Productive capacity names a quantity that no series on this site reports

Productive capacity is the quantity most often promised and least often defined. A government says it will build it. A company says it has added it. An analyst says a country has lost it. In all three sentences the word behaves like a stock: something there is a certain amount of, which can be raised by a stated sum and counted afterwards. Nothing in the published record behaves that way. The file behind this site holds fifty-eight annual series, each with its own source line and unit string, and one of the fifty-eight carries the word capacity in its title: Manufacturing capacity utilisation, series MCUMFN, from the Board of Governors of the Federal Reserve System. Its unit is Percent of capacity. It is a share, and a share is a statement about a denominator that is not in the file beside it.

Where this is written from

I am an analyst writing inside The Cefa Brief, working only from published statistics and public documents: the fifty-eight annual series in the file behind this site, redistributed by FRED and retrieved on 19 September 2026, each carrying its own source line and unit string. That position sees one thing clearly. It can read every title and unit in the file, count how many of them name capacity, and set out what each of the measures that stand in for it does and does not count. It cannot see the thing the argument is about. I have no sight of an order book, a shift roster, a bottleneck on a line, or an engineer's judgement about what a plant could run at if it were asked to, and none of the internal construction of the utilisation ratio, whose denominator is not published in this file. So I cannot say what American productive capacity is. I can show that the published record does not say either, and set out what it says instead. Where that distinction is carrying the argument, the text says so rather than let the figures appear to do work they cannot do.

A share of something the file does not carry

Manufacturing capacity utilisation runs 72.9 percent in 2020, 77.3 in 2021, 77.6 in 2022, 76.6 in 2023, 75.4 in 2024 and 75.3 in 2025. Each point is an annual average of a percentage and needs no further qualifier. The high is 77.6 in 2022; the 2025 reading sits 2.3 points below it, and the whole six-year range is 4.7 points. Read as a share, that is a complete statement and an unusually clean one. Read as capacity, it is half a statement. The series says what proportion was used and leaves what it was a proportion of outside the file.

Put an output series next to it and the halves do not close. Manufacturing output, series IPMAN, is an index on a 2017 base of 100: 93 in 2020, 97.6 in 2021, 98 in 2022, 97.3 in 2023, 96.4 in 2024 and 97.3 in 2025. Between 2022 and 2025 it gave up 0.7 index points. Industrial production, series INDPRO, on the same base, moved from 101 in 2022 to 101.2 in 2025, a rise of 0.2 points across three years. Over those same three years the utilisation ratio gave up 2.3 percentage points. Those are different units and I have not divided one by the other; nothing here computes a capacity index. What can be said is that three Federal Reserve series describing the same activity moved by different amounts over identical years, and the quantity that would reconcile them is the one the word capacity is meant to name.

Four measures stand in for the missing one

When capacity is claimed, four published measures are produced in support of the claim, and they answer four different questions. Output is the first, and the two indices above are it. Spending on plant is the second. Construction spending on manufacturing plant, series TLMFGCONS, is stated in billions of dollars at a seasonally adjusted annual rate, so each point is a yearly rate and approximates the year. It runs 75 in 2020, 82 in 2021, 125 in 2022, 202 in 2023, 244 in 2024 and 215 in 2025. The 2024 figure stands 169 billion dollars above the 2020 figure and is 3.25 times its size, dividing 244 by 75 and rounding to two decimal places. This is the series most often held up as capacity arriving. It is a flow of money into structures, and a structure is not yet a line running.

Headcount is the third. Manufacturing employment, series MANEMP, in millions of workers, reads 12.1 in 2020, 12.3 in 2021, 12.8 in 2022, 12.9 in 2023, 12.8 in 2024 and 12.6 in 2025. It peaked in 2023 and sits 0.3 million below that peak in 2025, while remaining 0.5 million above where the window opened. A payroll count records who is employed, not what they are able to produce. Output per hour is the fourth. Nonfarm business labour productivity, series OPHNFB, on a 2017 base of 100, reads 109.1 in 2020, 111.4 in 2021, 109.8 in 2022, 112.1 in 2023, 115.4 in 2024 and 118 in 2025, adding 8.9 index points across the window with a fall of 1.6 points in 2022 in the middle of it. That is the measure which would raise what a country can make without a single new building, and it is the one least often cited when capacity is being claimed.

A definition would have to be chosen

A figure for capacity cannot be read off a factory floor. It requires decisions taken before any measurement begins: how many shifts count as normal, whether a line is capable when its inputs are unavailable, at what price the marginal unit is worth making, and how long a rate must be sustainable to count at all. Those are conventions. A convention is chosen rather than observed, and whoever chooses it has settled the answer in advance. The utilisation ratio embeds one, because a share has to be a share of something defined, but this file carries the ratio and not the convention behind it. That is a limit on what I can see from here. It is also the reason the term stays loose, since the argument about the number would be an argument about the convention, conducted by parties with positions in it.

Why the term survives

That is the condition under which the term survives. Plant spending rising from 75 billion to 244 billion at a yearly rate supports the claim that capacity was built. Manufacturing output at 97.3 in 2025 against 98 in 2022 supports the claim that it was not. Employment peaking at 12.9 million in 2023 and standing 0.3 million lower in 2025 supports a third reading again. Each of those three parties can cite a real series, in its correct unit, from a named agency, and still disagree completely, because none of them is citing the quantity in dispute. A word with no series attached cannot be contradicted by one, and this word is wide enough to hold four measures that point different ways.

One sector looks like an exception and is worth stating carefully. U.S. semiconductor production, series IPG3344S, on a 2017 base of 100, reads 121 in 2021, 131 in 2022, 135 in 2023, 144 in 2024 and 165 in 2025: 44 index points added, which is 36.4 percent of the 2021 reading, dividing 44 by 121. The largest single step, 21 points, is the most recent one. The wider computer and electronic product output index, series IPG334S, went from 105.7 in 2020 to 126.8 in 2025. Those two index lines are steeper than any other index quoted here, though the semiconductor series is carried in this file only from 2021 and so covers one year less. The temptation is to read them as capacity installed. They remain output. A sector can lift output by running plant it already had, and an index of production does not separate that from plant that is new.

What to ask of a capacity claim

Three questions make a capacity claim checkable, and none of them requires a statistic that does not exist. Which published series is being cited. In what unit, stated as the source states it, since an index, a level and a yearly rate are three different objects and a reader cannot tell them apart from the number alone. And for which year, since the newest reading in this file is 2025 and several of these series turned during 2024. A claim that survives those three questions is a claim about something measured. A claim that does not is a claim about a word, and the word has no quantity behind it.