Investments

First published

RMB3.26 billion of orders shipped through ZKH. What did customers pay for AI?

Our interpretation

Finding the right spare part can keep a factory running. ZKH already earns revenue selling industrial supplies and services. Its latest results show a much smaller operating profit than its order volume suggests, and do not show what customers paid for AI part identification.

The Signal in brief

What changed

Third-quarter shipped order value rose 24.1%, net revenue 18.2% and net service revenue 33.3%. The financials are unaudited; an increase in service revenue does not isolate AI fees.

Why it matters

A wrong replacement part costs time and can force a return. Confirming a match may be worth paying for, but making a product sale more efficient is not the same as selling a separate service.

Who it affects

Factory maintenance teams, industrial distributors and readers following ZKH’s NYSE-listed securities.

Where

China; NYSE-listed ZKH

What to watch

Compare the next quarterly service mix, operating margin, cash from operations and returns. Look for a customer-approved pilot, repeat orders and documented payment; one to two quarters is a research window, not company guidance.

AI-specific fees, independently validated matching accuracy, customer-level cash receipts, current valuation and market mispricing have not been established.

Explore this signal

Emerging data

A maintenance team needs a replacement that fits, not a catalogue with the longest list. A mismatch can leave a machine idle while the wrong component makes a second trip.

In its 18 November release for the quarter ended 30 September, ZKH reported RMB3.256 billion of gross merchandise value, defined as shipped order value excluding taxes and net of returns. It recognized RMB2.753 billion of net revenue. Neither figure identifies payments from individual customers.

Gross profit reached RMB478.5 million, or 17.4% of revenue. After fulfilment, staffing and other operating expenses, operating profit was RMB25.1 million, about 0.9% of revenue. Operating cash flow was RMB124.5 million. These measures answer different questions.

Operating income is not net profit: ZKH reported RMB42.4 million of net profit, or 1.5% of sales. Operating cash of RMB124.5 million covered the whole company; the report does not connect it to an individual AI customer.

ZKH counted 78,174 transacting customers, up 10.4% from 70,800. That is the number active during each quarter, not a count of newly acquired customers or AI users.

Net service revenue rose from RMB43.0 million to RMB57.3 million. The accounts do not isolate AI revenue. Management reported an estimated 43,000 hours of company-wide AI productivity gains, not measured savings attributable to the parts model alone.

If validated part matching becomes a paid service, service mix could grow without matching inventory expansion. This remains unproven.

Earlier ZKH reports put this quarter in perspective: the company first crossed into operating profit in Q2 at a 0.2% margin. A second profitable quarter does not establish a durable margin or the source of the improvement.

The wider factory picture is mixed. In October, after ZKH’s quarter, China’s national purchasing managers survey put manufacturing new orders at 49.4, below the 50 threshold. Manufacturing value added among larger enterprises still rose 5.4% year on year. These cover different populations, periods and measures; neither identifies a ZKH buyer.

ZKH’s 2 November product announcement describes a receiving worker checking parts against delivery documents. That is not a certification that a part fits or is safe to install. In its results release, management also reported customer projects using other AI agents; this does not establish a paid LingLong Insight fee.

Report this passage

Orders, revenue and cash answer different questions
Orders, revenue and cash answer different questions
Q3 2026 measureRMB millionWhat it establishes
Shipped order value3255.8GMV net of returns, excluding taxes
Net revenue2752.9Recognised sales, not proof of settlement
Operating income25.1Profit from operations
Net profit42.4Profit including non-operating items
Operating cash flow124.5Company-wide operating cash
These are not successive stages of a single customer's payment.
Sources

The connection

Cataloguing, technical validation, delivery and accepted invoicing are separate steps. Faster matching may improve fulfilment without raising service fees.

Report this passage

The opportunity

With written permission, compare 20 anonymised parts matches against an experienced technician. Record errors, elapsed time, returns and quotes. Do not expose confidential drawings or substitute software for safety-critical approval.

What could happen next

  1. First-order consequence

    A manufacturer can compare verified matching accuracy, time and errors against its established workflow.

  2. Second-order consequence

    Accurate matches might mean fewer wrong-part orders.

  3. Third-order consequence

    Fewer returns might reduce warehouse handling and transport.

  4. Fourth-order consequence

    Reliable part records might help suppliers win repeat orders. No contracts were verified.

  5. Fifth-order consequence

    Less downtime might support factory output. No production gain across the chain was measured.

Costs, benefits and assumptions

Hypothetical pilot: RMB2,000 fee less RMB1,050 labour, RMB250 materials and RMB100 travel leaves RMB600 before tax and overhead. RMB1,400 is at risk before payment. No actual customer, fee or contract was verified.

