Medium | SM005 |
| CM015 | Mordor argues facial-recognition growth is increasingly shaped by edge architectures and stricter biometric-consent laws. | Medium | SM005 |
| CM016 | Mordor estimates the broader computer-vision market at $27.39 billion in 2025 and $68.38 billion by 2031. | Medium | SM006 |
| CM017 | Mordor says hardware still dominates computer-vision revenue while software-margin and edge deployment are rising fastest. | Medium | SM006 |
| CM018 | Verified Market Research values the computer-vision market at $13.04 billion in 2024 and $23.79 billion by 2032. | Medium | SM007 |
| CM019 | Statista's China smart-city topic tracks market size and pilot-project counts as still-relevant demand indicators for urban AI procurement. | Medium | SM008 |
| CM020 | Megvii's city-solution page centers smart traffic management, epidemic prevention, and urban-governance workflows. | Medium | SM009 |
| CM021 | Megvii's warehouse-solution page centers logistics-center scheduling, forecasting, and automated decision-making. | Medium | SM010 |
| CM022 | Megvii's Brain++ page shows the company competes not only as an application vendor but also as an AI-production platform vendor. | Medium | SM011 |
| CM023 | Face++ positions Megvii inside developer, identity, and enterprise computer-vision workflows rather than only closed government projects. | Medium | SM012 |
| CM024 | SenseTime's official about page shows a market peer spanning generative AI, vision AI, and infrastructure through SenseCore. | Medium | SM013 |
| CM025 | YITU's official English site still highlights smart-city and healthcare AI applications. | Medium | SM014 |
| CM026 | Hikvision and Dahua remain physical-security incumbents that define buyer expectations for surveillance hardware and integrated solutions. | Medium | SM015, SM018 |
| CM027 | Axis and Hanwha Vision demonstrate that global enterprise buyers still evaluate AI vision through the lens of established network-video and smart-visual vendors. | Medium | SM016, SM017 |
| CM028 | Clearview AI represents a law-enforcement-first facial-recognition model that differs from Megvii's broader AIoT platform ambition. | Medium | SM019 |
| CM029 | Megvii's real market boundary includes software, models, edge hardware, deployment services, and ongoing operations support, not only stand-alone APIs. | Medium | SM009, SM010, SM011 |
| CM030 | The most relevant buyers for Megvii are public-sector agencies, enterprise security and facilities teams, logistics operators, device makers, and fintech identity teams. | Medium | SM009, SM010, SM012, SM020 |
| CM031 | Budget ownership varies by use case, with CIO/CTO, municipal IT, security operations, logistics operations, and digital-channel leaders all relevant. | Medium | SM009, SM010, SM012 |
| CM032 | Recurring adoption triggers include fraud reduction, throughput gains, automation, safety, and compliance. | Medium | SM003, SM004, SM005, SM006 |
| CM033 | Recurring constraints include privacy regulation, export controls, consent requirements, switching cost, and reputational scrutiny. | Medium | SM005, SM023, SM024, SM025 |
| CM034 | Megvii's sanctions exposure makes its serviceable market narrower than the headline global computer-vision TAM. | Medium | SM023, SM024, SM025 |
| CM035 | Facial recognition is materially smaller than the broader computer-vision market, so Megvii needs city, logistics, and enterprise software expansion to outgrow a narrow biometric niche. | Medium | SM004, SM005, SM001, SM002 |
| CM036 | The reviewed public market reports disagree materially on both current size and long-range growth, so a valuation model should use ranges rather than a single TAM point. | Medium | SM001, SM002, SM006, SM007 |
| CM037 | No reviewed public source provides a clean China-only sanctioned SAM for Megvii after factoring in export controls and surveillance stigma. | Low | SM008, SM023, SM024, SM025 |
| CM038 | No reviewed public source provides precise procurement-cycle or budget-owner detail for Megvii's current 2026 sales motion by segment. | Low | SM020, SM021, SM022 |
| CP001 | SenseTime's official about page says the company was founded in 2014 and positions itself as a leading AI software company. | Medium | SP002 |
| CP002 | SenseTime says its business spans Generative AI, Vision AI, and innovation-driven segments underpinned by SenseCore infrastructure. | Medium | SP001, SP002 |
| CP003 | YITU's English site still foregrounds smart city projects and healthcare AI applications. | Medium | SP004 |
| CP004 | Hikvision's solutions pages show it remains a broad physical-security and video-solutions incumbent. | Medium | SP007 |
| CP005 | Dahua presents itself as a video-centric smart-IoT solutions provider. | Medium | SP015 |
| CP006 | Hanwha Vision presents itself as a smart visual intelligence company serving multiple industries. | Medium | SP009, SP010 |
| CP007 | Axis says it develops network solutions for safety, security, business intelligence, and efficiency. | Medium | SP011, SP012 |
| CP008 | Axis publicly discloses roughly 5,000 employees in over 50 countries. | Medium | SP011 |
| CP009 | Axis publicly disclosed 2025 sales of about $2.1 billion. | Medium | SP011 |
| CP010 | Clearview AI presents itself as a facial-recognition provider built for law enforcement, public safety, and secure commerce. | Medium | SP013 |
