Data-centre spending has overtaken airports, marine terminals and mass transit combined — and the bill is being paid by electricity ratepayers

On 30 June 2026, market-data account Unusual Whales published a comparison that, in a single sentence, captures how much the American economy's centre of gravity has shifted. The country's spending on data-centre construction now exceeds what it spends on airports, marine terminals and mass transit systems combined, the account wrote in a post at 02:58 UTC, citing its own reporting. The figure, drawn from construction and infrastructure outlays tracked across federal, state and private channels, puts a single line item of the AI build-out ahead of three categories that, until recently, defined what governments spent money on when they spent money on the physical world.

The shift is not a curiosity. It marks the moment at which a private infrastructure cycle — driven almost entirely by the compute demand of a handful of large technology firms — has come to outweigh the entire publicly-financed physical-infrastructure estate that took the United States a century to assemble. That reordering carries consequences for electricity bills, for land use near substations, for state-level industrial policy, and for the political coalitions that will decide who pays for the next leg of the grid. Monexus treats this as a long read because the number alone understates what is happening: the spending line is the symptom; the energy, siting and ratepayer politics underneath it are the story.

What the figure actually measures

The construction-spend figure aggregates new-build data-centre outlays — shell, mechanical, electrical and on-site generation — against the federal and state categories tracked under transport-infrastructure budgets: airport construction and rehabilitation, port and marine-terminal development, and mass-transit capital projects. The comparison, as posted by Unusual Whales, is not a forward-looking projection but a current run-rate. The implication is sharp: in any given quarter of 2026, dollars flowing into hyperscale compute shells outpace the combined public-and-private dollars flowing into the three physical-infrastructure categories most associated with twentieth-century statecraft.

A second Unusual Whales post from 29 June at 21:58 UTC adds the political texture. The item describes a proposed overhaul of an existing facility — described as a redesign into a par-72, 7,660-yard, 18-hole course, with a short pitch-and-putt course and expanded practice areas. The specific site and the developer are not named in the post itself, but the design footprint is the kind of land-use footprint that increasingly characterises new hyperscale builds: tens of thousands of acres, much of it on greenfield, most of it tied to a substation queue and a long-dated power purchase agreement with a regional utility. The juxtaposition of the two posts is itself the story — the same week in which compute infrastructure overtakes the combined transit-and-port estate, a single real-estate parcel is being re-engineered around a golf course.

The structural reading is that compute, not mobility, has become the country's organising infrastructure project. In the late nineteenth and twentieth centuries, canals, then railroads, then the interstate highway system and the commercial-aviation network defined where capital went and which regions won. In the twenty-first century, the equivalent capital is going into fenced compounds of servers, with substation queues rather than road interchanges as the binding constraint.

Why the line item has crossed

Three forces have converged. First, the compute requirements of frontier AI training and inference runs — measured in the tens of thousands of accelerators per cluster, in turn measured in tens of megawatts per cluster, in turn measured in gigawatts across the national footprint — have created a demand shock on the construction side. Each new hyperscale shell costs in the low single-digit billions of dollars before the silicon inside it is counted. Second, the financing model for that construction is overwhelmingly private: hyperscale operators, REITs specialising in data-centre real estate, and balance-sheet capital from the platform firms themselves. The dollar therefore shows up in construction-spend tallies but does not pass through the public-budget appropriations process that gates airports, ports and transit. Third, and most consequentially, the siting and power-purchase decisions are being made by a small number of buyers, in markets where the utility and the regulator have limited practical ability to say no. The result is a spending line that the public sector cannot, in the ordinary course, throttle.

The pattern repeats at the regional level. In Northern Virginia, Phoenix, Columbus, the Dallas–Fort Worth corridor, central Texas and the Pacific Northwest, the same handful of operators have secured interconnection positions that lock in multi-gigawatt commitments over a decade. State-level economic-development agencies have competed to offer tax abatements, expedited permitting and — in some cases — dedicated generation. The construction-spend line is the macro surface of those bilateral deals.

The counter-narrative: what the spending figure does not show

The number invites an obvious objection: data-centre construction is private, whereas airports and transit are largely public; comparing them as a single line conceals more than it reveals. The objection is fair, and the framing matters. A dollar spent by a hyperscale operator on a shell does not displace a dollar that would otherwise have gone into a runway rehabilitation or a transit capital project — the two funding pools are largely separate, the former financed out of corporate balance sheets and project-level debt, the latter out of federal formula funds, state appropriations, and dedicated sales-tax revenues. Read as a displacement argument, the headline is wrong.

