Is the ban leaking? Civil drones after the US UAS Covered List
Trade-flow companion to the price wedge (R72). DUAL-SCORE / alternative-track — never folded into any Tier-1 exposure score. The wedge measures scarcity; this desk measures the other half: is the control actually holding, or is the restricted trade re-entering the buyer by another door? Research, not investment advice. Circumvention is INFERRED from the structure of the flows plus public regulatory records, never asserted as evasion of any particular consignment.
Verdict — GATE 1
Between 2022 and 2024 China's declared share of US unmanned-aircraft imports fell 46.8% → 18.3% while the US market grew 65%. The volume did not disappear. It moved to Malaysia (already the #2 origin, then +111%) and — from a standing start of $169,012 — to Viet Nam, ×371, to 13.1% of the US market in two years.
But the reason this case belongs in the corpus has little to do with the table. It is the first row where layer 2 fails by construction, and the failure is the finding.
Every ownership tell the register has ever traced — Youngsun → Thai Unipet → Youngsun & Essen; Tsingshan at Morowali; Hongqiao at Ketapang; Huafu's Long An spindles — is a shareholding. Something a registry records. Here the controlled producer holds no equity in the vehicle at all. DJI licensed its design to an independent, US-registered company, which contracts a Malaysian factory, and takes a payment per unit instead of a stake. Congress's own words for it: a "passthrough company." A registry search returns nothing, because there is nothing in a registry to return. The link is a contract.
Call it mode H — licensed-design white-label. It is not mode B (no relocated capacity the producer owns) and not mode C (no transformation conferring origin). It is the controlled entity selling its identity-free technology and keeping the royalty — and it defeats the corpus's ownership test outright.
The second finding is about the instrument. The FCC's designation is written against place of production, not brand — which pre-emptively voids mode H, exactly as the Russian-seafood determination pre-emptively voided mode C. Three weeks later the Commission carved an exemption. The exemption is cut precisely where the real dependency sits.
Layer 1 — the buyer-side table
| US imports, unmanned aircraft (HS 8806), US$ | 2022 | 2024 | 2022 share | 2024 share | move |
|---|---|---|---|---|---|
| China (the restricted origin) | 136,568,009 | 87,872,139 | 46.8% | 18.3% | −36% |
| Malaysia | 119,631,346 | 252,048,044 | 41.0% | 52.5% | +111% |
| Viet Nam | 169,012 | 62,783,459 | 0.1% | 13.1% | ×371 |
| Canada | 17,432,079 | 20,300,603 | 6.0% | 4.2% | +16% |
| Korea | 986,360 | 10,589,515 | 0.3% | 2.2% | ×10.7 |
| Switzerland | 5,887,032 | 12,517,482 | 2.0% | 2.6% | +113% |
| World total | 291,577,813 | 480,504,491 | — | — | +65% |
Source: UN Comtrade free public preview API, reporter USA (842), flow M, annual, HS 8806. Reproduce: `https://comtradeapi.un.org/public/v1/preview/C/A/HS?reporterCode=842&period=2024&cmdCode=8806&flowCode=M`. 2023 is not served by the free preview for this reporter and the Census API requires a key, so the intermediate year is a labelled gap — the China-side mirror below covers it.
Two honest qualifications before reading anything into this.
(i) Malaysia is not a post-control surge. It was already 41.0% of the US market in 2022 — before the FY2025 NDAA, before the Covered List, before the white-label vehicle below shipped a unit. Malaysia is a pre-existing offshore assembly base that grew; it is not a phantom flag, and the naive volume-implausibility test does not fire on it. Penang-class electronics contract manufacturing is real capacity. As on Xinjiang cotton: the surgers can genuinely build.
(ii) Viet Nam is the new entrant, and its units are not consumer toys. At 18,302 units for $62.8m the 2024 Vietnamese line averages $3,430 and 6.0 kg per aircraft, against Malaysia's $718 / 1.45 kg and China's $1,048 / 2.61 kg (same Comtrade rows, qty and netWgt fields). Different class of machine, arriving from a country that shipped the US $169k of drones two years earlier.