| CP011 | CloudWalk's Wikipedia page identifies it as a public Chinese facial-recognition and AI company. | Medium | SP017 |
| CP012 | Tracxn says Megvii has 1,962 active competitors. | Medium | SP019 |
| CP013 | Tracxn says 330 of Megvii's active competitors are funded and 217 have exited. | Medium | SP019 |
| CP014 | The Company Check names competitors such as Dahua Technology, IntelliFusion, Oosto, Unico, and Evrotrust in Megvii's peer set. | Medium | SP018 |
| CP015 | Megvii competes across multiple archetypes rather than a single competitor set: AI platform peers, surveillance incumbents, developer-tool vendors, and niche biometric players. | Medium | SP018, SP019, SP022 |
| CP016 | SenseTime and Megvii both combine proprietary AI infrastructure with application-layer products, making them closer peers than hardware-only incumbents. | Medium | SP001, SP002, SP022 |
| CP017 | Hikvision and Dahua hold an installed-base and channel advantage because their offer begins with physical-security hardware and integrated surveillance systems. | Medium | SP007, SP015 |
| CP018 | Axis and Hanwha provide global distribution and brand trust that Megvii does not match outside China. | Medium | SP009, SP011, SP012 |
| CP019 | Face++ gives Megvii a developer and identity-distribution surface that hardware incumbents generally lack. | Medium | SP023 |
| CP020 | YITU, SenseTime, CloudWalk, and Megvii belong to the same China-native vision-AI cohort competing for government and enterprise workloads. | Medium | SP002, SP004, SP017, SP022 |
| CP021 | Clearview is narrower than Megvii because its public posture is centered on facial-recognition search and law-enforcement use rather than a broader AIoT stack. | Medium | SP013, SP022 |
| CP022 | Buyer comparisons in this market typically weigh accuracy, integration breadth, deployment support, regulatory posture, and channel reach rather than one transparent price list. | Medium | SP007, SP011, SP012, SP023 |
| CP023 | Project-based bundling and hardware-software integration make direct public pricing comparisons rare across the peer set. | Medium | SP007, SP012, SP015 |
| CP024 | Smart city and smart building projects create meaningful switching cost once cameras, access control, software rules, and operator workflows are integrated. | Medium | SP007, SP012, SP015, SP023 |
| CP025 | Warehouse and logistics use cases add another switching-cost layer because algorithms, edge hardware, and operating processes are embedded into site workflows. | Medium | SP022, SP023 |
| CP026 | Megvii's trust posture is weaker in overseas markets because the U.S. Entity List and Xinjiang-related scrutiny are directly attached to its brand. | Medium | SP021, SP024, SP025 |
| CP027 | Overseas incumbents such as Axis and Hanwha do not carry the same sanctions baggage in cross-border enterprise sales. | Medium | SP009, SP011, SP024 |
| CP028 | The stalled Hong Kong IPO reduced Megvii's access to public-market signaling and currency compared with public Chinese peers such as CloudWalk. | Medium | SP017, SP021 |
| CP029 | The Company Check competitor list shows that Megvii also faces identity-verification and video-analytics vendors outside the traditional China AI dragons. | Medium | SP018 |
| CP030 | SenseTime's infrastructure message around SenseCore suggests that compute and platform control are becoming a moat dimension, not just model accuracy. | Medium | SP001, SP002 |
| CP031 | Megvii's own moat is stronger when sold as a full-stack AIoT platform than when sold as stand-alone facial recognition. | Medium | SP022, SP023 |
| CP032 | Facial recognition risks commoditization when it is reduced to a feature inside broader hardware or software bundles. | Medium | SP007, SP012, SP015 |
| CP033 | Hardware incumbents are harder to displace on installed-base economics, while AI platform peers are harder to displace on algorithmic breadth and infrastructure ownership. | Medium | SP001, SP007, SP011, SP015 |
| CP034 | No reviewed source publishes a clean, apples-to-apples public price list across Megvii, SenseTime, Hikvision, and the other main peers. | Low | SP007, SP012, SP015, SP023 |
| CP035 | No reviewed source publishes peer win-rate, renewal-rate, or churn data that would let investors measure competitive durability directly. | Low | SP018, SP019, SP022 |
| CP036 | The competitive set splits into at least four buckets: China-native AI platforms, surveillance incumbents, global trusted vision vendors, and law-enforcement specialists. | Medium | SP001, SP004, SP007, SP011, SP013, SP017 |
| CP037 | Megvii's rivalry with SenseTime and YITU is strategically closer than its rivalry with Axis because the China peers overlap more directly in software-defined city and enterprise AI workloads. | Medium | SP001, SP002, SP004, SP022 |
| CI001 | Tracxn says Megvii has raised $1.98 billion across 10 funding rounds. | Medium | SI002, SI003, SI001 |
| CI002 | The Company Check also reports $1.98 billion of total funding across 10 rounds. | Medium | SI001, SI003 |
| CI003 | Tracxn records an undisclosed Series D round on April 8, 2025. | Medium | SI003, SI002 |
| CI004 | Tracxn names Ant Group, Legend Holdings, and Chongqing Industrial Investment Fund as participants in the 2025 round. | Medium | SI003 |
| CI005 | Reuters reported that Megvii raised $750 million in May 2019. | Medium | SI004, SI005, SI006 |
