China extends extra 10% tariff on US goods to 10 Nov 2026 under trade truce

The Analysis

Strictly speaking, nothing was imposed on 31 August. The blanket additional 10 per cent that China applies to goods of United States origin was already running, and what the record verified on 29 August establishes is narrower than a headline: the measure now runs to 10 November 2026 inside the current truce. An extension is not an imposition. An imposition forces a decision on every desk that touches the goods, while an extension lets the decision stay where it was, which is why most readers stop at the date.

The number the coverage leads with is 10 per cent. Plan against a different one. The sector measures dated February 2025 still sit on top of the blanket rate: LNG and coal come to 25 per cent in total, crude oil, farm machinery and large-engine vehicles to 20 per cent. With the MFN duty underneath, many lines land close to 30 per cent effective. Ten is the floor of this structure, not the rate that applies inside it.

What follows treats the measure as an instrument rather than an event. It takes the arithmetic apart, explains why extension was chosen over the other two moves available, separates what a customs database proves from what an instrument proves, follows the layers through five product families, sets out why the declaration date is the operative fact and where that stops being true, marks the scope boundary that is worth ten points, works three examples with the assumptions stated, and closes with the two conditions that would make me revise the reading.

A blanket rate and a sector rate behave differently, and the difference decides which number belongs in which document. The additional 10 per cent attaches to goods of United States origin as a class, which makes it broad, blunt and cheap to administer. The measures dated February 2025 attach to a defined scope and were drafted to produce a total rather than an increment. LNG and coal come to 25 per cent. Crude oil, farm machinery and large-engine vehicles come to 20 per cent. Read as increments, the sector layer is fifteen points on the first pair and ten on the other three. Read as totals, those are the figures that appear on the entry. Both readings are true, they answer different questions, and that is how two people can quote the same measure and still be five points apart in a meeting.

Underneath both sits the MFN duty, and the record is careful in how it puts this: effective rates near 30 per cent for many lines once it is added. Near, and many, are the honest form of that sentence. The gap between the headline ten and the effective thirty is not concealed anywhere. It is distributed across three instruments, and no summary of a measure carries an obligation to perform that addition on your behalf.

Nothing in the record verified on 29 August explains the choice of instrument, and I am not going to invent a motive for a government that did not state one. The structure of the choice can still be described. Extension is the cheapest of the three moves available. Removal would have to be presented as a concession and would immediately raise the question of what was received in return, a question nobody wants to answer four months before it can be answered well. An escalation ends the truce both sides are currently operating inside, and a truce being used is worth more than a point of tariff. Extension holds the revenue and keeps the bargaining position intact while leaving every option alive, and a measure that leaves every option alive has an end date that reflects scheduling rather than intent.

That distinction has a consequence which is easy to miss. The date is being read as a signal and it does not carry one. A window of a little over ten weeks is long enough to justify a review and short enough to make a structural change look reckless, so the rational response at most desks is neither to restructure nor to ignore but to defer. Deferral is not free. Its cost is the value of the decision not taken, and because no one invoices it, it never becomes a line. I have watched this pattern across several rounds of measures that were extended rather than resolved: the fourth extension is not more informative than the first, but by the fourth the organisation has quietly stopped treating the question as open, which is a different failure from deciding badly.

The record also needs a distinction it does not draw. A customs database verified on 29 August tells you what was in force on 29 August. It does not tell you what the instrument says, when it was made, under which power, or how it treats goods at the margin. Those are different categories of fact, and a contract recital can only use the second. If a supply agreement is triggered by a tariff being imposed, an extension is arguably not an imposition in the ordinary meaning of the word, since nothing new is charged. The precondition for relying on that reading is that the clause uses the word imposed and does not define it to include continuation. Where a clause is drafted around any change in duty, or defines the trigger by reference to the effective rate, the answer reverses and the extension itself becomes the trigger. Both drafts are in circulation, sometimes inside the same contract portfolio, signed eight months apart.

Where the layers land differs by product family, and the useful observation is not that there are five groups but that the transmission mechanism differs in each, so no single hedge covers them. Long-term LNG and coal arrangements absorb a rate through price renegotiation rather than through substitution, because the buyer of a United States cargo is usually buying the molecule and the route together and cannot change either without reopening the contract. Crude is more elastic in principle: a refinery with a flexible slate can change what it runs, and the ones that cannot are the ones carrying the 20 per cent in full.

