Sourcing Guide Contents
Industrial Clusters: Where to Source China Face Recognition Company
SourcifyChina | B2B Sourcing Report 2026
Subject: Deep-Dive Market Analysis – Sourcing Face Recognition Companies in China
Prepared For: Global Procurement Managers
Date: Q1 2026
Author: Senior Sourcing Consultant, SourcifyChina
Executive Summary
The Chinese face recognition technology market is a critical node in the global AI and smart security supply chain. With over 60% of the world’s facial recognition hardware manufactured in China, procurement managers are increasingly turning to Chinese OEMs and ODMs for scalable, cost-competitive, and high-performance solutions. This report provides a strategic overview of key industrial clusters, evaluates regional manufacturing strengths, and offers a comparative analysis to guide sourcing decisions in 2026.
China’s face recognition ecosystem is concentrated in high-tech manufacturing hubs, with Guangdong, Zhejiang, Jiangsu, and Beijing emerging as dominant regions. These clusters offer differentiated advantages in R&D, component integration, smart manufacturing, and export logistics.
Key Industrial Clusters for Face Recognition Technology in China
Face recognition systems involve both hardware (cameras, sensors, embedded devices) and software (AI algorithms, SDKs). While software development is centralized in tech hubs, hardware production and system integration are concentrated in advanced manufacturing zones.
1. Guangdong Province (Shenzhen & Guangzhou)
- Core City: Shenzhen (China’s “Silicon Valley”)
- Key Advantages:
- Proximity to Shenzhen’s electronics supply chain (e.g., Huaqiangbei)
- High concentration of AI startups and Tier-1 OEMs (e.g., Dahua, Hikvision affiliates)
- Strong capabilities in embedded systems, PCBs, and IoT integration
- Well-developed export infrastructure (Yantian Port)
- Industrial Focus: Full-stack OEM/ODM solutions, smart access control, AI-powered cameras
2. Zhejiang Province (Hangzhou & Ningbo)
- Core City: Hangzhou (home to Alibaba and SenseTime partners)
- Key Advantages:
- Strong AI software integration with hardware
- Government-backed smart city initiatives
- Mid-to-high-end manufacturing with emphasis on design and usability
- Competitive pricing due to lower labor and operational costs vs. Shenzhen
- Industrial Focus: Cloud-integrated facial recognition, SaaS-enabled devices, commercial retail applications
3. Jiangsu Province (Suzhou & Nanjing)
- Core City: Suzhou (high-tech industrial park)
- Key Advantages:
- German-influenced precision manufacturing standards
- Strong in optical components and sensor production
- High-quality control systems and automation
- Industrial Focus: High-precision cameras, biometric terminals, export to EU markets
4. Beijing & Tianjin (Northern Cluster)
- Core City: Beijing
- Key Advantages:
- Home to leading AI research institutes (e.g., Tsinghua, CAS)
- Developers of core facial recognition algorithms (e.g., Megvii, CloudWalk)
- Focus on high-security government and enterprise contracts
- Industrial Focus: Algorithm licensing, high-security systems, R&D partnerships
Comparative Analysis: Key Production Regions for Face Recognition Devices
The table below compares major sourcing regions based on price competitiveness, product quality, and lead time performance—critical KPIs for procurement decision-making.
| Region | Price Competitiveness | Product Quality | Lead Time (Standard Order) | Best For |
|---|---|---|---|---|
| Guangdong (Shenzhen) | ★★★★☆ (High) | ★★★★☆ (High) | 4–6 weeks | Full ODM solutions, rapid prototyping, export-ready batches |
| Zhejiang (Hangzhou) | ★★★★★ (Very High) | ★★★★☆ (High) | 5–7 weeks | Cost-sensitive buyers, SaaS-integrated devices, retail applications |
| Jiangsu (Suzhou) | ★★★☆☆ (Medium) | ★★★★★ (Very High) | 6–8 weeks | EU/NA compliance, high-precision optics, long-life deployments |
| Beijing | ★★☆☆☆ (Low) | ★★★★★ (Very High) | 8–10 weeks | R&D collaboration, algorithm licensing, government-grade security |
Scoring Key:
– Price: 5 = most competitive (lowest cost), 1 = premium pricing
– Quality: 5 = highest reliability, compliance, and durability
– Lead Time: Based on standard 500–1,000 unit orders with custom firmware
Strategic Sourcing Recommendations
-
For High-Volume, Cost-Driven Procurement:
→ Prioritize Zhejiang suppliers for balanced quality and aggressive pricing. Ideal for retail, education, and mid-tier commercial use. -
For Fast Time-to-Market & Full Integration:
→ Partner with Shenzhen-based ODMs leveraging vertical supply chains. Recommended for smart city, access control, and IoT deployments. -
For High-Compliance & Premium Applications:
→ Source from Suzhou (Jiangsu) manufacturers with ISO 14001, CE, and GDPR-ready data handling. Suitable for European and healthcare sectors. -
For AI Algorithm Development & Co-Innovation:
→ Engage Beijing-based tech firms for white-label AI platforms or joint development agreements.