| CI006 | Reuters reported that the 2019 financing valued Megvii at slightly above $4 billion. | Medium | SI004, SI008 |
| CI007 | Megvii said the 2019 proceeds would be used to strengthen deep-learning technology, accelerate commercialization, recruit talent, and support global expansion. | Medium | SI006, SI007, SI004 |
| CI008 | WOWLS still frames $4 billion in 2019 as the last known public valuation marker and warns current value could be lower. | Low | SI008 |
| CI009 | The Company Check page states that annual revenue is not publicly available. | Medium | SI001 |
| CI010 | The accessible CB Insights financials page does not disclose usable public revenue or profit figures in the reviewed output. | Medium | SI015 |
| CI011 | No reviewed source discloses ARR for Megvii. | Medium | SI001, SI015 |
| CI012 | No reviewed source discloses gross margin for Megvii. | Medium | SI001, SI015 |
| CI013 | No reviewed source discloses burn or monthly cash use for Megvii. | Medium | SI001, SI015 |
| CI014 | No reviewed source discloses runway or cash on hand for Megvii. | Medium | SI001, SI015 |
| CI015 | Megvii Pangu monetizes access control, attendance, visitor management, and regional security management workflows. | Medium | SI016, SI017 |
| CI016 | Pangu highlights contactless access, modular architecture, and open APIs, implying both software licensing and integration revenue. | Medium | SI016 |
| CI017 | Megvii's Device Authentication solution positions face verification and identity authentication as monetizable product surfaces. | Medium | SI018 |
| CI018 | Face++ remains a developer-facing facial-recognition and computer-vision platform, implying API or developer-channel monetization. | Medium | SI023 |
| CI019 | Megvii's Smart City Management solution implies project, integration, and operational revenue tied to urban governance workflows. | Medium | SI022 |
| CI020 | Megvii's Smart Warehouse solution implies project and workflow revenue tied to logistics-center automation and decision support. | Medium | SI021 |
| CI021 | Megvii's commercial Brain++ release says the platform shortened algorithm-development time by 80% and reduced algorithm-production cost by 55%. | Medium | SI020 |
| CI022 | Brain++ combines algorithm development, cluster construction, deployment, and support, making it relevant both to internal efficiency and external enterprise sales. | Medium | SI019, SI020 |
| CI023 | Megvii's public product mix implies a hybrid revenue model spanning software, hardware, APIs, and deployment services rather than a single recurring SaaS stream. | Medium | SI016, SI017, SI018, SI021, SI022, SI023 |
| CI024 | That hybrid mix implies heterogeneous gross margins across product lines. | Medium | SI016, SI021, SI022, SI023 |
| CI025 | Project-heavy smart-city and warehouse deployments likely lengthen sales cycles and increase implementation intensity relative to pure API sales. | Medium | SI021, SI022, SI009 |
| CI026 | The China Telecom cooperation shows Megvii monetizes partnership channels in smart community, home-security, and cloud-eye scenarios. | Medium | SI024 |
| CI027 | The Koala supermarket deployment shows Megvii can generate revenue from retail access and temperature-screening installations. | Medium | SI025 |
| CI028 | The Smart Identity Verification Device page shows Megvii also sells dedicated hardware tied to identity-verification workflows, reinforcing the company’s hybrid hardware-plus-software revenue profile. | Medium | SI027 |
| CI029 | Public operating proof exists, but it is deployment proof rather than audited revenue-quality disclosure. | Medium | SI024, SI025, SI001, SI015 |
| CI030 | Reuters reported that Megvii's Hong Kong IPO did not win approval at the 2019 committee hearing, delaying liquidity. | Medium | SI010 |
| CI031 | The March 2021 Shanghai Stock Exchange filing shows Megvii pursued a STAR Market / CDR listing path after the Hong Kong process stalled. | Medium | SI026 |
| CI032 | CNBC and Reuters reported that the Entity List bars Megvii from buying U.S. parts and components without government approval. | Medium | SI011, SI012 |
| CI033 | Human Rights Watch's Xinjiang surveillance reporting explains why Megvii's financing risk is not just timing risk but also reputational and policy risk. | Medium | SI013 |
| CI034 | The 2025 domestic-investor round indicates Megvii still has access to supportive capital even though amount and valuation were not disclosed. | Medium | SI003 |
| CI035 | No reviewed source discloses debt, project-finance obligations, or working-capital facilities for Megvii. | Medium | SI001, SI015 |
| CI036 | The last public headcount marker in the reviewed source set was more than 2,300 employees in 2020, implying a meaningful historical cost base. | Medium | SI025 |
| CI037 | Megvii looks financially more like a capital-intensive AIoT integrator than a pure software API business. | Medium | SI021, SI022, SI023, SI025, SI027 |
| CI038 | The absence of audited public 2025 or 2026 revenue and profit statements is the single largest diligence blocker in Megvii's financial chapter. | Medium | SI001, SI015, SI026 |
| CI039 | No reviewed source discloses CAC, payback, NRR, or win-rate data, so sales efficiency cannot be underwritten from public evidence alone. | Medium | SI001, SI015 |
| CE001 | Brain++ is Megvii’s proprietary AI productivity platform. | Medium | SE001 |
