Chip output rose 4.8% last quarter. Chip prices fell 1.5%.

Over the three months to June, US semiconductor output rose 4.8% while the producer price index for semiconductors fell 1.5%. Volume and price have now pulled in opposite directions for three straight years — output up 34.3%, prices down 6.0%. The gap against the rest of American manufacturing is the striking part: producer prices across all manufacturing industries rose 9.4% over the last year while semiconductor prices fell 0.6%. In an economy where industrial goods are getting dearer, chips are the exception, and they are the input everything else is being built out of.

The two series, and the one caveat that has to come first

The volume figure is the Fed’s industrial production index for NAICS 3344, semiconductors and other electronic components — which includes printed circuit assemblies, capacitors, resistors and connectors. The price figure is the BLS producer price index for NAICS 334413, semiconductors and related devices only, a sub-industry of the first.

The two are not the same population. A −1.5% price change and a +4.8% volume change are being read across slightly different industry boundaries, and no published pair of monthly series lines up exactly. The direction of the divergence is far too large to be an artefact of that mismatch — 34.3% against −6.0% over three years is not a classification effect — but the precise size of the gap carries that qualification.

Two more mechanics matter. Industrial production is seasonally adjusted; the producer price index is not. And the PPI is revised once, roughly four months after first publication, so the most recent months on this chart are the least settled.

The quarterly decline is really two months

The −1.5% over three months is true and slightly misleading, and the monthly path shows why.

Prices rose through April, to 30.602 — the highest reading since November 2024, seventeen months earlier — then fell in both of the two months that followed. A three-month window ending in June starts in March and so captures the entire fall while missing most of the rise. Measured over six months instead, the same index is +0.09% — flat.

That does not dissolve the finding; it relocates it. What happened is not a quarter of steady erosion but a sharp two-month reversal, down 3.0% from the April peak. And June’s −1.55% is genuinely unusual for a June: across 59 years of this not-seasonally-adjusted series, the average May-to-June change is −0.02% and the median is exactly zero. A −1.55% June is not seasonal.

Nor is the quarterly number extreme by this series’ own standards. A −1.45% three-month change sits at the 26th percentile of the 710 three-month windows since 1967. Roughly one quarter in four has been worse. This index has been falling for most of its life.

Chips are deflating inside an inflating manufacturing sector

The comparison that gives the number its weight is not with the chip industry’s own history. It is with everything else being manufactured.

Producer prices across US manufacturing rose 9.4% in twelve months. Semiconductor prices fell. The relative price of a chip against the average manufactured good has dropped about ten per cent in a year and roughly eighteen per cent in three.

The longer arc is steeper still. The semiconductor index is 70.3% below its December 1998 base, down 18.2% over ten years and 55.4% over twenty. June’s 29.695 sits 1.0% above the all-time low of 29.392 set in August 2025.

Falling prices are not falling revenues

The most common misreading of this pair is that a deflating price index means a shrinking industry. Multiply the two series together and the opposite appears: a rough nominal output proxy, output index times price index, is up 26.2% since June 2023 — because a 34.3% volume gain swamps a 6.0% price decline.

Volume growth of this size makes the price decline affordable at the industry level. It does not make it costless at the plant level, and that is where the two datasets meet. Capacity utilisation across NAICS 3344 is 72.2%, down from 92.2% in February 2022. Producers are absorbing lower unit prices and spreading fixed costs across emptier lines. Those two pressures compound.

Buyers get the mirror image. Anyone whose input is compute — data-centre operators, model developers, carmakers, industrial OEMs — is paying less per unit for more units, in a year when nearly every other manufactured input costs 9.4% more. That is the most concrete, verifiable subsidy running through the AI build-out, and it comes from the semiconductor industry’s own balance sheet rather than from anyone’s policy. The same asymmetry showed up in the electricity and capex data: the gains from this build-out are concentrated and visible, the costs diffuse and carried by whoever has least pricing power.

How much of the fall is price, and how much is a better chip

This is the boundary of what the index can say. BLS quality-adjusts the microcircuit indexes hedonically — modelling introduced for desktop microprocessors with February 2018 data and for server microprocessors from that July. A part sold at an unchanged dollar price with materially more performance registers in the index as a price decline.

So “chip prices fell 1.5%” means the price of a constant unit of semiconductor performance fell 1.5%. It does not establish that any invoice got smaller. Both readings are consistent with the data, and the series cannot separate them. For a buyer of compute the quality-adjusted measure is the relevant one; for a producer’s revenue line, the cash price is, and that is not published here.

Two further limits are worth stating plainly. The index cannot distinguish leading-edge logic from trailing-node parts, which have behaved very differently through this cycle. And a quality-adjusted deflator this steep flows into the national accounts, where it lifts measured real output in the industry — some of the 34.3% volume growth above is the deflator doing work, not additional wafers.

The combination is normal for this industry — and has become rare

Rising volume with falling prices is the semiconductor industry’s default state, not an anomaly. Counting every month since 1972 in which three-month output growth exceeded +4% while three-month price change was below −1%:

In the 1990s this was the industry’s condition almost half the time. In the 2020s it has occurred in 9% of months — seven in total, three of them clustered in late 2020, then January, February and July of 2025, and now June 2026.

The framing follows from that. What is happening is a return to the industry’s classic Moore’s-law configuration after a decade in which it was rare, and it is arriving at the same moment as the sharpest capacity build since 2013.

Our own Trade Stress Index reads 48.9 as of 24 July — Normal, at the 60th percentile of the last five years — with global trade activity at 10.4 on its component scale, firmly calm. Falling chip prices in this cycle are not a trade-disruption story. The index has not moved on this, and that absence is itself informative: the pressure is coming from supply the industry built itself.

What to watch

• Whether the two-month fall extends. May and June each fell about 1.5%. A third consecutive month near that pace would take the index below the August 2025 record low of 29.392 and turn a reversal into a trend.
• The 12-month gap against all-manufacturing PPI, now 10.0 points. This is the cleanest measure of how much cheaper compute is getting relative to everything else being built. A narrowing gap would mean the hardware layer’s cost advantage is eroding.
• Capacity utilisation, at 72.2%. Prices and utilisation falling together is the signature of supply running ahead of demand. Utilisation recovering while prices keep falling would mean something different — genuine cost deflation rather than an overbuild.
• The four-month PPI revision. June’s figure is a first estimate. The two-month decline that carries this piece is the part most likely to be restated.

Every figure above comes from the linked series and can be reproduced by downloading them.