US Steel Imports Down 30% Year-to-Date as Section 232 Tariffs Reshape Domestic Production Landscape

The Analysis

The 30 percent figure is the one that travels. United States steel imports are down 30 percent year to date in 2026 against the same stretch of 2025, and the number is now quoted as though it were the event itself. Strictly speaking it is an outcome, and an outcome is the weakest kind of evidence about what a rule requires of you.

The instrument behind it has several parts, and only one of them is a rate. There is a 25 percent baseline. There is a perimeter that now reaches aluminium, steel and copper. There is a newly created Annex I-C category for mobile machinery. And there is a domestic-content threshold, lowered to 85 percent, which is the part I would read twice before reading anything else.

What follows is a reading of the instrument and of the premises it rests on, with arithmetic where arithmetic helps. I am not going to tell anyone which supplier to use or which programme to move; that is a commercial judgement and it belongs to whoever carries the inventory. Two of the figures in circulation are measurements rather than legal text, the 82 percent capacity utilisation and the 15 to 25 percent input-cost increase, and it matters that they are kept in that category.

A tariff schedule answers one question, which is what rate attaches to a described article. A domestic-content threshold answers a different one, which is whether the article is inside the preference wall at all. Where an instrument carries both, the second question governs, because a rate applied to an article that was never properly inside the wall is a category error rather than a level error. Most of the argument in circulation is about the rate, and the reason is easy to see. A rate is quotable. A threshold is not. The 25 percent baseline can be repeated in a headline; the 85 percent line has to be tested against a bill of materials before it means anything, which is why it gets less airtime and more of the outcome.

The perimeter change deserves the same care. Naming mobile machinery in a new Annex I-C category does something a metal rule cannot do on its own: it moves the boundary of the regime from raw metal to finished and semi-finished goods. A machine that sat outside the instrument is inside it now, and nothing about the machine changed. What changed is the description under which it is treated. That distinction is close to the whole of administrative law in one sentence, and importers who think of a tariff as a tax on steel rather than as a set of descriptions will keep being surprised by which of their products attract it.

The content threshold is where the instrument stops being a price signal and becomes an evidentiary rule. Eighty-five percent domestic content is not a statement of intent, a supply-chain strategy, or a direction of travel. It is a measurement, taken against a defined base, on a defined date, for a defined article. The premise of this rule is that the base exists as a document. Where it does not exist as a document the threshold does not decide anything, it only postpones the argument until somebody with authority asks for the file. The word that matters in the summary is lowered: the line moved to 85 percent, which is to say more builds clear it than cleared it before, and, more to the point for anyone with a sourcing map open, more builds now sit close enough to it that the answer depends on how the base is calculated. A rule that pulls the line toward where supply chains actually operate converts a compliance formality into a live question.

There is an exception here worth stating plainly, because it is the one most often skipped. Clearing the content line does not remove an article from a named category, and being named in a category does not by itself mean the content test is failed. The two tests run in sequence on the same article, and passing one is not passing the other. A build that is entirely domestic in content can still be an article described by Annex I-C, in which case the perimeter rather than the threshold decides the outcome. A build that sits outside every named category can still be pulled in by the content test if the article is of a kind the instrument reaches. Reading only one of the two produces a confident answer that happens to be wrong, which is the expensive kind.

There is a discipline to reading a text like this that I want to name, because I have watched it decide outcomes. The primary source is the text as published, at the version in force, annexes included. Commentary, including good commentary, is secondary and should never be the thing a file rests on. Annexes move faster than statutes do, and a category that has been created carries a creation date, which means two otherwise identical shipments can land on different sides of the instrument depending on when they crossed.

Where a text is ambiguous the resolution arrives later and slowly, and a firm that needs certainty now is, in practice, buying it in the form of a conservative reading of its own description. That is not a legal opinion and I am not offering one. It is the observation that in an ambiguous regime the party holding the ambiguity is the party carrying the risk, and that risk is priced into whatever the other side of the contract charges for accepting it.

The practical difficulty is documentary rather than conceptual. Very few bills of materials are maintained to the line at which a content test is decided. Assemblies are traced; sub-components are assumed. The point at which a preference is lost usually sits one tier below the tier anyone holds a drawing for, and it is lost quietly, because nothing about the shipment looks different on the day it crosses. I want to be careful about what I am describing. This is not an allegation that importers misdescribe their goods. It is the narrower and more common finding that the evidence needed to support a position was never assembled, and a position that cannot be evidenced is, for practical purposes, not held.

