Is the germanium ban leaking? Germanium (and gallium) after China's Dec-2024 US export ban

Trade-flow companion to the price wedge (R72). The wedge measures scarcity (what a controlled material costs ex-China). This case measures the other half: is the control actually holding, or is banned material re-entering the US through a laundered origin? This is a DUAL-SCORE / alternative-track signal — never folded into any Tier-1 exposure score. The divergence between "who China says it stopped selling to" and "who suddenly started selling to the US" IS the signal. Research, not investment advice; origin-relabelling is INFERRED from statistical implausibility plus public production/ownership records, never asserted as smuggling on any single shipment.

Live chart + method: [/situation-room → "Is the ban leaking?"](/situation-room).

Verdict — GATE 0 (dual/mixed: genuine substitution + a muted, basket-obscured tell)

Recomputed 2026-10-02 on the newest window (Jan-Jul 2026 vs Jan-Jul 2025, reporter USA, same two HS lines). Both halves held direction and one sharpened in a way the annual table could not show: see the two "recompute" tables below each line's annual table.

On 3 December 2024 China's MOFCOM (Announcement No. 46) banned dual-use exports of germanium and gallium (with antimony and superhard materials) to the United States — the same instrument that produced the clean antimony trans-shipment fingerprint ([antimony case](2025-antimony-ban-circumvention-transshipment.md)). Germanium and gallium are the semiconductor / fibre-optic / IR-optics twins in that ban; China refines ~60% of germanium and ~98% of gallium (USGS MCS 2026 / IEA / EU-JRC). Yet germanium's post-ban flow signature is not the antimony story. It splits in two, and the two halves point in different directions:

(1) Germanium OXIDE — genuine-ally substitution, NOT laundering. US imports of germanium oxide (isolated inside the polluted HS 282560 line by unit value — see method) come from exactly the refiners USGS already lists as the genuine US germanium supply base, and their value climbed as the price surged:

Recompute, Jan-Jun 2026 vs Jan-Jun 2025 (monthly, >$100/kg rows only — July 2026 excluded here because `netWgt` is not yet reported for any of the three partners that month, so the unit-value filter cannot run on it; July's raw dollar values are Belgium $839k, Canada $10.97M, Germany $1.16M, stated without a filtered weight):

Belgium's near-total exit from the priced oxide line is a genuinely new finding the annual table (which still shows 2025 full-year Belgium at $12.1M) could not surface — H1 2026 is not a continuation of 2025's three-way split, it is Canada absorbing essentially the whole US oxide demand this case tracks. No explanation is claimed (outage, re-allocation, and a reporting lag on Belgium's July row are all consistent with the data); it is a fact to watch, not yet a verdict. The substitution-not-laundering read is unaffected — Canada is the same genuine USGS-listed primary producer the original verdict already named.

China does not appear in these rows: China's HS 282560 line to the US is low-value zirconium dioxide (~$12–22/kg, thousands of tonnes), a different molecule that shares the six-digit code. The germanium-priced rows are refined Western germanium. The pattern is the bismuth pattern ([bismuth case](2025-bismuth-licensing-circumvention-transshipment.md)): the control bites through price (unit values ~5–7× on the clean 2023→2025 comparison), and the gap is filled by real allied refiners (Umicore, Teck), not a zero-capacity relabel. No implausible origin. No ownership pipe.

(2) Ga/Ge METAL + scrap basket — a muted, basket-obscured relabel tell. The unwrought-metal-and-scrap line (HS 811292) tells the other half — but it is a basket (it pools gallium, germanium, hafnium, indium, rhenium, thallium, etc.), so nothing here can be pinned to germanium/gallium alone:

Source: UN Comtrade, reporter USA (842), flow imports (M), annual, HS 282560 and 811292, pulled via the free public preview API. Production shares: USGS MCS 2026, `lib/iptm-material-country-production.ts` (germanium refining split is a modelled estimate — USGS publishes no country table for germanium — anchored on China as leading producer + US-import-source shares BE 41% / CN 23% / CA 17% / DE 14% + IEA/EU-JRC ~60% China).

Recompute, Jan-Jul 2026 vs Jan-Jul 2025 (monthly, customs value):

The zero-capacity-origin tell is no longer a 2023→2025 artefact: both Taiwan and Malaysia kept growing into H1 2026 while China's share stabilised rather than continuing to fall. Still unattributable to germanium/gallium specifically at HS6 granularity (the basket caveat below is unchanged) — the tell is sharper, not newly resolved.

China's basket line to the US collapsed −70% in the first full post-ban year, and the fastest-growing origin is "Other Asia, nes" (predominantly Taiwan) — a place with no primary gallium or germanium capacity — which more than quintupled to become the #2 supplier, with Malaysia (also ~0% capacity) appearing new. That is the textbook trans-shipment shape (the same "Other Asia, nes" surge the [tungsten case](2025-tungsten-licensing-circumvention-transshipment.md) flagged on APT). But the basket cannot say the surge is germanium or gallium rather than indium, hafnium or rhenium; the surger unit values (~$375–400/kg) are consistent with gallium or indium, not germanium metal (~$1,500–3,000/kg). So the tell is real at the basket level and unattributable at free-data granularity.