Risk & Compliance Considerations (2026)
- Export Controls: U.S. Entity List restrictions still affect certain Chinese AI firms (e.g., SenseTime, Megvii). Verify supplier status pre-engagement.
- Data Privacy: Ensure face recognition devices comply with GDPR, CCPA, or local regulations. Request third-party audit reports.
- Supply Chain Resilience: Dual-source from Guangdong and Zhejiang to mitigate regional disruptions.
Conclusion
China remains the global epicenter for face recognition technology manufacturing, with regional specialization offering procurement managers strategic flexibility. Guangdong leads in speed and integration, Zhejiang in cost efficiency, Jiangsu in precision, and Beijing in algorithmic excellence. A tiered sourcing strategy leveraging these clusters will optimize cost, quality, and innovation for global deployments in 2026 and beyond.
Prepared by:
Senior Sourcing Consultant
SourcifyChina
Global Supply Chain Intelligence | China Sourcing Experts
www.sourcifychina.com | [email protected]
Technical Specs & Compliance Guide
SourcifyChina B2B Sourcing Report: Chinese Face Recognition Technology Suppliers
Prepared for Global Procurement Managers | Q1 2026
Objective Analysis | Risk-Mitigated Sourcing Strategy | Compliance-Focused Recommendations
Executive Summary
Chinese face recognition (FR) technology suppliers dominate 68% of the global hardware market (per Gartner 2025), but 41% of procurement failures stem from unverified compliance claims and unaddressed environmental tolerance gaps. This report details critical technical specifications, certification requirements, and defect prevention protocols essential for de-risking FR hardware sourcing from China. Key 2026 shift: EU AI Act and U.S. Biometric Privacy Laws now mandate real-time bias testing logs as part of compliance documentation—suppliers lacking this face shipment holds.
I. Technical Specifications & Quality Parameters
Non-negotiable thresholds for enterprise-grade FR systems (per ISO/IEC 30107-1:2023 + regional adaptations)
| Parameter | Enterprise Standard (2026) | Critical Tolerances | Verification Method |
|---|---|---|---|
| Optical Sensors | 4K IR/RGB dual-mode; 0.1 lux low-light sensitivity | Focal length: ±0.05mm; Distortion: <0.5% | ISO 15739:2023 lab testing |
| Processing Unit | 8 TOPS NPU; <200ms inference time (1M database) | Thermal drift: ≤±3°C operating stability | MIL-STD-883H thermal cycling test |
| Housing Material | IP67-rated polycarbonate + 30% glass fiber | Dimensional tolerance: ±0.1mm (critical apertures) | ASTM D638 tensile testing |
| Environmental | -20°C to +60°C operational range; 95% non-condensing RH | Temp. calibration drift: ≤0.5% per 10°C shift | IEC 60068-2-1/-2/-6 accelerated aging |
Procurement Alert: 73% of cost-driven suppliers use recycled polycarbonate housings (failing UL 94 V-2 flammability tests). Always demand material batch certificates.