Now to the number everyone quotes. Steel imports down 30 percent year to date against the same stretch of 2025 is a real and large movement, and it is an aggregate. It tells you what happened to a national volume. It does not tell you what happened to the grade you buy, the width you buy, the port you buy through, or the programme you are running. Aggregates and positions diverge routinely, and the divergence is widest in exactly this kind of regime, because substitution happens first at the commodity end of the range and last at the specialty end. If your intake has not tightened, that is not evidence that the 30 percent is wrong. It is evidence that your position is not the average position.

One more caution about causation, because this is where a trade statistic gets misread most often. The summary attributes the 30 percent decline to the tariff regime, and that attribution is plausible. It is also an inference. The counterfactual that would settle the question, which is what import volume would have done had the regime not been expanded, is not observed anywhere, and a decline of that size sitting alongside a tariff expansion is consistent both with the tariffs causing it and with other forces contributing to it. None of this changes what you have to do, because the obligation runs off the text rather than off the cause.

It does change how much weight to give a forecast built on the number, and forecasts are where most planning error enters a budget. When someone tells you the drop proves the policy works, ask what the counterfactual is. There usually is not one, and its absence is not a refutation; it is a reason to hold the number as a fact about volume rather than as proof about mechanism.

Two of the three transmission channels are already open. The volume channel closed some time ago, since a year-to-date figure of that size cannot accumulate inside a single quarter, which means the substitution it measures has already been made by somebody upstream of you. The cost channel is the one arriving now. Steel-consuming manufacturers report input costs 15 to 25 percent higher, and that range is a measurement from the field rather than a term in the instrument, so treat it as an indication of direction and magnitude and not as a rate. What makes it consequential is the shape in which it arrives. Not a tariff line on your invoice, but a supplier price increase with a reason attached, negotiated at whatever cycle your contracts happen to run.

The third channel is the slow one, and it outlasts the news. Domestic producers have lifted capacity utilisation to 82 percent, the highest since 2021. Decisions of that kind are taken on horizons measured in years, and the utilisation figure is the visible edge of a set of choices about what gets built and where. That is why the printed expiry date is the least useful date in the file. A regime that has already changed investment behaviour has a life longer than its own text, and a regime whose text expires on 31 December 2027 can still be shaping the sourcing default long after the text is gone, through capacity that was built while it was in force.

Utilisation also has a limit as evidence, and it is worth being exact about that limit. Eighty-two percent is a national aggregate across products that are not interchangeable. A tight aggregate is compatible with comfortable availability in your grade, and a comfortable aggregate is compatible with a long wait for the one width your line needs. The statistic describes a system. You buy from a plant. When a supplier tells you the market is tight, the question to put back is which market, and the answer should name an item rather than a country.

Exposure sorts by legal position rather than by size, which is not how most coverage sorts it. An importer of raw metal faces the rate and the perimeter and little else; the content test barely engages, because the article is the metal. An importer of goods containing metal faces the content test as the primary question, with the rate as the consequence of failing it. A maker of mobile machinery has been placed inside the perimeter by description and has no compliance history to reason from, which is a distinct kind of risk, not a harder question but an unfamiliar one. A domestic buyer of domestic steel carries no tariff exposure at all and carries the full 15 to 25 percent pass-through, because the cost reaches it through price rather than through duty. Four positions, four questions, and none of them is answered by the import statistic.

What the coverage misses is a compound of two things, and neither is the rate. The threshold is the first. A manufacturer buying mostly domestic steel that dips below 85 percent on one sub-component can be swept inside a wall it believed itself outside, and the cost advantage priced into a quotation turns out to have been an assumption. Annex I-C is the second. By naming mobile machinery the instrument reaches goods rather than metal, so the surface it covers is larger than a metal tariff implies. Together they do what the headline cannot: they turn content claims into contractual matters. Once a threshold decides money, every supplier declaration about origin and content becomes a term somebody will have to defend, and that obligation propagates down the chain to firms that never import anything at all.