| CE002 | Megvii says Brain++ covers algorithm production, model training, and deployment-related workflow support. | Medium | SE001, SE002 |
| CE003 | Megvii says the commercial Brain++ release shortened algorithm-development time by 80% and reduced algorithm-production cost by 55%. | Medium | SE002 |
| CE004 | Pangu packages access management, attendance, visitor management, and regional security management into one software layer. | Medium | SE003 |
| CE005 | Pangu’s open API / embedded-SDK positioning shows Megvii expects partners to embed its algorithms into third-party software and hardware. | Medium | SE004 |
| CE006 | The Face Recognition and Access Control Terminal supports on-device offline recognition and libraries of up to 100,000 entries. | Medium | SE005 |
| CE007 | The Smart Identity Verification Device page advertises accuracy above 99% and verification speed below 400ms. | Medium | SE006 |
| CE008 | The Smart Network Camera page says Megvii combines proprietary face-recognition algorithms with cloud-edge-device deep learning in camera products. | Medium | SE007 |
| CE009 | Megvii’s Device Authentication offering shows the company extends beyond access control into identity-authentication and device-side vision workflows. | Medium | SE008 |
| CE010 | Megvii’s face-recognition technology page says the stack is powered by the MegEngine deep-learning framework. | Medium | SE009 |
| CE011 | The face-recognition technology page also says Megvii’s face stack is built on big data and designed for diverse real-life scenarios. | Medium | SE009 |
| CE012 | MegEngine is publicly available as an open-source repository on GitHub. | Medium | SE010 |
| CE013 | The MegEngine repository describes the framework as unified for both training and inference. | Medium | SE010 |
| CE014 | The MegEngine repository also highlights quantization and dynamic-shape / image-preprocessing support. | Medium | SE010 |
| CE015 | MegFlow is publicly available as a separate official repository positioned as an efficient ML solution for long-tailed demands. | Medium | SE011 |
| CE016 | The Face++ China site says the platform offers Web APIs and SDKs and serves developers and enterprise users across more than 220 countries and regions. | Medium | SE012 |
| CE017 | Megvii maintains an enterprise FaceID download page, indicating a packaged identity-verification product for business users. | Medium | SE013 |
| CE018 | The FaceID solution page shows Megvii applies computer vision to industrial campus or smart-campus workflows. | Medium | SE014 |
| CE019 | The Facestyle solution page shows Megvii also applies vision technology to online marketing and interactive media use cases. | Medium | SE015 |
| CE020 | The Smart Park page shows Megvii packages building-access intelligence as an AIoT building solution rather than a standalone algorithm. | Medium | SE016 |
| CE021 | The Access Control for Smart Building page shows Megvii adapted the stack for temperature measurement and pandemic-era access workflows. | Medium | SE017 |
| CE022 | The SMB attendance page shows the company has lighter-weight building and attendance packages in addition to large-enterprise offerings. | Medium | SE018 |
| CE023 | The Smart Warehouse solution shows Megvii applies perception and automation tech to logistics-center workflows. | Medium | SE019 |
| CE024 | The Smart City Management solution shows Megvii translates vision algorithms into urban-governance and public-space management workflows. | Medium | SE020 |
| CE025 | The Singapore smart-access case shows Megvii can deploy its access stack outside mainland China. | Medium | SE021 |
| CE026 | The China Telecom partnership shows Megvii’s stack can be embedded into operator-led digital-life ecosystems. | Medium | SE022 |
| CE027 | The Beijing supermarket deployment shows Megvii can productize temperature-screening and recognition tech at multi-site retail scale. | Medium | SE023 |
| CE028 | The Jinyu property-management cooperation shows Megvii positions smart-building and property-tech as part of its commercial stack. | Medium | SE024 |
| CE029 | The 2015 Megvii face-recognition paper shows the company’s technical story includes original research, not just product marketing. | Medium | SE025 |
| CE030 | That paper reported 99.50% accuracy on the LFW benchmark, illustrating early research strength in face recognition. | Medium | SE025 |
| CE031 | RepVGG is an academic paper by Megvii researchers, showing ongoing contribution to mainstream computer-vision model design beyond one narrow product line. | Medium | SE026 |
| CE032 | Megvii’s public stack is broader than a pure facial-recognition API because it spans frameworks, devices, industry software, and vertical solutions. | Medium | SE001, SE003, SE005, SE007, SE019, SE020 |
| CE033 | Megvii’s public stack is more software-and-model centric than a pure hardware-security vendor because it exposes frameworks, APIs, and developer tooling in addition to devices. | Medium | SE004, SE010, SE012, SE005, SE007 |
| CE034 | The strongest switching-cost layer appears when Megvii’s models, devices, access workflows, and management software are deployed together. | Medium | SE003, SE005, SE016, SE018 |
| CE035 | The stack likely requires meaningful implementation effort in smart-city, warehouse, and property deployments because solutions are sold as workflow systems rather than drop-in widgets. | Medium | SE019, SE020, SE024 |