The China-side mirror — the line that fell is the direct one
| China's exports, HS 8806, US$ (reporter CN, flow X) | 2022 | 2023 | 2024 | 2022 share | 2024 share |
|---|---|---|---|---|---|
| → USA | 498,569,654 | 254,585,967 | 193,117,466 | 27.6% | 8.9% |
| → Netherlands (EU distribution hub) | 103,163,777 | 161,308,785 | 232,965,795 | 5.7% | 10.8% |
| → Hong Kong | 1,174,814 | 29,412,895 | 83,239,295 | 0.07% | 3.85% |
| → Malaysia | 18,064,813 | 33,910,311 | 59,267,326 | 1.00% | 2.74% |
| → Viet Nam | 18,908,927 | 27,376,717 | 45,249,625 | 1.05% | 2.09% |
| → Mexico | 29,800,182 | 37,048,535 | 51,510,025 | 1.65% | 2.38% |
| → Russia | 22,892,369 | 13,315,428 | 583,281 | 1.27% | 0.03% |
| World total | 1,806,027,794 | 1,805,289,752 | 2,162,950,877 | — | +19.8% |
Source: UN Comtrade, reporter China (156), flow X, annual, HS 8806. Reproduce: `https://comtradeapi.un.org/public/v1/preview/C/A/HS?reporterCode=156&period=2024&cmdCode=8806&flowCode=X`. Partner descriptions are null in the preview payload; codes were mapped by M49 (842 USA, 528 NL, 344 HK, 458 MY, 704 VN, 484 MX, 643 RU).
This is the share column doing the work, exactly as on plywood. China's world drone exports grew 19.8%, and its 2022→2023 world total was flat to four significant figures — while the US line halved in that same flat year. The fall is origin-specific, not demand-wide. By 2024 the United States is no longer China's largest declared drone destination; the Netherlands is, and Hong Kong — an entrepôt whose 2022 imports of drones ($173.0m) exceed its exports ($125.3m), i.e. it re-exports essentially everything it takes in — has gone ×71 to sit just behind.
The Russia row is the control group: $22.9m → $0.58m, −97%, the direct line extinguished under sanction. What replaced it is not in this heading, and is covered by the dual-use-microelectronics row rather than here.
A reporter gap that must be stated, not buried
China says it shipped the US $193.1m of drones in 2024. The US says it received $87.9m of Chinese-origin drones. The gap is $105m, and it is not a one-off — in 2022 it was $362m ($498.6m vs $136.6m). A stable 2–3.7× discrepancy in the same direction is a classification artefact, not a smoking gun: HS 8806 was created in HS2022, and camera-carrying consumer drones are classified inconsistently across reporters (US practice has historically reached for the 8525 camera headings).
The consequence is a rail, not a footnote: the two tables above are not level-comparable across reporters. Shares within a reporter are informative; subtracting one reporter's number from another's is not. This case therefore does not run the magnesium-style mirror-gap-as-volume test, even though the arithmetic would look spectacular.
The input-line ratio — and what it ranks
The plywood/cotton diagnostic: read the INPUT line into the surger, not the finished-good origin line. Drones have an adjacent parts heading, HS 8807.30 (parts of goods of headings 88.01/88.02/88.06), so China's own export statistics report the feedstock — with the caveat that 8807.30 also pools civil-aeroplane parts, so it is an upper bound on drone content.
| Chinese input into the origin ÷ that origin's US-bound drone shipments, 2024 | 8806 only | 8806 + 8807.30 |
|---|---|---|
| Viet Nam ($45.25m + $15.97m in; $62.78m out) | 72.1% | 97.5% |
| Malaysia ($59.27m + $19.75m in; $252.05m out) | 23.5% | 31.4% |
Chinese input: UN Comtrade reporter China (156), flow X, HS 8806 and HS 880730. US-bound: reporter USA (842), flow M, HS 8806.
VN 72–97% ≫ MY 24–31%. Because the numerator and denominator come from different reporters, and the China line is the one that over-reports relative to the US line, both ratios are upper bounds — but both are inflated by the same factor, so the ranking is the robust part, and the ranking says the leg with almost no domestic value added is Viet Nam, not Malaysia.
Malaysia's ~31% is independently corroborated from a completely different direction: the House Select Committee found "roughly half of Anzu's parts come from China" — a company-specific figure against a country-wide ratio, same order of magnitude, arrived at by inspection rather than by customs arithmetic.
And here is the plywood petition-perimeter rule firing again. Congress and the press went to Malaysia — the lower-ratio origin — because that is where the named vehicle assembles. Viet Nam, the higher-ratio leg, is un-named in any of the public record reviewed here. The enforcement perimeter tracks the entity that got written about, not the flow.