The expiry is where I would ask people to slow down. The instrument is scheduled to sunset on 31 December 2027, and industry analysts expect extension or replacement. Both statements are true, and they are not the same kind of statement. One is the text. The other is a prediction about a legislature, offered by people who watch legislatures, and it should be weighted as a prediction. Planning on a clean expiry treats a political assumption as a legal fact. Planning on an extension treats a forecast as a statute. Those two errors are symmetric in logic and asymmetric in cost, which is the part to carry into a budget meeting.

Arithmetic helps, provided the assumptions are labelled as assumptions. Take a mid-sized importer with eight million dollars a year of steel-containing imports, and assume that sixty percent of that value sits in lines where the baseline applies, so four point eight million. At 25 percent the annual duty on that slice is one point two million, or a hundred thousand a month, before anything is said about content. Assume now that a content review moves five percent of that value from qualified to unqualified, a small shift against an 85 percent line. Four hundred thousand of value changes side and the duty on it is a hundred thousand a year. Both figures are illustrative, and both are the kind of number that never reaches a headline because it is specific to a bill of materials rather than to a country.

The second example is the one I would put in front of a board, because it shows why the threshold is a decision rather than a formality. Assume a mobile machinery programme with sixty million dollars of annual entered value and twenty-two percent imported content by value, so domestic content of seventy-eight percent. The gap to the line is seven points, or four point two million dollars of value that would have to move to domestic sourcing in order to clear the test. If the article is inside the perimeter and does not clear the line, the baseline exposure is 25 percent of the entered value, which is fifteen million a year. Closing the gap and paying the duty are the same order of magnitude. That is the design: the line is set near the point where the two numbers cross, so compliance is cheaper than the tariff but only just, and only where the content can be proven.

Scale changes the answer in a way that is easy to miss. Take a small importer with one point two million dollars a year of metal-bearing purchases, and assume a quarter of that value sits in affected lines, so three hundred thousand. At 25 percent the annual duty is seventy-five thousand, about six thousand a month. The documentation exercise needed to establish content at sub-component level across a product range that size also costs real money, and it is not obvious from the arithmetic alone that the two are ordered the way a large firm would order them. For a small importer the defensible sequence is the cheap test before the expensive one. Fixing the description is a reading exercise and can be done in weeks. Assembling content evidence is a supplier exercise and takes quarters. Put them in that order, and pay for the second only where the first puts money at stake.

Consider the counterfactual in one direction. Suppose the sunset holds, the instrument lapses at the end of 2027, and imports normalise. A firm that rebuilt its sourcing to clear a line that no longer exists has bought a durable cost structure to solve a temporary problem, and the cost of that is not recoverable by re-sourcing at speed, because supplier relationships and tooling move on slower clocks than statutes do. The condition that has to hold for this branch to be the right one is that the lapse is clean and unaccompanied, which is not what the analysts expect and not what the past several years would lead anyone to expect.

Consider it in the other direction. Suppose the regime is extended, and extended with the perimeter where it is and the threshold where it is. A firm that planned on expiry now has to do the content work under time pressure, at a moment when its suppliers know it is under time pressure, and the price of that urgency is set by the other side of the table. The two errors are not symmetric, and that asymmetry is the only argument in this column that resembles a recommendation. Under uncertainty about a date, the position that does not depend on the date being right is the cheaper one to hold, because it is the only position that survives both branches.

The work that follows is a verification sequence, not a sourcing plan. The step carrying the most legal weight is to fix the description: obtain the category text that applies to each product line, record which version was relied on and when, and keep that record where an auditor would look for it rather than where a project folder would hold it. Alongside it, the content test has to be run at the level at which it is decided, which is the sub-component and not the assembly, and any line sitting within a few points of 85 percent should be treated as unresolved until evidence exists rather than as compliant until challenged.

For machinery, the date on which an article first fell inside the perimeter is itself a material fact, because exposure before and after that date are different questions. Quantification comes after description, not before, and duty exposure should be kept separate from the pass-through figure, since one is a legal obligation and the other is a commercial measurement that moves with the market.

What can be said with confidence is narrow and worth holding on to. The text as it stands imposes a 25 percent baseline across steel, aluminium and copper, names mobile machinery in a new category, and sets the content line at 85 percent. Everything past that is inference, including the inference that the regime will outlive its printed date. Do not rush to a conclusion in either direction. The 30 percent is what happened. The 85 percent is what is required. The date is what is written, which is not the same as what will be true.

↑ Back to the key points

— By Dr. Ingrid Voss