| CE036 | Public technical evidence is strongest on breadth of product surfaces and weakest on independently validated benchmark economics, patent counts, and current deployment scale by product line. | Medium | SE001, SE010, SE025, SE026 |
| CE037 | Many performance statements in Megvii’s product pages are company claims rather than benchmark results reproduced by independent evaluators. | Medium | SE002, SE006, SE009 |
| CE038 | The Papers With Code page for RepVGG shows Megvii research outputs are discoverable in mainstream developer and model-discovery workflows. | Medium | SE027 |
| CE039 | The CVF open-access page confirms RepVGG as a published CVPR 2021 paper rather than only an arXiv preprint. | Medium | SE028, SE026 |
| CE040 | Megvii Research has a public organization page on Hugging Face, indicating presence in a widely used model-sharing ecosystem even if public artifacts are limited. | Medium | SE029 |
| CE041 | Replicate hosts Megvii-research/NAFNet as a runnable API model, showing at least one Megvii research asset distributed through an external developer platform. | Medium | SE030, SE032 |
| CE042 | Developer-distribution evidence spans multiple external ecosystems, but the reviewed sources still do not quantify active users, downloads, or model-call volume. | Medium | SE027, SE029, SE030, SE031, SE032 |
| CU001 | Public sources show Megvii serves a mix of government, enterprise, property, logistics, retail, telecom-ecosystem, and developer-facing customer types. | Medium | SU005, SU006, SU007, SU010, SU015, SU016 |
| CU002 | Megvii publicly disclosed a smart-access deployment for a landmark development in Singapore. | Medium | SU001 |
| CU003 | Megvii publicly disclosed cooperation with China Telecom spanning smart community and digital-life scenarios. | Medium | SU002 |
| CU004 | Megvii publicly disclosed deployment of its screening solution at nearly 200 supermarkets in Beijing. | Medium | SU003 |
| CU005 | Megvii publicly disclosed strategic cooperation with Jinyu property management. | Medium | SU004 |
| CU006 | The Smart City solution and industry pages show government and public-space operators are a core customer segment. | Medium | SU005, SU015 |
| CU007 | The Smart Warehouse solution and Megvii Robotics case page show logistics and manufacturing operators are a core enterprise customer segment. | Medium | SU006, SU011 |
| CU008 | The Smart Park, Smart Building, and Pangu materials show building owners, campuses, and property operators are a core customer segment. | Medium | SU007, SU008, SU016 |
| CU009 | The FaceID solution indicates schools, campuses, and related operators are addressable customers. | Medium | SU009 |
| CU010 | Face++ indicates Megvii also serves developers and enterprise software integrators rather than only site-specific solution buyers. | Medium | SU010, SU025 |
| CU011 | The AMD case study shows Megvii’s customer delivery stack depends partly on external technology partners. | Medium | SU012 |
| CU012 | Nextomoro and The Company Check both describe Megvii as serving smart city, logistics, and enterprise security markets. | Medium | SU013, SU022 |
| CU013 | Pangu’s open API positioning implies Megvii can reach customers indirectly through integrators and OEM-like partners. | Medium | SU017 |
| CU014 | Device Authentication implies customer demand from device makers or identity-sensitive digital services. | Medium | SU018 |
| CU015 | The access-control terminal and identity device pages imply commercial building, campus, and enterprise-entry scenarios. | Medium | SU019, SU020 |
| CU016 | The smart network camera page implies security, public-space, and video-analytics buyers. | Medium | SU021 |
| CU017 | Megvii’s named customer evidence is stronger in deployments and partnerships than in Western-style customer testimonials. | Medium | SU001, SU002, SU003, SU004, SU011 |
| CU018 | Much of the reviewed customer evidence is partner proof or company proof rather than independently audited buyer references. | Medium | SU002, SU012, SU024, SU025 |
| CU019 | The Singapore project is the clearest reviewed proof of an international customer deployment. | Medium | SU001 |
| CU020 | China Telecom is the clearest reviewed proof of ecosystem distribution through a large domestic partner. | Medium | SU002 |
| CU021 | The Beijing supermarket deployment is the clearest reviewed proof of repeatability across many sites. | Medium | SU003 |
| CU022 | The warehouse case-study page suggests Megvii Robotics targets large industrial accounts rather than small self-serve buyers. | Medium | SU011 |
| CU023 | Face++ and related enterprise identity assets suggest at least some developer and integration-led customer acquisition in addition to direct enterprise selling. | Medium | SU010, SU025, SU026, SU027, SU028, SU029 |
| CU024 | Megvii’s public customer list appears sector-diverse, but public disclosure is not sufficient to measure actual revenue concentration. | Medium | SU013, SU022 |
| CU025 | Public-sector and smart-city exposure likely make government-adjacent demand strategically important even if exact revenue share is undisclosed. | Medium | SU005, SU015, SU021 |
| CU026 | The property and building stack suggests Megvii’s customer motion often runs through operators responsible for access, attendance, security, and tenant experience. | Medium | SU007, SU008, SU016 |