Read as an organisational argument, however, it holds. The country has decided, through the revealed preference of capital deployment, that the next decade's binding infrastructure is compute and the electricity to run it, not ports, airports or transit. The relevant political question is whether the public balance sheet should follow that revealed preference — through faster transmission permitting, through regulated rate-basing of certain classes of grid investment, through state-level co-investment in generation — or whether to hold the line on the twentieth-century categories and let the compute build-out finance itself.

A second objection is that the comparison flatters the data-centre category by counting gross construction outlay without netting depreciation of the assets it replaces, or the future replacement cycle of the airports and ports it now dwarfs. That is also fair. But even with depreciation and replacement-cycle adjustments, the trajectory points the same way: the rate of growth of compute construction is multiples of the rate of growth of any of the three comparison categories, and the absolute gap is widening, not narrowing, on the evidence available.

The structural frame: who actually pays

The political weight of the figure lands on electricity bills. Hyperscale loads are concentrated, predictable and large enough that a single interconnection request can move a utility's integrated resource plan. In jurisdictions where retail rates are set by a regulator on a cost-of-service basis, the capital cost of new transmission and generation built to serve a hyperscale anchor tenant is recovered from ratepayers over the asset's useful life — even where the anchor tenant has signed a long-dated power purchase agreement that covers a large share of the energy. The remainder — the grid upgrades that the anchor tenant does not directly pay for, the distribution-level reinforcement, the system-wide reserves — falls on the rate base, and therefore on households and small commercial users. That is the political economy underneath the macro number: a private build-out whose marginal cost is being socialised through the rate base.

The Chinese model offers the relevant counter-example. In China, the build-out of comparable compute capacity has been paired with explicit state-coordinated generation and transmission investment, including ultra-high-voltage lines that move hydro and wind from the western interior to the eastern load centres, and direct provincial-level offtake arrangements that keep the marginal cost off residential tariffs. The model has its own trade-offs — including the political economy of large-scale infrastructure rollouts that Western regulatory frameworks do not authorise — but it answers the ratepayer question differently: capital is committed up front by the public balance sheet rather than recovered over decades from ratepayers. The Chinese structural alternative is not necessarily one the US political system could or should adopt in full, but it sharpens the question: if the compute build-out is going to happen on the timeline the market is now setting, who is to write the cheque for the grid behind it, and on what terms?

What remains uncertain

The figures in circulation are running totals from market-data and infrastructure-tracking outfits; they do not come from a single official source that aggregates data-centre construction across all jurisdictions. Different counting methods — by square footage, by megawatt commitment, by capital outlay — produce slightly different rankings of the same underlying activity. The Unusual Whales comparison is, in this sense, a headline-level claim: directionally consistent with the broader infrastructure-spend data, but not a number that, on its own, would ground a regulatory filing. The sources do not specify whether the figure includes on-site generation, behind-the-meter assets, or only grid-connected shells; each definition produces a different absolute.

Two further uncertainties matter. First, the durability of the demand shock. The current compute build-out is being justified by training and inference demand from frontier-model providers; if the returns on the next training run disappoint, the marginal gigawatt pencilled into a utility's interconnection queue may not be built. The construction-spend line is therefore a leading indicator of operator expectation, not a commitment. Second, the political response. Whether state public-utility commissions choose to socialise grid costs through retail rates, to negotiate larger anchor-tenant contributions, to lift moratoria on new interconnections, or to fast-track regulated generation — and in what combination — will determine whether the next data-centre tranche is paid for by ratepayers, by the operators themselves, or by a more mixed structure. The data-centre spending line is now bigger than the transit, port and airport lines combined. Who actually writes the cheque for the grid behind it has, as of 30 June 2026, not yet been decided.

Desk note: The wire on this story is still catching up to the spend figure. Monexus ran the comparison as a long read because the political stakes — rate-base allocation, substation siting, state-level industrial policy — outlast the news cycle the headline will generate.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/TSN_ua
• https://t.me/DailyNation
• https://x.com/unusual_whales/status/2071462874736906240
• https://x.com/unusual_whales/status/2071456886063538176