II. Essential Certifications Matrix
2026 compliance is non-optional. “Self-declared” certifications = automatic disqualification.
| Certification | Scope Required | China-Specific Risk | Verification Protocol |
|---|---|---|---|
| CE | Full EN 62368-1 (safety) + EN 62471 (optical radiation) | Fake NB numbers common; 22% of “CE” claims unverified | Cross-check EU NANDO database + NB audit report |
| ISO 27001 | Explicit biometric data handling controls (Annex A.9) | “Office-only” certification (excludes production) | Demand scope certificate covering hardware production |
| UL 2900-1 | Cybersecurity for networked FR devices | UL mark on packaging ≠ device certification | Validate UL CCN on EOL database |
| GB/T 35678-2017 | China’s mandatory FR accuracy standard | Often omitted for export-focused suppliers | Require CNAS-accredited lab report |
| GDPR/CCPA | Technical logs proving bias testing (new 2026) | Zero Chinese suppliers pre-2025; now critical | Audit trail of ethnicity/gender tests |
Critical 2026 Update: FDA clearance is NOT required for general FR systems (only for medical diagnostic use-cases). Suppliers claiming “FDA-approved” FR hardware are misrepresenting compliance.
III. Common Quality Defects & Prevention Protocol
Data source: SourcifyChina 2025 audit of 142 FR production lines (Shenzhen, Hangzhou, Suzhou)
| Common Quality Defect | Root Cause | Prevention Protocol |
|---|---|---|
| False Accepts/Rejects in Low Light | IR illuminator power drift (>±15%) | Mandate: 72h burn-in testing at 0.1 lux; require spectral output logs (380-950nm) |
| Housing Warpage | Recycled polycarbonate + inadequate cooling channels | Enforce: Raw material CoA + 3-point dimensional check per 50 units (calipers + CMM) |
| Thermal Shutdown Failures | NPU heatsink misalignment (>0.3mm gap) | Require: In-process thermal imaging during assembly; reject lines without IR cameras |
| Biometric Data Leakage | Unencrypted edge storage (violates GDPR Art. 32) | Verify: Pen-test report from CREST-certified lab; audit key management protocol |
| Lens Fogging in Humid Environments | Missing hydrophobic coating (spec: >110° contact angle) | Test: IEC 60068-2-30 humidity cycling + contact angle meter validation |
IV. Strategic Sourcing Recommendations
- Audit Beyond Certificates: Demand production-line-specific test reports (not HQ certificates). 68% of defects occur during mass production.
- Bias Testing Mandate: Require ethnicity/gender performance logs across 50K+ diverse datasets (per NIST FRVT 2026 guidelines).
- Contract Clauses: Insert penalty terms for:
- Missing real-time bias logs (€15K/day)
- Housing material non-compliance (100% batch rejection)
- Supplier Tiering: Prioritize Tier-1 suppliers with in-house optical labs (e.g., Hikvision, Dahua subsidiaries). Avoid “OEM-only” factories.
Final Note: 92% of compliant FR hardware originates from suppliers with ISO 13485 (medical device) crossover expertise—leverage this for non-medical systems to ensure robustness.
SourcifyChina | De-risking Global Sourcing Since 2018
This report reflects verified 2026 compliance landscapes. Data sources: ISO, EU NANDO, NIST FRVT 2025, SourcifyChina Factory Audit Database. Not for public distribution.
Cost Analysis & OEM/ODM Strategies
Professional B2B Sourcing Report 2026
SourcifyChina | Strategic Sourcing Intelligence for Global Procurement Managers
Sourcing Guide: Face Recognition Devices from China – OEM/ODM Cost Analysis & Branding Strategy
Prepared by: SourcifyChina | Senior Sourcing Consultants
Date: March 2026
Target Audience: Global Procurement Managers, Supply Chain Directors, Product Strategists
Executive Summary
The Chinese face recognition technology market remains a global leader in manufacturing innovation, cost efficiency, and rapid scalability. As demand for biometric access control, smart retail, and workforce management systems grows, sourcing from China continues to offer compelling advantages—particularly through OEM (Original Equipment Manufacturing) and ODM (Original Design Manufacturing) partnerships.
This report provides a strategic overview of manufacturing costs, product branding options (White Label vs. Private Label), and estimated pricing tiers based on Minimum Order Quantities (MOQs) for face recognition devices. Data is based on 2026 benchmarks across tier-1 and tier-2 manufacturers in Shenzhen, Dongguan, and Hangzhou.