| CU027 | The warehouse and retail evidence suggests Megvii’s customer value proposition includes operational efficiency as well as security. | Medium | SU003, SU006, SU011 |
| CU028 | The current public evidence does not disclose named top customers by revenue, renewal rates, or contract duration. | Medium | SU022, SU023 |
| CU029 | Privacy scrutiny and sanctions likely make some overseas or trust-sensitive customers harder to win even where Megvii has technical fit. | Medium | SU023, SU013 |
| CU030 | The company’s customer proofs are strongest where Megvii controls both workflow software and on-site devices. | Medium | SU007, SU008, SU019, SU020 |
| CU031 | Megvii appears to have both lighthouse-style named references and many broader vertical claims that are not attached to named accounts. | Medium | SU014, SU015, SU016 |
| CU032 | Retail, telecom, building, city, and logistics references together imply customer diversification by use case rather than reliance on one narrow product. | Medium | SU002, SU003, SU006, SU007, SU015 |
| CU033 | The reviewed sources show more evidence of enterprise and institution buyers than of true consumer demand. | Medium | SU001, SU002, SU006, SU007, SU009 |
| CU034 | SMB and lighter-weight access or attendance packages exist publicly, but named SMB customers are not disclosed in the reviewed evidence. | Medium | SU007, SU008 |
| CU035 | No reviewed source discloses net retention, churn, or customer lifetime metrics for Megvii’s customer base. | Medium | SU022, SU013 |
| CU036 | The customer chapter supports real deployment activity, but not a clean investor-grade customer concentration model. | Medium | SU001, SU002, SU003, SU004, SU022 |
| CU037 | Named-customer disclosure is sufficient to prove commercial traction, but insufficient to prove repeatable revenue quality or sector balance. | Medium | SU001, SU002, SU003, SU004, SU011 |
| CR001 | The U.S. government added Megvii to the Entity List in October 2019. | Medium | SR001, SR002 |
| CR002 | The Federal Register notice ties the listing to involvement in surveillance technology contrary to U.S. foreign-policy interests. | Medium | SR001 |
| CR003 | CNBC and Reuters reported that the blacklist forced Goldman Sachs to evaluate its role in Megvii’s IPO. | Medium | SR002 |
| CR004 | Reuters reported that Megvii’s Hong Kong IPO was hit by a regulatory setback after the blacklist. | Medium | SR003 |
| CR005 | Human Rights Watch documented Xinjiang policing and surveillance concerns that frame the company’s reputational risk. | Medium | SR004 |
| CR006 | Megvii’s controversy profile is persistent enough that it remains central to third-party summaries and profiles. | Medium | SR005, SR006 |
| CR007 | The stalled Hong Kong path and later STAR Market filing show that public-market access has been attempted more than once without success. | Medium | SR003, SR008 |
| CR008 | Because Megvii remains private, investors still lack audited ongoing public-company disclosures on revenue quality, burn, and customer concentration. | Medium | SR015, SR016 |
| CR009 | China’s PIPL treats biometric information as sensitive personal information. | Medium | SR009 |
| CR010 | PIPL imposes separate-consent and notice obligations that matter directly for facial-recognition deployment. | Medium | SR009 |
| CR011 | China’s 2025 face-recognition measures add more detailed operational constraints on facial-recognition use. | Medium | SR010, SR011 |
| CR012 | Those rules increase compliance burden for any public-place or identity-sensitive deployment strategy. | Medium | SR009, SR010, SR011 |
| CR013 | The EU AI Act enforcement framework shows Europe is moving toward strict oversight of high-risk biometric AI. | Medium | SR012, SR013 |
| CR014 | EDRi’s position reflects strong civil-liberties pressure against public facial recognition in Europe. | Medium | SR014 |
| CR015 | Even if Megvii’s current revenue base is concentrated in China, stricter foreign biometric rules still narrow long-term overseas expansion options. | Medium | SR012, SR013, SR014, SR018 |
| CR016 | The Entity List raises both supply-chain and customer-trust risk because access to U.S. technology and acceptance by sensitive buyers can both be constrained. | Medium | SR001, SR002, SR005 |
| CR017 | Trust-sensitive international customers are harder to win when sanctions and surveillance allegations attach directly to the brand. | Medium | SR005, SR006, SR018 |
| CR018 | Megvii’s product breadth across smart city, warehouse, building, and developer surfaces does not eliminate the trust overhang created by sanctions and rights scrutiny. | Medium | SR019, SR020, SR021, SR022, SR023 |
| CR019 | Smart-city exposure likely increases policy sensitivity because public-sector and public-space use cases face the most scrutiny. | Medium | SR019, SR009, SR011 |
| CR020 | Warehouse and industrial products partly diversify demand away from the most controversial surveillance use cases. | Medium | SR020 |
| CR021 | Building-access and identity products still sit close enough to biometric compliance questions that regulation remains commercially relevant. | Medium | SR021, SR022, SR009 |
| CR022 | Developer and API exposure create additional privacy and misuse risk because identity functionality can be embedded by third parties. | Medium | SR022 |