Farm machinery lands on a dealer network holding inventory ordered months earlier, so the duty attaches to stock that was priced before anyone knew the date, and the dealer discovers the cost after the sale. Large-engine vehicles are the mirror image in scale, few units at high value, a large amount per unit and a small amount in aggregate, which makes them loud in a board discussion and minor in the annual number. Everything else takes the blanket ten, where the amount per unit is modest, the SKU count is enormous, and the cost surfaces as administrative load rather than as a tariff line.

That last mechanism deserves its own paragraph, because it is systematically under-recorded. A per-unit amount small enough to absorb, multiplied by a SKU count large enough that absorption stops being a decision and becomes a drift, produces a cost that is real, recurring and invisible in the tariff column. It shows up as more lines to characterise, more entries to reconcile, more exceptions to explain when a file is pulled two years later. Finance will not see it because it is not a duty, and compliance will not report it because it is not an exception. The blanket layer is the one most importers describe as manageable, and it is the one that consumes the most staff time per dollar of duty.

Timing runs through the declaration rather than the shipment. The rate that governs is the one in force when the entry is declared, which places cargo loaded in the last week of October and declared in November on the far side of the date. Read that as a statement about which rate applies to an ordinary entry, not as a rule that survives every procedure. There is an exception here, and it is not a small one: where a declaration is lodged before the date but the duty point is fixed at release, or where a procedure fixes the duty at an earlier stage, the date is not the operative fact at all. Those cases exist, they are not exotic, and they are precisely where a general rule lifted from a summary answers with great confidence and the wrong number. The pivot belongs to the person who will lodge the declaration, and I would not take it from an article.

Once the timing is expressed that way, the calendar stops being one date and becomes three. The legal date is 10 November 2026. The last comfortable sailing is earlier by the length of your longest lane from load to release, and on an assumed five weeks that puts the practical cargo deadline in the first days of October, roughly five weeks from publication. The contractual date is earlier again, because a twelve-month agreement signed in the fourth quarter will run past 10 November by construction, and the allocation has to be settled before signature rather than after. A company that plans only against the legal date is planning against the latest of its three deadlines, which is the one with the least room left.

The seam the coverage does not open is scope. The sector layers are defined by product scope, so the difference between ten and twenty, or between ten and twenty-five, is decided by how goods are characterised rather than by the schedule the rate sits in. A line entered as farm machinery and a line entered as a part of farm machinery do not necessarily occupy the same position, and where they diverge the distance is ten points. That is the largest single variable in the structure. It is also the only one a company can genuinely work on, because scope is a question about the goods while the truce calendar is a question about two governments. Attention is fixed on the calendar. The exposure sits on the boundary.

The tempting inference from that is that scope is a labelling exercise. Strictly, it is not, and the difference is where companies get hurt. Scope is settled by the objective characteristics of the goods, by the technical documentation that describes them, and by the entry history that shows how they have been treated. Getting the characterisation right is legitimate work, and it is worth having: a part that has been entered as a complete machine, or a machine whose documentation supports a narrower description than the one in use, is a real finding and it is worth ten points. A description that outruns the goods is not a planning technique. It is an exposure with a delayed fuse, and the fuse is the length of the audit cycle, which is comfortably longer than the political cycle that produced the measure. The return on that trade is a few points today against a recovery, plus whatever the jurisdiction adds to it, at a date you do not choose.

The two layers also place the burden of proof in different places, which is why they fail differently when a file is examined. The blanket layer is answered by origin: where the goods were produced, from what, and whether processing somewhere else changed their character. The sector layer is answered by scope: what the goods are, how they are described, and what the technical file says they do. A company can hold a complete origin file and still lose the scope argument, and the reverse happens about as often. When a recovery is assessed the two questions are put separately, by different people, against different documents, and a file that answers one of them well tends to create the impression that it answers both.

