Public purchasing is an industrial instrument, and none of these series counts it as one

Every government is a buyer before it is a planner. It buys vehicles, uniforms, medicines, software, steel, buildings and the labour to install them, and it does so continuously, under rules written long before any current strategy. That purchasing is the one industrial instrument a state already owns outright. It needs no new appropriation in order to exist, no new agency, and no legislative theory about which sectors matter. It needs only a decision about what to buy, from whom, and for how long. Whether such decisions are being made is not something the statistics below record, and that silence is where this begins.

Where this is written from

This is written from the position of an analyst working only from published statistics and public documents: the federal series redistributed by FRED and retrieved for this site on 19 September 2026, and the agency notes that define what each one counts. That position is good for one thing. It can say exactly what the public record does and does not contain, and it has no stake in describing a programme more favourably than that record allows. What it cannot see is the inside of a procurement. It holds no contract files, no bid histories, no supplier accounts, and no record of which firms added capacity against which award and which did not. Nothing below rests on a conversation with a contracting officer or a supplier, because none took place. The argument is therefore about what the statistics can and cannot register, not about how any particular purchase was made. A practitioner who has run a programme office knows things about this subject that no series records, and this piece cannot contradict them.

Paying for capacity, and ordering it

A state that wants more productive capacity has two routes to it. It can pay for the capacity directly, through a grant towards a plant, a credit against equipment, or money for research. That is a transfer, and it appears as spending. Or it can promise to buy the output, in stated volumes, over a stated number of years. The promise costs nothing on the day it is made and it moves demand risk off the producer and onto the state. The first route is a payment. The second is an order. Only the second answers the question a lender actually asks, which is whether the plant will still have a customer in year seven.

The first route is the one the statistics follow most closely, though what they follow is the spending and not the payer. Construction spending on manufacturing plant ran at 75 billion dollars in 2020, then 82 in 2021, 125 in 2022, 202 in 2023 and 244 in 2024, before falling to 215 in 2025. Each point is a seasonally adjusted annual rate, so each is a yearly rate and approximates the year. The distance between 2020 and 2024 is 169 billion dollars. Research and development inside GDP, stated the same way, went from 754 billion dollars in 2020 to 1,081 in 2025, an increase of 327. Neither series states who paid, so neither separates a public transfer from a private decision to build.

The measure such spending is usually judged against has barely moved. Manufacturing output, indexed with 2017 at 100, reads 93 in 2020, 97.6 in 2021, 98 in 2022, 97.3 in 2023, 96.4 in 2024 and 97.3 in 2025. Eight years after the base year the index is still 2.7 points below it, and the 2025 reading sits 0.7 below the 2022 reading. Capacity in use runs in the same direction, from 77.6 percent in 2022 to 75.3 in 2025, a fall of 2.3 points. A large and visible build-out of plant sits beside an output series that is close to flat. Those two facts are not a causal finding and nothing here turns them into one. They are what the record contains.

The order book nobody opens

Private buyers run the demand-side instrument constantly, and the statistics count it when they do. Manufacturers new orders averaged 437 billion dollars a month in 2020 and 611 in 2025; each annual point is the average of that year's months, not the year's total. Core capital-goods orders averaged 62 billion a month in 2020, reached 77 in 2022, fell to 75 and then 74, and returned to 77 in 2025, the same figure as three years earlier. Those series are watched because an order is a commitment and a commitment can be financed against. Nothing equivalent exists for what the state orders. Public purchases enter the national accounts as an expenditure category, and none of the series used here breaks them into what was bought, from which producers, or under commitments of what length.

Why the instrument resists the use

Procurement rules were built to solve a different problem. Their purpose is to stop public money being steered to friends, and they pursue it by forcing competition on price and by making every award defensible after the event. An industrial commitment has the opposite shape. It selects a producer before that producer has proved cheapest, accepts a higher price for a period, and holds the award through years in which a rival is quoting less. An official who does that is exposed on every point the rules exist to test, and the exposure is personal while the industrial benefit is diffuse. What the system rewards is the lowest compliant bid, awarded annually, cancellable.

Duration is the second obstacle, and it is arithmetic rather than culture. Industrial capacity is financed across decades. The ten-year Treasury yield, the rate against which long commitments are discounted, stood at 0.89 percent in 2020, 1.45 in 2021, 2.95 in 2022, 3.96 in 2023, 4.21 in 2024 and 4.29 in 2025. Against a discount rate of that size, a promise to buy for one year, renewable at the buyer's discretion, is worth very little to the party being asked to build. A promise to buy for ten is worth a great deal. Where appropriations run annually, an annual appropriation cannot by itself make a ten-year promise, and no amount of strategic language closes that gap.

What is not counted is not governed

A third obstacle is the one this position sees most clearly. A programme is managed against the figures published about it. Plant spending has a series. Research inside GDP has a series. Output, capacity in use and orders each have one, revised on a schedule, with a named agency behind them. The industrial content of public purchasing has none of that in the material used here. No line states how much of what was bought last year was made domestically, or what share went to producers who added capacity because of the award. An instrument with no published measure cannot be evaluated, cannot be defended when it costs more than the alternative, and cannot be corrected when it fails. It ends up being administered as housekeeping, because housekeeping is the only thing the reporting can confirm it did.

None of this establishes that procurement run as industrial policy would work. It would cost more per unit than the alternative, by construction, and what it bought would be a capability rather than a price. Whether that trade is worth making is not a question these series can settle, and an analyst without the contract files cannot settle it either. The narrower point is harder to argue with. The spending that is counted rose steeply and the output index is close to where it started. Whether the instrument that is not counted was used as an instrument is something these series cannot say either way, and the absence of a measure is the reason they cannot.