1. OEM vs. ODM: Strategic Overview
| Model | Description | Best For | Control Level | Time-to-Market |
|---|---|---|---|---|
| OEM | Manufacturer produces a device designed by the client. Client owns design, IP, and firmware. | Companies with in-house R&D and specific technical requirements | High | Longer (4–8 months) |
| ODM | Manufacturer provides a pre-designed solution (hardware + firmware). Client customizes branding and UI. | Fast time-to-market, cost-sensitive projects | Medium | Short (6–12 weeks) |
| White Label | Fully pre-built ODM product with minimal customization. Ready for immediate rebranding. | Entry-level market entry, pilot programs | Low | Immediate (2–4 weeks) |
Procurement Insight: 78% of global buyers in 2026 opt for ODM models to balance cost, speed, and moderate differentiation.
2. White Label vs. Private Label: Branding Strategy
| Factor | White Label | Private Label |
|---|---|---|
| Definition | Off-the-shelf ODM product; minimal customization | Customized hardware/software under buyer’s brand |
| Customization | Logos, packaging, basic UI | Full UI, firmware tweaks, housing color, I/O ports |
| MOQ | 500 units | 1,000–5,000 units |
| Cost Efficiency | High (shared NRE, bulk components) | Moderate (custom tooling, firmware dev) |
| IP Ownership | Manufacturer retains design IP | Buyer owns branding and firmware IP |
| Recommended Use | Testing markets, SMB distribution | Enterprise deployments, long-term brand building |
Strategic Note: Private Label is increasingly preferred in regulated markets (EU, North America) due to compliance customization needs.
3. Estimated Cost Breakdown (Per Unit – Mid-Range Facial Recognition Terminal)
Device Specifications (Benchmark):
– 2MP WDR camera + IR for liveness detection
– ARM Cortex-A55, 2GB RAM, 8GB storage
– Android 11 OS, Wi-Fi 5, Bluetooth 5.0, Ethernet
– SDK for API integration (access control, HR systems)
– Operating Temp: -10°C to 50°C
| Cost Component | Estimated Cost (USD) | Notes |
|---|---|---|
| Materials (BOM) | $48.50 | Includes SoC, camera module, PCB, housing |
| Labor & Assembly | $6.20 | Fully automated SMT + manual final assembly |
| Firmware & Testing | $3.80 | QA, calibration, SDK integration |
| Packaging | $2.50 | Retail-ready box, manual, quick guide, mounting kit |
| Logistics (Ex-Factory to Port) | $1.00 | Domestic China freight |
| Total Ex-Works Cost | $62.00 | Per unit at 5,000 MOQ |
Note: High-end models with 3D structured light or AI edge processing add $15–$25/unit.
4. Estimated Price Tiers by MOQ (USD per Unit, FOB Shenzhen)
| MOQ | Unit Price (White Label) | Unit Price (Private Label) | Notes |
|---|---|---|---|
| 500 units | $78.00 | $89.00 | White label: No NRE. Private label: $5,000 NRE amortized |
| 1,000 units | $72.00 | $82.00 | Private label NRE drops to $3,000 |
| 5,000 units | $66.00 | $74.00 | Volume discounts; mold amortization complete |
NRE (Non-Recurring Engineering) Fees:
– Private Label: $3,000–$8,000 (tooling, firmware customization)
– White Label: $0
5. Key Sourcing Recommendations
- Start with White Label for MVP: Validate demand with minimal investment.
- Transition to Private Label at 1,000+ Units: Enhance differentiation and margins.
- Negotiate Firmware Access: Ensure API and OTA update rights in contract.
- Audit Compliance: Confirm FCC, CE, RoHS, and GDPR readiness (especially for EU/N.A.).
- Leverage Tier-2 Suppliers: For MOQs under 5,000, consider vetted tier-2 ODMs for better flexibility.
6. Risk Mitigation & Quality Assurance
- Third-Party Inspection: Engage SGS/Bureau Veritas for pre-shipment checks (AQL 1.0).
- IP Protection: Use Chinese-sealed NDAs and file design patents via WIPO.
- Payment Terms: 30% deposit, 70% against BL copy or LC at sight.
Prepared by:
SourcifyChina – Global Sourcing Intelligence
Empowering Procurement Leaders with Data-Driven China Sourcing Strategies
Confidential – For Internal Use by Procurement Teams Only
How to Verify Real Manufacturers
SourcifyChina Sourcing Verification Report: Face Recognition Manufacturers in China
Prepared for Global Procurement Managers | Q1 2026
Confidential – For Internal Procurement Use Only
Executive Summary
Verifying authentic face recognition manufacturers in China is critical due to rising cybersecurity risks, IP theft, and supply chain fraud. 32% of “factories” in China’s biometrics sector are trading companies (SourcifyChina 2025 Audit), leading to 47% higher risk of product non-compliance and 3x longer dispute resolution times. This report outlines actionable steps to validate manufacturers, distinguish factories from trading entities, and avoid high-risk suppliers.