Service revenue grew; AI fees were not broken outRMB millions
Horizontal scale 0 to 60Ticks 0 · 10 · 20 · 30 · 40 · 50 · 60
View chart values
Service revenue grew; AI fees were not broken out
QuarterRevenue
Q3 202543.0
Q3 202657.3
Reported service revenue is not identified as paid LingLong fees.
Sources

What would change our view?

Thin operating margin, errors and returns could outweigh efficiency. A shrinking service mix or weaker operating cash would challenge the thesis.

Alternative explanation

Workflow efficiencies may reduce internal processing cost without generating any new paid service.

What would confirm it

A paid customer pilot, accepted recurring matches, actual settlement and a disclosed contribution to service income would strengthen the case.

What would disprove it

Repeated errors, no accepted paid orders, or weak service conversion despite growing shipped order value undermine the commercial interpretation.

What would prove customers pay for part matching?
  1. ScopeApprove a written test
  2. AccuracyCompare matches with a technician baseline
  3. OrderVerify a paid order and acceptance
  4. SettlementMatch invoice to payment and check repeat demand
Conditional evidence path, not an observed transaction.
Sources

Evidence and sources

  • ZKH third-quarter 2026 unaudited results

    published date precision: day

    Source publication date 2026-11-18

    Coverage and independence

    Issuer primary financial statements.

    View source
  • ZKH Form 6-K Exhibit 99.1

    published date precision: day

    Source publication date 2026-11-18

    Coverage and independence

    Same company filing; not independent corroboration.

    View source
  • ZKH results distributed via PR Newswire

    published date precision: day

    Source publication date 2026-11-18

    Coverage and independence

    Issuer distribution at 06:00 EST, not separate confirmation.

    View source
  • OnTheRice Learning

    Coverage and independence

    Historic method only; not ZKH evidence.

    View source
  • ZKH Q2 2026 unaudited accounts, Form 6-K Exhibit 99.1

    Publisher: ZKH Group Limited

    published date precision: day

    Source publication date 2026-08-21

    Coverage and independence

    The first operating profit came in Q2 at a 0.2% operating margin; same issuer, different quarter.

    View source
  • ZKH Q3 2025 unaudited financial results

    Publisher: ZKH Group Limited

    published date precision: day

    Source publication date 2025-11-20

    Coverage and independence

    Original prior-year quarterly comparison; not independent of ZKH.

    View source
  • ZKH fourth quarter and fiscal year 2025 results

    Publisher: ZKH Group Limited

    published date precision: day

    Source publication date 2026-03-19

    Coverage and independence

    Historical platform mix: marketplace accounted for 12.9% of full-year 2025 GMV. It does not establish paid AI services.

    View source
  • ZKH fiscal 2025 Form 20-F

    Publisher: ZKH Group Limited

    published date precision: day

    Source publication date 2026-04-24

    Coverage and independence

    Annual company filing. Accounts receivable and revenue are distinct from collected cash; no Q3 AI-customer settlement follows.

    View source
  • Purchasing Managers’ Index for October 2026

    Publisher: National Bureau of Statistics of China

    published date precision: day

    Source publication date 2026-11-01

    Coverage and independence

    NBS national survey: October manufacturing PMI 49.3, new orders 49.4. This is neither a ZKH order count nor a quarter-on-quarter sales comparison.

    View source
  • Industrial Production Operation in October 2026

    Publisher: National Bureau of Statistics of China

    published date precision: day

    Source publication date 2026-11-17

    Coverage and independence

    NBS reported year-on-year manufacturing value-added growth of 5.4% among enterprises above designated size. Same statistical authority as PMI, different measure; not evidence of AI demand.

    View source
  • ZKH and Intel Bring Industrial Multimodal AI to the Edge with LingLong Insight 1.0

    Publisher: ZKH Group Limited

    published date precision: day

    Source publication date 2026-11-02

    Coverage and independence

    Earlier issuer product announcement describes receiving teams checking product details against delivery records. It does not certify machine fit or show a paid fee.

    View source

Gaps in the evidence

The latest release, SEC exhibit and wire duplicate one ZKH account. Earlier quarterly and annual financial records are distinct dated documents from the same issuer. Both national manufacturing measures come from China’s statistics bureau. Neither source family verifies a paid AI contract, independent accuracy, an executable share price or market mispricing. OTR Learning is undated methodological context only, not counted among dated records.