| CR023 | Project-heavy AIoT deployments likely create execution and working-capital risk that a pure software model would not carry to the same extent. | Medium | SR019, SR020, SR021 |
| CR024 | The 2019 Megvii funding round proved the company could raise large private capital, but it does not by itself resolve present liquidity risk. | Medium | SR017, SR016 |
| CR025 | The 2025 domestic-investor round shows capital access is not shut, but terms remain opaque. | Medium | SR016 |
| CR026 | Visible customer and partner proofs show market access remains partly open in China despite sanctions. | Medium | SR018, SR024, SR025 |
| CR027 | Singapore remains the clearest public international proof point, which also highlights how thin the reviewed overseas evidence is. | Medium | SR018 |
| CR028 | Megvii’s biggest downside risk is not one isolated product flaw but the interaction of sanctions, privacy regulation, and trust-sensitive demand. | Medium | SR001, SR009, SR014 |
| CR029 | The least quantifiable risk from public evidence is customer-quality concentration because public sources do not disclose sector revenue mix or renewal depth. | Medium | SR015, SR016, SR025 |
| CR030 | No reviewed public source discloses current litigation, fine history, or active official audits in enough detail to close compliance diligence. | Medium | SR009, SR010, SR011 |
| CR031 | Megvii’s surveillance branding can depress exit optionality even if the company’s technology remains commercially useful. | Medium | SR002, SR003, SR006 |
| CR032 | The risk chapter is mitigated somewhat by product diversification and continued domestic commercial activity. | Medium | SR020, SR021, SR022, SR024, SR025 |
| CR033 | The risk chapter is worsened by the lack of current audited operating disclosures that would let investors price downside with confidence. | Medium | SR015, SR016, SR008 |
| CR034 | European and Chinese biometric rules matter even before large foreign scale is proven because they shape procurement confidence and future design constraints. | Medium | SR009, SR012, SR013, SR014 |
| CR035 | The combination of rights scrutiny and export controls makes Megvii a structurally higher-risk asset than a comparable AI company without surveillance exposure. | Medium | SR001, SR004, SR006 |
| CR036 | Megvii still appears investable only if an investor is comfortable underwriting policy, compliance, and trust risk as core variables rather than tail risks. | Medium | SR001, SR009, SR016, SR024 |
| CR037 | The public risk record supports a conservative view: sanctions and regulatory friction are durable constraints, not temporary headline noise. | Medium | SR001, SR003, SR010, SR012 |
| CR038 | Sanctions Finder independently reflects Megvii’s sanctions / entity-list status, showing the designation is visible in commercial compliance tooling as well as official notices. | Medium | SR026 |
| CR039 | Politico’s coverage of the AI Act debate shows opposition to public facial recognition has strong political backing in Europe beyond pure compliance administration. | Medium | SR027, SR014 |
| CR040 | DigiChina’s translation reinforces that PIPL explicitly treats biometric information as sensitive personal information, supporting the compliance reading used in this chapter. | Medium | SR028, SR009 |
| CR041 | Reuters reporting via AInvest shows the broader U.S. blacklist environment remains fluid, which means Megvii’s policy risk should be monitored as part of a wider export-control trajectory. | Medium | SR030 |
| CR042 | Additional translated or secondary legal references do not materially soften the chapter’s conclusion: biometric regulation is tightening, not easing, in key jurisdictions. | Medium | SR028, SR029, SR010, SR012 |
| CV001 | Tracxn says Megvii has raised $1.98 billion across 10 rounds. | Medium | SV001, SV002, SV003 |
| CV002 | The Company Check also reports $1.98 billion of total funding across 10 rounds. | Medium | SV003, SV002 |
| CV003 | Reuters reported that Megvii raised $750 million in May 2019. | Medium | SV005, SV006, SV007 |
| CV004 | Reuters reported that the 2019 round valued Megvii at slightly above $4 billion. | Medium | SV005, SV004 |
| CV005 | Tracxn records an undisclosed Series D round on April 8, 2025. | Medium | SV002 |
| CV006 | The 2025 round appears to preserve capital access but does not publicly reveal valuation or terms. | Medium | SV002, SV003 |
| CV007 | WOWLS still frames $4 billion in 2019 as the last known public valuation marker and suggests current value may be lower. | Low | SV004 |
| CV008 | The Hong Kong IPO setback weakened Megvii’s access to public-market price discovery. | Medium | SV008, SV009 |
| CV009 | The U.S. Entity List introduces a structural discount because it raises both technology-access and trust-sensitive demand risk. | Medium | SV009, SV010 |
| CV010 | Human-rights and surveillance controversy deepen that discount by limiting the set of comfortable investors and counterparties. | Medium | SV011, SV004 |
| CV011 | Market-report sources indicate that computer vision remains a large and growing category, preserving long-run optionality if Megvii can execute. | Medium | SV012, SV013 |
| CV012 | SenseTime is a relevant public China AI comparator because it monetizes computer-vision and broader AI software while still carrying loss-making growth characteristics. | Medium | SV014 |