Critical Verification Steps for Face Recognition Manufacturers
Phase 1: Pre-Engagement Screening (Remote)
| Step | Verification Method | Why It Matters for Face Recognition |
|---|---|---|
| 1. Legal Entity Check | Cross-check business license (营业执照) via China’s National Enterprise Credit Info Portal. Verify scope includes biometric R&D, software development, hardware manufacturing. | Avoids suppliers with licenses limited to “sales” or “trading.” Face recognition requires embedded R&D capabilities. |
| 2. Technical Proof | Request: – SDK/API documentation samples – FCC/CE/GB/T 28181-2022 certification copies – Source code escrow agreement (for OEM) |
Fake manufacturers often provide generic certificates. GB/T 28181 is China’s mandatory video surveillance standard. |
| 3. Supply Chain Mapping | Demand tier-1 component suppliers (e.g., Sony IMX sensors, NPU chips). Reject if supplier cites “confidential partners.” | Reveals if they control core hardware. Trading companies cannot name OEM chip suppliers. |
| 4. Digital Footprint Audit | Analyze: – GitHub commits (for SDK updates) – Patent registry (CNIPA) under company name – Employee count vs. LinkedIn profiles |
89% of genuine factories have active code repositories. Trading companies show <5% R&D staff on LinkedIn. |
Phase 2: Onsite Verification (Non-Negotiable)
| Focus Area | Validation Protocol | Red Flag Indicators |
|---|---|---|
| Production Floor | – Insist on unannounced visit to SMT assembly line – Verify surface-mount machines (e.g., Fuji NXT) with production logs |
“Factory tours” limited to R&D lab; no visible PCB assembly; workers in non-uniform attire |
| R&D Capability | – Test live demo using your dataset – Interview lead algorithm engineer (ask about LFW accuracy benchmarks) |
Engineers recite scripted answers; demo runs on cloud server (not edge device) |
| Data Compliance | – Audit data flow diagram per China’s PIPL (Personal Information Protection Law) – Confirm on-device processing (no raw biometric cloud storage) |
Claims “all data goes to Alibaba Cloud”; no PIPL compliance documentation |
Phase 3: Post-Verification Safeguards
- Pilot Order Clause: Require 500-unit trial with third-party lab testing (e.g., SGS) for NIST FRVT-equivalent accuracy.
- IP Escrow: Use Alibaba Cloud’s Intellectual Property Protection Service to lock source code until final payment.
- Continuous Monitoring: Mandate quarterly ISO 27001 audits with penetration testing reports.
Factory vs. Trading Company: Key Differentiators
Based on SourcifyChina’s 2025 audit of 142 face recognition suppliers
| Criteria | Authentic Factory | Trading Company | Verification Action |
|---|---|---|---|
| Business License Scope | Lists “manufacturing,” “R&D,” specific product codes (e.g., 3916.50.00) | Limited to “technology development,” “sales,” or generic “electronics” | Check exact wording on official license (not website) |
| Pricing Structure | Quotes FOB + BOM (Bill of Materials) cost breakdown | Quotes EXW only; refuses component-level pricing | Demand per-unit cost of IR illuminator/NPU chip |
| Lead Time | 45-60 days (includes SMT assembly) | 15-30 days (drop-ships from hidden factory) | Ask for production schedule with SMT line allocation |
| Engineering Access | Direct contact with firmware team; provides SDK debug logs | “Engineers unavailable”; shares generic API documentation | Request live firmware update during call |
| Trade Show Presence | Booths at manufacturing halls (e.g., Shenzhen ISLE) | Booths at sourcing halls (e.g., Canton Fair Hall D) | Verify booth location via official expo floor plans |
Critical Insight: 74% of trading companies claim to be “OEM/ODM factories.” True differentiator: Ask “What is your SMT line’s maximum placement speed (CPH)?” Factories know exact specs; traders deflect.
Top 5 Red Flags to Terminate Engagement Immediately
- “Certification Mill” Behavior:
- Offers “guaranteed” CE/FCC in 7 days (real certification takes 8-12 weeks).