One more structural point that never appears in the coverage. The declaration date is a property of an entry, not of a company. An importer declaring at four entry points does not face one cliff on 10 November; it faces four, and they are not synchronised, because each depends on when the goods arrive and when the declaration is lodged. The same holds inside a single port where different procedures are used for different flows. A group that manages the date centrally as one item will discover the divergence in the second week of November, which is the worst week in which to discover it.

Scale changes which part of the structure matters, and it changes it out of proportion. A large importer with a standing compliance function can carry three buckets, sees the exposure monthly and can staff the boundary work; its problem is coordination across entry points rather than ignorance. A mid-size importer usually has one person wearing the classification hat alongside two other jobs, which means boundary work happens only when something forces it, and the exposure stays invisible until an audit or until a counterparty raises it in a dispute. A small importer is mostly a consumer of the blanket layer, and the sector layer bites only if it happens to buy into one of the five families, which is exactly the case where a single purchase order can move it from ten to twenty. The distribution is not a gradient. It is a set of different problems sharing a date.

Direction adds a second axis. Goods of United States origin are what the blanket measure reaches, and origin is itself a determination rather than a notation on the invoice. A product assembled outside the United States from United States components is a different case from a product made in the United States, and a product whose character is altered by processing elsewhere raises the substantial transformation question that decides where it sits. I am not going to resolve that question in a paragraph, and neither should a summary. What I will say is that the origin file is the first thing an audit asks for and the last thing companies prepare, and that the extension gives you four more months in which preparing it is cheap.

Two worked examples, with assumptions stated because the record does not supply them. Assume a buyer importing USD 3 million of United States LNG a month. At 25 per cent the duty is USD 750,000 a month; at the blanket 10 per cent alone it would be USD 300,000; the sector layer is therefore USD 450,000 a month, or USD 5.4 million a year, and that is the standing position rather than a stress case. Assume a second buyer importing USD 3 million of United States crude a month at 20 per cent. The duty is USD 600,000, the blanket alone would be USD 300,000, and the sector layer is USD 300,000 a month, USD 3.6 million a year. Identical volume, half the exposure, and the only thing that changed is the family the goods sit in.

A smaller case on the same logic, again with the assumption on the table. Assume a mid-size importer buying USD 400,000 a month of goods characterised as farm machinery. At 20 per cent the monthly duty is USD 80,000; at the blanket ten it would be USD 40,000; the sector layer is USD 40,000 a month, USD 480,000 a year. On a business that size the sector layer is not a variance, it is a fixed cost that did not exist in the budget, and the decision it forces is not about sourcing but about whether the product line stays. Aggregate figures never show this case, because 480,000 is a rounding error in a national total and a structural problem in a company of that size.

The contract question is narrower than it looks and harder than it sounds. Most twelve-month agreements carry a duty-change clause that says what happens when a rate changes. Very few say what happens when a measure expires, because expiry is not a change of rate, it is the disappearance of a layer. Silence is not neutral. Where the clause is silent, allocation falls back to the general position and to whoever holds title at the duty point, and neither of those necessarily matches what the commercial team assumed when the price was agreed. The precondition for that fallback is that the contract does not allocate the risk somewhere else, in a price formula, a hardship clause or an indexation mechanism, and those three are where I would look before drafting anything new.

Two counterfactuals decide how much work is justified. If the measure is extended again before 10 November, the calendar moves and nothing else does: the same USD 5.4 million a year keeps running, the planning horizon gains a few months, and the analysis you did not do is postponed by a quarter at the price of another quarter of the same exposure. If the sector measures were rolled back, the structure collapses to the blanket ten for most lines, most of the boundary work loses its value, and the sensible response becomes to budget the ten and stop restructuring. Nothing in the record verified on 29 August points to the second. The condition that would flip my conclusion is a change in the sector lists, and a change in the sector lists is a different instrument from an extension of the date, made by a different process, carrying its own transition. Coverage watches the date. The date is the part that moves without changing anything.

What a policy desk can do that a trading desk cannot is keep the record straight. The rates are public and nobody inside a company can move them. The exposure is private, assembled from your own entries, your own contracts and your own characterisations, and it can be improved in a week. The extension did not change the rates. It changed the amount of time available to establish which of them actually applies to you, and four months is sufficient for that if it begins now rather than in the last week of October.

↑ Back to the key points

— By Dr. Ingrid Voss