A record June for chip output — but not a seven-month climb

US semiconductor and electronic component output reached 181.9 in June, the highest reading in the 654 months the Federal Reserve has published the series. The seven-month run that produced it was not seven months of climbing: output fell in February and fell harder in March, ending that month below where January had left it. The record is real and the trend behind it is real. But the number doing the most work is not production. It is capacity, which has grown at double-digit rates for four consecutive years and pushed utilisation back to where it sat in 2016.

The record is real; the run of gains is not

The whole seven-month path, in the Fed’s index where 2017 = 100, seasonally adjusted:

Two of the six monthly changes are negative. The index spent February and March going backwards, and closed March at 173.56 — below the 176.71 it had reached in January. The longest unbroken run of monthly gains ending in June is three months.

The distinction matters because this series has produced real streaks and they look nothing like this. The record is 78 consecutive rising months, ending in July 1999. Against that, a series that fell in four of the last twelve months is not on a streak; it is in an uptrend with ordinary monthly noise on top of it.

The uptrend is not in doubt. Ten of the last twenty-four months set an all-time high. Output is up 11.5% on the year, 26.8% over two years, 34.3% over three, and 111.3% over ten. Nothing here argues the boom is fictional. It argues that “seven months of climbing” describes a chart nobody has.

June itself was an ordinary month

The +0.53% posted in June is not a strong print. Since January 2020 this series has averaged +0.63% a month, with a median of +0.47%. June landed between the two — typical, and weaker than the average month of the last six years.

The record was set in May, on a +2.54% gain; June merely failed to give it back. That separation matters, because “another record” implies acceleration and the monthly path shows the opposite. April, May and June ran +1.66%, +2.54%, +0.53%. The approach to the high was losing pace.

One caveat sits under all of it. Industrial production figures are revised as source data and seasonal factors are updated, and a June print published in mid-July is a first estimate for June and a revised estimate for everything before it. The shape is unlikely to invert; the individual monthly changes will move.

Nothing else in American manufacturing is doing this

Set against the sector it belongs to, the semiconductor number reads differently again.

Total US manufacturing output grew 1.8% over three years. One sub-industry inside it grew 34.3%, from a base already at record levels.

Total manufacturing also remains below its own 2017 base, at 98.70. Nearly a decade on, the aggregate has not recovered its reference year, while the chip industry has nearly doubled off it. This is not a manufacturing recovery. It is a single-industry event sitting inside a stagnant sector, and it means the headline manufacturing statistic is increasingly hostage to one industry’s cycle.

Capacity is the number that actually moved

Capacity utilisation across NAICS 3344 was 72.2% in June, the 18th percentile of all 654 months since January 1972. In March it touched 71.0%. The industry ran at 92.2% as recently as February 2022, so the entire shortage-era tightening — twenty points of it — has been given back in four years.

The context most commentary drops is that this is not even unusual for the industry. Utilisation sat below June’s level for 22 straight months in 2012–14 and for 16 more in 2015–17. Before 2026, the last month below 72.2% was February 2017.

Record output at low utilisation is arithmetically possible only one way: capacity is growing faster than production. Utilisation is output divided by capacity, so the capacity index falls out of the two published series directly.

On annual averages, implied capacity has grown +14.4%, +12.0%, +16.1% and +13.6% across 2023, 2024, 2025 and the first half of 2026. In the four years before that — 2019 through 2022 — the same measure ran +9.9%, −0.8%, −3.4% and +6.3%. The industry spent the shortage years adding almost nothing and the four years since adding at double-digit rates.

The 2013–16 precedent is the useful one. Capacity then grew 19.3%, 11.5%, −0.9% and 13.8% year on year while utilisation sat between 66% and 73%. It took until 2018 for utilisation to recover to 81.6%, and demand did eventually arrive. That episode is a reason to be careful about reading today’s 72.2% as distress — and also a reminder that the gap took four years to close.

What that costs, and whom

Low utilisation alongside rising volume redistributes; it does not destroy.

It lowers cost for buyers. Semiconductor producer prices are down 6.0% over three years while output rose 34.3%. Anyone consuming chips — data-centre operators, carmakers, industrial equipment builders — is buying more units at lower unit prices. That is the cleanest gain in the dataset, and the same signature the electricity and capex data showed earlier this month.

It raises cost per unit for producers. A fab is close to pure fixed cost; depreciation on a leading-edge line runs whether or not wafers move through it. Utilisation falling from 81.8% to 72.2% over three years spreads a larger fixed base over a proportionally smaller share of it. Volume growth of 11.5% a year is fast, but it has been slower than the capacity arriving, and that gap is where fab-level margin pressure lives even in a record year.

It changes what a shortage would mean. In 2021 and 2022, above 90% utilisation, an outage at one large facility had nowhere to go. At 72.2% the system carries roughly 28 points of headroom on the Fed’s measure. Whether that headroom is usable is a different question — idle trailing-node capacity does not substitute for scarce leading-edge logic, and this index cannot tell the two apart. The aggregate cushion is nonetheless far larger than three years ago.

What the data cannot settle is why capacity grew this fast. The series records that it did. It says nothing about how much came from federal incentives, how much from private commitments made during the 2021 shortage, and how much is foreign capacity relocating. Attribution needs plant-level data that none of these series contain.

Our own Trade Stress Index reads 48.9 as of 24 July — Normal, at the 60th percentile of the last five years — and its global trade activity component scores 10.4, the calm end of the scale. Goods are moving, and nothing in the trade-friction data is obstructing this build-out. The tension in this story is domestic capacity, not cross-border flow.

What to watch

• Utilisation, at 72.2%. The five-year low is 71.0%, set in March. A sustained break below it while output keeps rising would mean capacity is still outrunning demand. A recovery toward 78–80% would mean the new lines are filling, as they did after 2016.
• The monthly output path, not the record count. A record is a low bar for a series that has doubled in ten years. The informative number is whether monthly gains hold near the +0.63% post-2020 average. June’s +0.53% did not.
• Implied capacity growth, running near +13% a year. Four consecutive double-digit years is already longer than the 2013–14 build. A fifth would put the utilisation gap beyond anything in this series outside a recession.
• Total manufacturing at 98.70. If the aggregate stays below its 2017 base while NAICS 3344 keeps setting records, US industrial growth is concentrating further into one industry.

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