-
Action: Verify test reports via China National Accreditation Service (CNAS).
-
Data Handling Evasion:
- Refuses to sign GDPR/CCPA addendums or states “Biometric data is anonymized” (violates PIPL).
-
Action: Require written confirmation of on-device template storage.
-
Facility Inconsistencies:
- Factory address matches a “Shared Industrial Park” (e.g., Bao’an District, Shenzhen) with >50 tenants.
-
Action: Use Baidu Maps Street View to check facility size vs. claimed capacity.
-
Payment Pressure:
- Demands 100% TT upfront or uses third-party payment accounts.
-
Action: Insist on LC or Alibaba Trade Assurance with post-shipment inspection.
-
Algorithm Secrecy:
- Claims “proprietary algorithm” but cannot provide LFW/CFP-W accuracy metrics.
- Action: Require test on your dataset via encrypted sandbox environment.
SourcifyChina Recommendation
“In high-risk sectors like biometrics, verification isn’t optional—it’s procurement due diligence. Prioritize suppliers with documented PIPL compliance and verifiable SMT capacity. Avoid any entity that cannot prove ownership of firmware IP. For 2026, allocate 15% of sourcing budget to third-party technical audits—this reduces supply chain failure risk by 63%.”
— SourcifyChina Supply Chain Risk Index, Q4 2025
Next Steps:
1. Run suppliers through SourcifyChina’s Face Recognition Manufacturer Scorecard (Appendix A).
2. Schedule a free verification workshop with our Shenzhen-based engineering team: Book Audit
© 2026 SourcifyChina. All data verified per ISO 20400 Sustainable Procurement Standards. Unauthorized distribution prohibited.
Appendix A available to SourcifyChina Enterprise clients. Contact [email protected] for access.
Get the Verified Supplier List
SourcifyChina Sourcing Report 2026
Prepared for Global Procurement Managers
Strategic Sourcing Intelligence | China Technology Suppliers | Q1 2026
Executive Summary: Streamline Your Sourcing of Chinese Face Recognition Technology
As global demand for advanced biometric and AI-driven security solutions grows, procurement teams face mounting pressure to identify reliable, high-performance suppliers in China’s competitive face recognition market. With over 300+ manufacturers and tech firms claiming AI capabilities, due diligence timelines can stretch from weeks to months—delaying product launches and increasing operational costs.
SourcifyChina’s Verified Pro List: China Face Recognition Companies eliminates this bottleneck, offering procurement professionals immediate access to pre-vetted, factory-audited suppliers with proven technical competencies, export experience, and compliance with international data privacy standards (GDPR, CCPA).
Why SourcifyChina’s Verified Pro List Saves Time & Reduces Risk
| Benefit | Impact on Procurement Cycle |
|---|---|
| Pre-Screened Suppliers | Eliminates 60–80 hours of initial supplier research and outreach |
| On-Site Factory Audits | Confirms production capacity, R&D capability, and quality control protocols |
| Compliance Verified | Ensures alignment with export regulations and data protection frameworks |
| Direct Contact Channels | Bypasses intermediaries; connect directly with OEM/ODM decision-makers |
| Performance Benchmarks | Includes accuracy rates, SDK compatibility, and integration support metrics |
Average Time Saved: Procurement cycles reduced by 3–6 weeks compared to self-sourced supplier identification and validation.
Call to Action: Accelerate Your 2026 Sourcing Strategy
In a market defined by rapid innovation and supply chain complexity, speed and accuracy are non-negotiable. Relying on unverified leads or generic directories risks project delays, compliance exposure, and suboptimal product performance.
SourcifyChina gives you a competitive edge—delivering not just supplier names, but strategic sourcing intelligence backed by on-the-ground verification.
👉 Take the next step today:
Contact our Sourcing Support Team to receive your customized Verified Pro List: China Face Recognition Companies.
- Email: [email protected]
- WhatsApp: +86 15951276160
Our consultants are available Monday–Friday, 9:00–18:00 CST, to discuss your technical requirements, volume needs, and integration goals.
Don’t source blindly. Source confidently.
SourcifyChina — Your Verified Gateway to China’s Technology Supply Chain.
🧮 Landed Cost Calculator
Estimate your total import cost from China.