| CV013 | Hikvision is a useful scale and hardware-channel comparator, but not a clean multiple comparator because it is a mature profitable AIoT incumbent. | Medium | SV015, SV029 |
| CV014 | Palantir is a useful trust-premium and disclosure-quality comparator, but not a product-like comparator, because it is profitable, transparent, and U.S.-listed. | Medium | SV017, SV018, SV019 |
| CV015 | Axis is a useful trusted physical-security comparator for buyer confidence and channel quality, even if it is less software-platform-centric than Megvii. | Medium | SV016 |
| CV016 | Megvii’s mix of Brain++, Face++, smart warehouse, smart city, and building software means it should not be valued as a single-product facial-recognition vendor. | Medium | SV021, SV022, SV023, SV024 |
| CV017 | That same business mix also makes pure SaaS multiples inappropriate because parts of the company look project-heavy or hardware-attached. | Medium | SV023, SV024, SV025 |
| CV018 | China Telecom, Singapore, and Koala deployments support the view that Megvii still has commercial relevance beyond research pedigree. | Medium | SV025, SV026, SV027 |
| CV019 | The last public $4B-plus marker is too stale to use without a major discount because listing failure, sanctions, and private opacity all intervene. | Medium | SV004, SV008, SV010 |
| CV020 | A scenario where Megvii only merits a sub-last-round valuation is easy to justify from public evidence alone. | Medium | SV004, SV008, SV011 |
| CV021 | A scenario where Megvii regains or exceeds the last public mark requires successful proof on audited growth, compliance, and exit optionality, none of which is public today. | Medium | SV002, SV008, SV028 |
| CV022 | The market can support sizable AI winners, but Megvii’s risk-adjusted value should sit below the headline opportunity implied by TAM reports. | Medium | SV012, SV013, SV010 |
| CV023 | SenseTime’s audited 2025 revenue and still-negative earnings show that public investors can value large China AI platforms even before full profitability, but only with full disclosure. | Medium | SV014 |
| CV024 | Hikvision’s scale and profitability show what a mature trusted AIoT/security platform looks like, underscoring how far Megvii is from a mature incumbent valuation profile. | Medium | SV015 |
| CV025 | Palantir’s public filing trail highlights the valuation premium that transparency, profitability, and U.S. market access can command versus Megvii’s opacity. | Medium | SV017, SV018, SV019 |
| CV026 | Axis is best treated as a trust and channel benchmark, not an economic benchmark, for Megvii. | Medium | SV016 |
| CV027 | Public evidence supports at least three valuation buckets for Megvii: sanctioned China AI platform, AIoT/vision workflow vendor, and developer-identity platform. | Medium | SV021, SV022, SV023, SV024 |
| CV028 | The sanctioned China AI platform bucket should command the heaviest discount. | Medium | SV010, SV011, SV014 |
| CV029 | The AIoT/vision workflow bucket supports value because it connects software, devices, and operations in physical-world use cases. | Medium | SV023, SV024, SV025 |
| CV030 | The developer-identity bucket supports some strategic option value because Face++ broadens distribution beyond named solution projects. | Medium | SV022 |
| CV031 | A precise public revenue multiple is not supportable because current revenue, margin, and recurring-revenue mix are not disclosed. | Medium | SV003, SV001, SV002 |
| CV032 | Private investors should assume both a liquidity discount and a disclosure discount relative to public comps. | Medium | SV008, SV017, SV018 |
| CV033 | Public-comp quality differences matter: a transparent profitable company like Palantir can trade on economics, while Megvii must still trade on narrative and optionality. | Medium | SV017, SV018, SV019 |
| CV034 | A plausible public-evidence valuation range for Megvii is roughly $2.0B to $4.5B, with the center of gravity below the stale 2019 anchor. | Medium | SV004, SV008, SV010, SV016 |
| CV035 | A downside case near $1.5B to $2.5B is supportable if investors treat sanctions, opacity, and delayed exit as dominant variables. | Medium | SV004, SV008, SV011 |
| CV036 | An upside case near $4.0B to $6.0B would require believing that commercial breadth, domestic capital support, and future disclosure can outweigh the policy discount. | Low | SV002, SV021, SV025 |
| CV037 | A strategic-acquirer outcome is conceptually possible through industrial or domestic ecosystem buyers, but no reviewed source makes such a path concrete. | Low | SV020, SV025, SV026 |
| CV038 | The most defensible public valuation output is a scenario range, not a point estimate. | Medium | SV003, SV004, SV014 |
| CV039 | Megvii’s valuation ceiling is capped more by trust and exit constraints than by lack of market opportunity. | Medium | SV010, SV011, SV012, SV013 |
| CV040 | Megvii’s valuation floor is supported by substantial historical capital raised, ongoing domestic commercial proof, and persistent AI market demand. | Medium | SV001, SV025, SV027, SV012 |
| CV041 | The single largest blocker to tighter valuation is absence of audited 2025 or 2026 operating disclosure. | Medium | SV001, SV002, SV003 |
| CV042 | The valuation chapter should therefore be read as a disciplined range around stale anchors, not as a precision-mark exercise. | Medium | SV004, SV028, SV014 |