An infrastructure delivery model allocates risk, capital and blame, and it reduces none of the three

An infrastructure delivery model is usually presented as a choice between ways of building the same thing. It is better read as a contract about where three things sit once building begins: the risk that the work costs more or arrives later than planned, the capital that funds it until it earns, and the accountability that attaches to whoever the public holds responsible if either goes wrong. A model moves those three between parties. It does not shrink any of them. Two can be bought and sold and one cannot, and the one that cannot is often the one deciding which model is chosen.

Where this is written from

The vantage needs stating, because it bounds every sentence that follows. I am an analyst writing inside The Cefa Brief. My materials are published statistics and public documents: annual series redistributed by FRED and retrieved for this site on 19 September 2026, together with the issuing agencies' own descriptions of what each series counts. I have not sat on either side of a delivery contract, have not seen a risk allocation matrix, and hold no project's tender papers.

What that position can do is narrow. It can read a national aggregate against its own unit and say whether a claim made on it holds. It can name what the record does not contain, which in this subject is nearly everything specific. No national annual series identifies a procurement route. None records a concession, an availability payment, a completion guarantee, a variation order or the party that paid for one. The argument below is about the shape of the instruments rather than about any project that has passed through them, and where it turns on something unmeasured the absence is stated instead of filled.

Risk is priced, never removed

Take the first of the three. Under a traditional procurement the sponsor designs the work, tenders the construction and keeps the consequences of most of what then goes wrong. Under a design and build contract the sponsor pays a contractor to accept completion risk at a fixed price. Under a concession a private party takes more still, in some structures including whether the asset earns anything once it opens. Every step hands a hazard to a party that will charge for holding it, and that charge sits inside the price the sponsor pays. Risk transfer is a purchase. A sponsor who books it as a saving has read one side of the entry.

The transfer holds only to the limit of the balance sheet standing behind it. A contractor who has accepted a fixed price and cannot meet it does not absorb the overrun. The overrun returns to the sponsor through insolvency, delay and a re-tender, and it returns late, when the work is half done and every alternative is worse than it was at award. That is the part the word transfer obscures. What moves is a claim, and a claim is worth the party it is written against. No series carried here records the financial standing of any party to a contract, so nothing in the national statistics can tell a reader whether a given transfer was real or nominal.

The capital leg is quoted, and the quote moved

The second allocation has a market price, which makes it the part of the argument most often settled by the calendar rather than by the merits. The ten-year Treasury constant maturity yield runs 0.89, 1.45, 2.95, 3.96, 4.21 and 4.29 percent across 2020 to 2025, a rise of 3.4 percentage points from the first observation to the last. The federal funds effective rate runs 0.38, 0.08, 1.68, 5.02, 5.14 and 4.21 percent over the same years, a range of 5.06 points between its lowest reading in 2021 and its highest in 2024. Neither is a rate at which a sponsor borrows. Both are annual observations of the price of public money, and both moved further inside five years than most delivery arguments assume the cost of capital can move at all. A structure chosen against 0.89 percent and delivered against 4.29 is not the same instrument twice.

Against that, the comparison actually deciding between public borrowing and private capital is a spread, and the spread is not published. Nothing held here carries a project's cost of private capital, an equity return, a rate struck at financial close or the margin on a construction facility. The record of delivery is no finer. Nonresidential construction spending runs 856, 845, 968, 1193, 1295 and 1309 billion dollars across 2020 to 2025, each point at a seasonally adjusted annual rate, so each is already a figure for a year and not for a month;one month's outlay is roughly a twelfth of 1309 billion. Inside that line there is one fall, of 11 billion into 2021, one large step of 225 billion into 2023, and a most recent step of 14 billion. None of those movements belongs to a delivery model, because the series does not record one.

The third allocation has no market

Accountability does not move. A contract can send the cost of a failure to a contractor, a lender, an insurer or a concessionaire, and the public will still address the body that commissioned the work. My view is that this asymmetry does more work in the choice of model than the delivery case does. A structure placing the financial consequence elsewhere while leaving the political consequence in place is attractive to a sponsor for reasons separate from whether it builds anything faster or cheaper. That reasoning is rarely written down, and no statistical agency is in a position to collect it.

There is a fiscal version of the same asymmetry. A delivery model can change when and where an obligation appears in a public account without changing the obligation itself. The one national measure of the burden carried here is federal debt as a share of GDP, which runs 122.3, 120.9, 118.4, 117.7, 120.4 and 120.7 percent across 2020 to 2025. It ends the window 1.6 points below where it began, with 4.6 points between its highest reading in 2020 and its lowest in 2023. That series measures federal debt. A delivery decision taken by a state authority or a municipal sponsor does not enter it at all. Nothing in the national record can therefore show whether a given structure moved an obligation or only moved where it is written down.

What follows

My position is that a delivery model should be published as an allocation and judged as one. Name each material risk, name the party holding it, name the price paid for that party to hold it, and name what happens if that party cannot pay. State the assumed cost of capital as a number with the date it was struck, because the last five years show the date doing more to the arithmetic than the choice of route does. Where a model is chosen partly because it keeps an obligation off a particular account, say so in the same document; that is a reason, and reasons survive being stated. What this vantage cannot do is rank the models. Four annual series of national aggregates cannot reach a contract, cannot date a decision inside a year and cannot tell one delivery route from another. They establish that the price of the capital leg moved 3.4 points in five years while the record of what was delivered stayed a single undifferentiated line, and that on both sides of every one of these decisions a comparison is being made that the public cannot check.