The permit counts say permitting is not the binding constraint on American delivery

Two candidates are usually offered for what holds American building back: the time it takes to get permission, and the price of the money. I take the second, and the evidence I would put first is not a rate. It is the count of permissions themselves. Housing units authorised by permit have fallen in every year since 2021. The annual rate was 1734 thousand units in 2021, then 1685 in 2022, 1516 in 2023, 1474 in 2024 and 1431 in 2025.

A binding constraint produces a queue. Were procedure the thing that stops American building, the count of permissions granted would sit against a ceiling with applications stacked up behind it, and the number would be flat. It is not flat. It has fallen four times in a row, by 49, then 169, then 42, then 43, which comes to 303 thousand units a year below the 2021 figure, just under 17.5 per cent of it. A ceiling does not descend.

What moved against the people who would have been applying is the price of borrowing. The thirty-year fixed mortgage rate averaged 2.96 per cent in 2021, 5.34 in 2022, 6.81 in 2023, 6.72 in 2024 and 6.6 in 2025. Twice the 2021 average is 5.92, so every reading from 2023 onward is more than double the rate at which the permit series peaked. Building moved with it. Housing starts, which are stated at the same annual rate but in millions of units rather than thousands, ran at 1.6 million in 2021 and 1.36 million in 2025.

The category that should have been impossible

Now take the case that runs the other way. Construction spending on manufacturing plant went from an annual rate of 75 billion dollars in 2020 to 82 in 2021, 125 in 2022, 202 in 2023 and 244 in 2024. That last figure is 169 billion a year above the first and 3.25 times it, to two decimal places. Across those same years the federal funds effective rate climbed from its 2021 low of 0.08 per cent to 5.14 in 2024, a move of 5.06 points and the dearest money in the window.

Every one of those factories passed through American permitting: environmental review where it applied, state siting, local consent. None of it was suspended while the spending line was tripling. In 2022, 2023 and 2024 that line rose in every year the permit count fell. If procedure were the binding constraint, this is the category that could not have happened. What separated plant from housing was not that one of them found a faster path to permission. It was what the finished asset could expect to earn, which federal policy moved deliberately and which this brief holds no dataset to quantify.

Where this argument is weakest

The permit series counts authorisations issued. It does not count applications filed and it records no waiting time at all. So I am arguing from the absence of a queue in a series that could not show me one, and that is the weakest joint in the case. I would rather say so than write past it.

That weakness also tells me what would change my mind. A published count of applications received against decisions issued, with a median time to decision, at the monthly frequency on which the Census already publishes permits, would settle the question in either direction. Nothing in the files this brief works from carries such a series. The permitting case is argued from examples; the capital case is argued from six years of prices. That asymmetry is not proof, but it is the reason the two arguments are not being held to the same standard.

A second weakness sits inside my own account of causation. Plant spending fell in 2025, from 244 billion dollars a year to 215, a fall of 29 billion and the only decline among the five steps the series holds. The policy rate came down that year, from 5.14 per cent to 4.21. Cheaper money, less building. A simple story in which rates rise and building stops does not survive that observation, and I am not going to defend one.

What survives is narrower. The quantity that binds is the gap between what a finished asset is expected to earn and what its capital costs, and either side of that gap can move it. Between 2022 and 2024 the cost of capital rose hard and lost anyway, because the expected return on plant rose harder. In 2025 the cost of capital eased slightly and plant spending fell, which says the return side moved the other way. Both years point to the same variable, from opposite directions.

Nor has expensive money stopped the wider build. Nonresidential construction spending, the series that carries manufacturing plant and much else besides, reads 856 billion dollars a year in 2020, 845 in 2021, then 968, 1193, 1295 and 1309 in 2025. It is not a measure of public infrastructure and cannot be broken into categories here. One fall at the start and four rises after it, ending 464 billion a year above the 2021 reading, with 14 billion of that added in the most recent year. Dear money did not halt that. Neither did procedure.

What follows if capital is the one

Permitting reform, on this account, should be appraised as a financing measure and priced like one. A consent shortened by a year is worth the carrying cost of the capital that would otherwise sit idle for that year, plus whatever the schedule shift does to the timing of the revenue. Put that way the case is often a good one and I would support it. Put as projects unlocked, it is a claim about a queue that nobody has counted.

Public money, meanwhile, buys more on the return side than on the cost side. The plant figures are the demonstration available: the lever that moved was the income the completed asset could expect, not the speed of the consent. For public infrastructure the equivalents are availability payments, offtake commitments and revenue certainty, together with one unglamorous item that is usually left unallocated, which is who carries the interest rate between award and financial close. A sponsor who wins a procurement at one rate and reaches close at another has absorbed a cost that no permitting reform touches.

The last thing that follows is a request for a number. An agency confident that its consenting process is not the binding constraint should be willing to publish applications received, decisions issued and days elapsed, monthly, as a series. So should one that believes the opposite, because the number would prove its case. The absence of that series is the most striking fact about a debate that has run for years on the claim that a queue exists.

None of this makes permitting costless or reform pointless. Delay is real, it is expensive, and shortening it is worth doing on its own terms. The claim I am defending is about which constraint binds first, and on the evidence that can be checked it is the price of money set against the return that has to beat it. The permit counts have fallen for four straight years. Nobody stopped issuing them.