(Neither Malaysia nor Viet Nam serves HS 8806 in the free Comtrade preview, so the surger-side mirror cannot be checked. Labelled gap.)
Layer 2 — the licence that leaves no ownership footprint (mode H)
On 27 August 2024 the House Select Committee on the CCP published its letter to Anzu Robotics (dated 20 August 2024). The findings, verbatim from the Committee:
> "Anzu's Raptor T is essentially a DJI Mavic 3 painted green, with its remote > control and application all running on DJI technology."
> DJI is "using Anzu as a passthrough company in an attempt to avoid current and > anticipated U.S. restrictions on DJI products."
The chain, as the Committee established it:
> DJI (Shenzhen — the restricted producer, licensor) → Anzu Robotics (US-registered; > CEO Randall Warnas, at DJI 2015–2017) → contract manufacture in Malaysia → > US enterprise / public-safety buyers.
The consideration is the tell. Anzu stated publicly that no royalties are shared with the licensor, while acknowledging that DJI receives a payment for every drone Anzu orders from its Malaysian manufacturer — plus "priority technical support." The Committee also flagged Cogito Tech Company Ltd to Commerce in the same breath, so the pattern is not a single vehicle.
Why this breaks the detector. Run the corpus's standard layer-2 test — common ownership, shared name-root, shared address, a registry hop — and it returns clean. There is no shareholding to find. Restated as a rule:
> The licensing rule. A control that binds on who owns the seller is defeated by > a producer that sells its design rather than its equity. Layer 2 must > therefore test for a licence and a per-unit payment, not only for a shareholder > — and where the registry is silent, the substitute evidence is product identity > (a teardown showing the same airframe, firmware and controller) rather than > corporate identity.
Outcome. The route did not scale. Anzu ended Raptor production, announced 17 February 2026, the company citing component shortages after demand "driven in part by behavior tied to the NDAA 2025" depleted stock early. The company says it is working on a next-generation product. The public reporting does not attribute the discontinuation to the Covered List — treat cause as open.
Layer 2b — the dependency that never left, and the exemption cut around it
The FCC's designation of 22 December 2025 (DA 25-1086) does not name a brand. It adds, on an interagency National Security Determination of 21 December 2025, all UAS and UAS critical components "produced in a foreign country" — and the Determination enumerates them:
> "UAS and UAS critical components, including data transmission devices, > communications systems, flight controllers, ground control stations, controllers, > navigation systems, batteries, smart batteries, and motors produced in a foreign > country…"
> "UAS and UAS critical components must be produced in the United States."
Written that way, mode H is void on arrival: a US-branded, US-designed drone produced abroad is covered anyway, because the trigger is the production fact, not the flag or the badge. This is the C⁰ move the register first isolated on Russian seafood — attach the prohibition to an extraction/production fact and the substantial-transformation door never opens. The corpus's announcement-time test passes on a second, completely unrelated instrument.
Then, on 7 January 2026, DA 26-22 exempted two classes until 1 January 2027: UAS and critical components on the Blue UAS Cleared List (over 39 platforms, over 165 items), and equipment qualifying as a "domestic end product" under Buy American — at least 65% US component cost.
Seven weeks before that exemption, DefenseScoop reported (20 November 2025) that "the majority of the unmanned aerial systems cleared through this effort have motors that are sourced in China," and that per a former senior defense official the top three Chinese components in Blue UAS platforms are "motors, the batteries and the electric speed controllers" — the parts long treated as low-risk "dumb" components. AUVSI's Michael Robbins: "Motors were not considered one of those critical components. That has led to there being a very immature drone motor market outside of China."
Set the three documents side by side:
1. The National Security Determination names motors as a covered critical component. 2. Reporting one month earlier finds most Blue-UAS-listed drones run Chinese motors. 3. The exemption is granted to the Blue UAS list.
The dependency is not being trans-shipped. It is arriving, correctly declared, as a component, and the buyer's own domestic industry assembles it inside the wall — the register's mode G, first isolated on EU nitrogen fertiliser, reproduced on a manufactured good. And the cheapness is the mechanism, not a detail: the same official gives "a small brushless motor is $12 to $25 per motor, from China. The best that you can find in the United States… is like $100 to $225 per motor."
