The Border Price

What the Sales Figure Contains

The figure is compiled for the association by World Semiconductor Trade Statistics and is a three-month moving average of nominal sales, so it carries no adjustment for what was sold or at what quality. Growth on a year earlier was reported at 171.3 percent into the Americas, 134.9 percent into Asia Pacific and other regions, 123.6 percent into China, 85.2 percent into Europe and 50.8 percent into Japan.

The composition is not in dispute. Memory is where the price has moved: contract prices for conventional dynamic random access memory and for flash have been reported rising by double digits in every quarter of 2026 after larger increases at the start of the year, and the producers have said publicly that their capacity is committed. This note does not use those reported contract prices in any exhibit. They are proprietary, they are quoted at varying degrees of remove from an actual transaction, and there is a better instrument available.

The better instrument is that a price increase in a component is a cost increase for everybody who buys the component, and the United States operates three statistical programmes that measure exactly that at three different points: what crosses the border, what leaves an American factory, and what a household pays. All three are published monthly. All three are free. Between them they say something the revenue figure cannot.

The Price Is Being Paid at the Border

The Bureau of Labor Statistics prices imports and exports on the same industrial classification, which means the price of what America buys in a category and the price of what it sells in the same category can be set beside each other.

For ten years the import price index for semiconductor and other electronic component manufacturing sat between 74.9 and 80.1. In the twelve months to July 2026 it went to 99.4, a rise of 32.7 percent, and it is now within a point of where it stood in December 2005. Over the same twelve months the export price index for the identical classification fell 4.1 percent, and the narrower semiconductor and related device line inside it fell 5.4 percent. The domestic producer price index for semiconductor and related device manufacturing fell 2.7 percent, which is roughly what it does in an ordinary year.

Two features of how these indexes are built dispose of the first two objections a reader will raise. The Bureau states plainly that the prices used to calculate them exclude duties, and that the majority of import prices are quoted free on board at the foreign port. Nothing in the import series is a tariff. The presidential proclamation of 14 January 2026, which imposed a 25 percent ad valorem duty on a narrow category of advanced computing chips and derivatives with effect from the following day, does not appear in this index at all. What appears is the price the foreign seller charged before any of that was added.

The second is quality. The Bureau states that its international price programme attempts to hold the quality of the items being priced constant so that pure price change can be isolated. A 32.7 percent increase is therefore not a reader being shown a faster chip and told it costs more. It is, on the agency's own method, the same thing costing a third more.

Nothing Else Imported Moved

The remaining objection is that this is not about semiconductors at all: that a weaker dollar, or freight, or a general import inflation would move any index of goods bought from the same places. That is testable, because the Bureau prices four other categories inside the same computer and electronic product family.

Audio and video equipment rose 1.2 percent, laboratory instruments 1.1 percent, communications equipment 0.7 percent and computer terminals and parts 0.7 percent. These are goods assembled in substantially the same places, shipped on substantially the same vessels and paid for in the same currency. The aggregate for all computer and electronic product imports rose 9.4 percent, and it rose because the first bar is carrying the other four.

Whatever repriced, it repriced in one category and did not touch the goods next to it. That is not what a currency move looks like, and it is not what a freight or tariff-driven general increase looks like either.

Where It Has Reached

A cost that lands at the border has to go somewhere. It can be absorbed by the assembler, absorbed by the brand, or charged on. The next two exhibits date the point at which it began to be charged on.

The producer price index for electronic computer manufacturing, which is the price at which an American factory sells a computer, rose 0.22 percent over the eleven months from August 2025 to June 2026. In July 2026 alone it rose 2.71 percent. Against a July 2025 reading that was itself elevated, the twelve-month change is 2.2 percent.

At the household end the consumer price index for computers, peripherals and smart home assistants was 3.9 percent above a year earlier in July. In the twenty-eight July-on-July comparisons available since 1999 there have been two increases, 2021 and this one, and this one is the larger. The index is published on a December 2007 base and reads 36.641, so a constant-quality computer still costs about thirty-six cents in the dollar of what it cost then. The direction has changed; the level has not been undone.

That the consumer index had turned was noticed before this note. Counterpoint Research published the observation on 30 June, working from the April reading. It is recorded here as the end of a chain rather than as a finding.

Investment Implications

The firm reads four things from this. First, that this is a cost event with a date on it, and the date is recent. The producer and consumer prints for July 2026 are the first month in which the pass-through is visible in an American statistic. Any hardware gross margin underwritten on trailing twelve-month figures is underwritten on a period that contains almost none of it, and the distance between the border figure and the factory-gate figure is the part that has not yet been allocated to anybody.

Second, that exposure is not the useful distinction here, because everybody who assembles electronic hardware has it. The distinction that will separate one result from another is whether the memory was contracted before the repricing or is being bought into it, and that is a contract-structure question which almost no filing discloses in a form that can be read off. It is answerable company by company from inventory disclosures and purchase commitments, and that is the work rather than a screen on end markets.

Third, that the tariff arithmetic is smaller than the price arithmetic. The January proclamation set a 25 percent duty on a narrow category. The index that excludes duties altogether moved 32.7 percent across the whole category. Policy aimed at the terms on which chips enter the United States is being debated at a scale below the one at which the sellers have already moved the price, and a position built on the tariff is a position on the smaller of the two numbers.

Fourth, that the location asymmetry is real and is about production rather than about ownership. What crosses into the United States repriced by a third and what crosses out did not. That is a statement about where the constrained capacity sits, and the constrained capacity is not in the United States. It carries no implication about any listed company's results, for the reason set out above.

The work this note has not done, and which would turn a reading into a position, is the supply side: the announced capacity additions with their commissioning dates, against the share of output committed under multi-year agreements. That is what decides how long the border price stays where it is, and none of the four exhibits here speaks to it.