The supplied research cutoff was approximately 01:00 SGT on 19 November 2026; its exact forecast issuance time was not recorded. The original issuer release was distributed at 19:00 SGT on 18 November. The site publication time is separate.

Explore the research

Research layers

Research entries

  1. Layer name

    01 Money Flow

    Research and limitations

    Reported GMV RMB3.256bn is shipped platform order value, not recognised revenue; net revenue RMB2.753bn; operating cash generated RMB124.5m in Q3. Source: unaudited issuer statement.

  2. Layer name

    02 Early Orders / Deposits

    Research and limitations

    78,174 transacting customers and shipped GMV are reported, not future deposits. No buyer-funded AI-verification order or cash deposit independently verified.

  3. Layer name

    03 Bottlenecks

    Research and limitations

    Product identification mistakes, inventory availability and last-mile delivery may constrain value. Specific observed delays or rejection rates NOT VERIFIED.

  4. Layer name

    04 Second / Third-Tier Beneficiaries

    Research and limitations

    Potential beneficiaries: industrial spare-part cataloguers, authorised quality testers and distributors. Named downstream paid awards NOT VERIFIED.

  5. Layer name

    05 Listed Company Finances And Quote

    Research and limitations

    NYSE ZKH: Q3 revenue RMB2.753bn; 17.4% gross and 0.9% operating margin; reported cash/investments RMB1.70bn versus RMB1.92bn end-2025. Debt maturity, customer concentration, valuation and current exchange quote/time NOT VERIFIED.

  6. Layer name

    06 Transparent Customer Unit Economics

    Research and limitations

    ILLUSTRATIVE: RMB2,000 paid job less RMB1,400 direct spend = RMB600 before overhead. This is not ZKH pricing, client profit or an observed order.

  7. Layer name

    07 Insider And Institutional Activity

    Research and limitations

    Company reported repurchasing roughly 3.14m ADSs for US$9.54m by 30 Sep; officer trades and institutional fund-flow changes NOT VERIFIED.

  8. Layer name

    08 Hiring

    Research and limitations

    Company described AI operations and a Lanzhou warehouse; incremental hiring attributable to the 18 Nov report NOT VERIFIED.

  9. Layer name

    09 Public Procurement And Budgets

    Research and limitations

    No linked public tender or government budget was verified.

  10. Layer name

    10 Foreign-Language Warning

    Research and limitations

    The issuer release reviewed is in English. No separate Chinese contract was verified; translation provenance was not established.

  11. Layer name

    11 Behaviour And Payment

    Research and limitations

    Aggregate revenue and operating cash are reported, not customer-level receipts or AI service invoices.

  12. Layer name

    12 Contradictory Market Readings

    Research and limitations

    GMV rose 24.1%, revenue rose 18.2%, and the quarterly operating margin was 0.9%. High growth can coexist with a thin quarterly operating margin.

  13. Layer name

    13 Observed Missing Expected Activity

    Research and limitations

    The 18 Nov report gives no independently audited AI-part-matching error-rate study or named new paid service contract; absence in this statement does not prove none exists.

  14. Layer name

    14 Commercial Pilots And Approvals

    Research and limitations

    LingLong Insight 1.0 was described as launched in September. New external client payments, certified accuracy and permissions NOT VERIFIED.

  15. Layer name

    15 Competitive Weaknesses

    Research and limitations

    Existing distributors and human part-match specialists may outperform an AI tool on rare, safety-critical parts. Test against actual alternative.

  16. Layer name

    16 Cross-Border Spillovers

    Research and limitations

    Company cites international GMV above RMB160m for nine months; where this translates into lasting overseas contribution and margins NOT VERIFIED.

  17. Layer name

    17 Dated Prior Otr Outcomes / Failures

    Research and limitations

    OTR Learning (checked 19 Nov) warns that announced commitments are not completed cash outcomes. No dated prior OTR ZKH prediction or fail was verified.

  18. Layer name

    18 Investor Awareness / Priced-In Test

    Research and limitations

    NYSE listing and public results mean awareness exists; a dated, live NYSE quote and 24h post-report price reaction NOT VERIFIED. Do not claim cheap stock.

  19. Layer name

    19 Strongest Disconfirming Facts

    Research and limitations

    Near-zero operating margin, falling cash/investment balance and the risk of execution errors are stronger cautions than a growth headline.

  20. Layer name

    20 Action Clock / First Proof / Money / Stop Rule

    Research and limitations

    First test: written approved catalogue sample, then paid pilot. Upfront illustrative RMB1,400, horizon 1-2 quarters. Stop if wrong matches, no paid order or lower cash conversion.

Back to Signals

AI powered research