That yields the case's second transferable rule, and it is the drone twin of the contained-element rule:
> The cheap-part rule. A value-based domestic-content threshold cannot see a > dependency that is cheap. At an 8× price disadvantage, the Chinese motors, > batteries and ESCs in an airframe consume a trivial share of bill-of-materials > cost while being the parts with no Western supply — so a drone can clear > "65% US component cost" and remain 100% dependent on a single foreign source for > the components that decide whether it flies. Grade domestic content in > substitutable-supplier count, never in cost share.
What it means for the buyer
1. Origin data on this good is compromised in both directions. A risk model keyed to declared origin now reads "18% China" on a market where the largest origin is a contract-assembly base and the fastest-growing one imports ~72–97% of its US-bound value from China. Read the input line and the licence, not the flag. 2. The registry test alone will now miss the vehicle. Mode H is cheaper to build than mode B — no plant, no equity, no consolidation — and leaves a contract-shaped hole where the corpus expects a shareholding. Expect it wherever a dominant producer faces a brand-specific or entity-specific control. 3. The instrument that works is the one aimed at production, not at a name. The FCC wrote a category-wide, production-fact designation and mode H was void before it scaled. The register now has two independent instruments (US seafood, US UAS) showing the same drafting move working. 4. Check where the exemption is cut. The designation closed the door; the Buy-American cost threshold reopened it for exactly the components with no alternative supplier. The exemption sunsets 1 January 2027 — that date, not the December designation, is the live watch item.
Method & honesty rails
- Trade data: UN Comtrade free public preview API. Reporter USA (842) and China
(156), annual, HS 8806 (unmanned aircraft) and HS 880730 (parts). Every reproduce URL is printed under its table.
- Known data limits, all labelled above: US 2023 absent from the free preview;
Malaysia and Viet Nam do not serve 8806 at all; HS 8806 is an HS2022 heading with cross-reporter classification inconsistency (a stable 2–3.7× China-vs-US gap), so levels are not comparable across reporters and no mirror-gap-as-volume claim is made; HS 880730 pools civil-aeroplane parts, so the input ratios are upper bounds.
- Inference, not accusation. No shipment is asserted to be unlawful. The
passthrough characterisation is the House Select Committee's, quoted as a public congressional finding, not an adjudication; Anzu has publicly disputed the characterisation of its independence. The Blue UAS component finding is press reporting attributed to unnamed defense officials, and DIU's response is quoted as given.
- Alternative-track only: never folded into
buyerRelativeScoreor any Tier-1
exposure score.
- Open questions: where Hong Kong's ×71 drone intake goes (HK's 2023–24 export
side is not served by the preview); whether the Vietnamese leg has a named vehicle; whether the 1-Jan-2027 sunset is extended or the Blue UAS list is broadened to absorb the motor/battery/ESC dependency before it bites.
Sources
- FCC Public Notice DA 25-1086 (released 22 Dec 2025), incl. the interagency
National Security Determination of 21 Dec 2025 — https://docs.fcc.gov/public/attachments/DA-25-1086A1.txt
- FCC Public Notice DA 26-22 (released 7 Jan 2026), Blue UAS / domestic-end-product
exemption — https://docs.fcc.gov/public/attachments/DA-26-22A1.pdf ; scope summary (39+ platforms, 165+ items, ≥65% US component cost, expires 1 Jan 2027) via Morgan Lewis client alert, Jan 2026 — https://www.morganlewis.com/pubs/2026/01/fcc-exempts-certain-drones-and-components-from-covered-list-to-address-national-security-risks
- House Select Committee on the CCP, "Moolenaar, Krishnamoorthi Expose PRC Drone
Company Masquerading as U.S. Firm," 27 Aug 2024 — https://chinaselectcommittee.house.gov/media/press-releases/moolenaar-krishnamoorthi-expose-prc-drone-company-masquerading-us-firm ; underlying letter to Anzu Robotics, 20 Aug 2024 — https://chinaselectcommittee.house.gov/sites/evo-subsites/selectcommitteeontheccp.house.gov/files/evo-media-document/2024-08-20%20anzu%20letter.pdf
- DefenseScoop, "Pentagon's growing list of 'made in America' drones has a loophole
for certain parts made in China," 20 Nov 2025 — https://defensescoop.com/2025/11/20/dod-drones-blue-uas-list-chinese-parts-motors/
- DroneDJ, "Anzu Raptor drone discontinued," 17 Feb 2026 —
https://dronedj.com/2026/02/17/anzu-raptor-drone-production-end/
- UN Comtrade free public preview API — reproduce URLs printed under each table.