Net: the dominant, clean reading is genuine allied substitution (Ge oxide, Umicore/Teck); the laundering tell exists only on the polluted metal basket and cannot be isolated to the controlled molecules. No common-ownership pipe traced. → GATE 0. Germanium sits between antimony (a clean leak) and bismuth (a clean genuine-substitution): the oxide leg is bismuth-like, the metal leg is a weaker, basket-obscured tungsten-like tell. 2026-10-02 recompute: both legs hold the same direction on the freshest window — the oxide leg concentrated further into a single ally (Teck/Canada, after Belgium's H1 2026 near-exit), the basket's zero-capacity-origin growth (Taiwan, Malaysia) continued rather than being a 2023-2025 blip, while China's own basket share stopped falling. Still GATE 0: no new ownership pipe, nothing isolable to germanium/gallium specifically.

Why it matters for the buyer

1. The same ban produces opposite flow signatures across its own materials. Antimony (Announcement 46) leaked through Thailand/Mexico with a named corporate pipe; germanium oxide (Announcement 46) did not leak — it re-sourced to Umicore and Teck at a much higher price. A risk team cannot assume "China banned it → therefore it's being laundered." Some controls are absorbed by real non-Chinese capacity; the flow data tells you which, per material, per product form. 2. Product form decides everything. Germanium oxide has genuine Western refiners (Belgium, Canada, Germany) and re-sourced cleanly. Germanium/gallium metal and scrap sits in a basket where a zero-capacity origin (Taiwan, Malaysia) surged — the leg where relabelling, if it is happening, would hide. A buyer sourcing "germanium metal from Taiwan" post-ban should know Taiwan refines no primary germanium. 3. It confirms the control is binding — via price, not (mainly) via leakage. The 5–7× unit-value jump on genuine allied germanium oxide is the same fact the price wedge prices: ex-China germanium is genuinely scarce and expensive. Flow re-routing to real refiners and a widening wedge are two readings of one thing — the tap is real, even where the origin label is honest.

Method & honesty rails

• Trade data: UN Comtrade, US imports, annual, HS 282560 (germanium oxides /

zirconium dioxide — shared code) and HS 811292 (unwrought gallium, germanium, hafnium, indium, rhenium, thallium… + waste/scrap). Free public preview API, no key. netWgt is product gross weight (kg), not contained Ge/Ga; primaryValue is customs value (USD). Late-2025 months lag ~2–3 months, so 2025 may be slightly under-counted.

• The unit-value separation (the key move): zirconium dioxide trades at

~$3–10/kg; germanium oxide at ~$500–1,200/kg — a ~100× gap. Filtering HS 282560 to rows above $100/kg cleanly isolates germanium oxide from the zirconia that shares the code. This is a heuristic, stated as one; the transitional 2024 Belgium/Canada rows ($100–135/kg, 80–103 t) look zirconia-contaminated and are treated as noisy — the clean read is 2023 (pre-ban) vs 2025 (first full post-ban year), where prices are unambiguously germanium.

• The basket caveat (why not GATE 1): HS 811292 pools several controlled

minor metals. The China collapse and the Taiwan/Malaysia surge are real at the basket level, but cannot be attributed to germanium or gallium specifically at free HS-6 granularity — a US HTS-10 breakout (behind an API key) or a national customs portal that isolates gallium/germanium would be needed to lift this leg toward GATE 1. Until then the relabel tell stays labelled and un-gated.

• "Other Asia, nes" (Comtrade partner 490) is predominantly Taiwan in US

customs reporting; treated as such here, as in the tungsten case.

• 2026-10-02 recompute replication: monthly reporterCode=842,

cmdCode=282560,811292, flowCode=M, periods 202501-202507 and 202601-202608 (August not yet reported). The subscription COMTRADE_API_KEY was quota-exhausted when this recompute started (HTTP 403 "Out of call volume quota") — fell back to the free public preview endpoint (https://comtradeapi.un.org/public/v1/preview/C/M/HS), one period+code per call (28 calls, ~2.7s pace), deduped with scripts/py/comtrade_dims.dedupe_dimensions. Partner codes: Belgium 56, Canada 124, Germany 276, China 156, Malaysia 458, "Other Asia, nes" 490, World 0.

• Alternative-track only: this never touches buyerRelativeScore or the base

exposure — it sits next to them, exactly like the China–West price wedge.

• Inference, not accusation: substitution is read from genuine USGS refiner

shares; the metal-basket tell is inferred from an origin's ~zero USGS production. No individual shipment is asserted to be illegal.

• Generalises (R72): completes the Announcement-46 trilogy — antimony (clean

leak, GATE 1), germanium/gallium (mixed, GATE 0 here), against the same two-layer detector.