When the servers come to town: the data-centre build-out is rewriting the local politics of American construction

On 30 June 2026 the data journalism outfit Unusual Whales put a single line on the timeline that should make every county auditor in the United States sit up. The country's projected spend on data-centre construction this year, the post argued, now exceeds what the country spends on airports, marine terminals and mass-transit systems combined. The figure was not a forecast from a research boutique; it was the kind of round number that anchors a shift in how America allocates concrete, copper and grid capacity. The arithmetic has been creeping into public discourse for at least two years, but 2026 is the year the spending line crossed the others. The question is what kind of country it leaves behind.

This publication finds that the data-centre build-out is no longer a side-show in the American capital-expenditure story. It is the main act. The projects are concentrated in fewer than two dozen counties, the capital is largely foreign-sourced, and the operating model — power-purchase agreements, accelerated depreciation, ten-year tax abatements — has turned local property taxes into the industry's quietest subsidy.

The shape of the spend

The line in the headline is structural, not cyclical. Hyperscalers and the large model labs they now serve need warehouse-scale buildings, dedicated substations, cooling plants and fibre back-haul. Each megawatt of IT load pulls with it, by industry rules of thumb, roughly 0.5 to 0.8 acres of building and a comparable envelope of land for the substation and chiller yard. Multiply that across the 5–10 GW pipelines that Northern Virginia, Phoenix, Columbus, Dallas and the Atlanta–Augusta corridor have been advertising, and the country is on track to commit several hundred billion dollars of construction-cost alone to a single asset class in a single calendar year.

Unusual Whales's framing of the figure is precise: airport, marine-terminal and mass-transit outlays combined. Public-works spending of that scale has historically been the core business of federal infrastructure programmes. The fact that private, foreign-financed server campuses now exceed the sum of all three is, on its own, a marker of how thoroughly the centre of gravity in American capital expenditure has shifted toward digital infrastructure and away from the assets that move humans.

The construction itself is heavy. Hyperscale shells are steel-framed tilt-up or pre-engineered metal; the heavy engineering is in the electrical room, the chiller yard and the medium-voltage distribution. The domestic supply chain for large transformers, switchgear and back-up generation has been on allocation for most of 2025 and 2026. So has the supply chain for the cement and the engineered steel used in the slabs. In other words, the build-out is not just absorbing capital. It is competing for it — with housing, with grid reinforcement, with the reshoring of advanced manufacturing.

The counter-narrative: it's just buildings

The pushback, heard in the trade press and on hyperscaler earnings calls, runs roughly as follows. Data centres are taxable real property. Once operational, they pay county real-estate and personal-property taxes, plus sales tax on the construction-period inputs. They employ hundreds of construction workers at peak and dozens of full-time technicians once live. They do not send their product to a landfill and they do not put children in school. The build-out, on this telling, is the closest thing American fiscal policy has had to a free lunch since the 1950s.

There is real evidence behind that case. Loudoun County, Virginia — the historical centre of the Northern Virginia cluster — has used data-centre tax revenue to hold down residential property-tax rates while still funding school construction. The trade-off has been visible and local: voters know exactly which buildings paid for which schools. Whether the model survives the next decade of larger, more power-hungry campuses is the open question.

The argument from the industry's critics, sharpened in places like the Princeton SPIA working group and now surfacing in a growing number of state-legislative hearings, runs differently. It is that the up-front fiscal bargain has been engineered. Counties grant ten- or fifteen-year tax abatements to attract the site, the project draws in a transient construction workforce that does not enrol children in local schools, and once the abatement expires the site typically applies for an extension under the rubric of "competitive disadvantage." The recurring complaint is not that data centres pay nothing — it is that the public ledger cannot reconcile what the buildings are assessed at with what they actually cost to wire, cool and serve with upgraded transmission.

Structural frame: when the user is the customer and the product

What the build-out is doing to the country's industrial map is harder to see than the construction permits, because the asset is not a building in the conventional sense. A data centre is a piece of the grid that happens to have a roof. The political economy of the United States for the last hundred years has assumed that the largest load on the system would be a mill, a refinery, a smelter or — most recently — an electric arc furnace for rebar. Each of those anchors a regional economy, a workforce and a tax base.

The hyperscale data centre is a different sort of anchor. Its throughput is rented, not sold. Its consumer is, by design, the user of an application that the consumer did not pay for. The landlord of the load is a foreign-headquartered platform firm whose relationship with the locality is contractual rather than civic. The county in which the building sits is selling, in effect, a finite physical resource — substation capacity, water for cooling, the right of way for new transmission — to a counterparty that is free to leave when the power price moves against it.

The deeper question, and the one that the industry is going to be forced to answer in the back half of this decade, is what the United States is buying in exchange for the concrete. The post-Cold War assumption was that dominance in digital infrastructure translated into dominance in finance, media and the platform economy. That assumption is now being tested in two directions at once. One is the concentration of the cloud market in three to four buyers and the corresponding concentration of bargaining power over electricity supply. The other is the rate at which the marginal unit of compute is being shifted abroad, to jurisdictions with cheaper power, looser water rules, and explicit industrial-policy courtship of the same customers.

The local balance sheet

For the counties that host the buildings, the arithmetic plays out on a much shorter ledger. A typical hyperscale campus of 250 MW IT load in the Eastern Interconnection will draw roughly 300 MW of utility supply at peak. That is the equivalent load of a mid-sized American city. The local utility has to build, finance and rate-base the generation, transmission and substation to serve it — most of which the ratepayer base of the same utility will eventually subsidise, because the cost-of-service model for regulated utilities does not differentiate cleanly between a load that is here for thirty years and a load that is here for a lease.

The trade-off that Loudoun and its peers made in the early years was rational at the time. A 250 MW load attached to a constrained part of the PJM grid is the most valuable load a utility can pick up, because it is large, flat, and willing to be curtailed in exchange for a tariff discount. The construction-era sales tax, in jurisdictions that collect it, covers the local government's incremental service demand. The long-tail risk is the post-abatement step-up — whether the assessed value of the equipment inside the building, when the abatement rolls off, is anywhere close to the replacement cost the utility will have incurred on its side.

What is changing is the cost of the power. PJM's capacity auction in 2025 cleared at a record, and the 2026 clearing price is widely expected to be higher still. The data-centre campuses, as the largest single buyers in those auctions, are now in the position of pricing their own power upward through their own demand. The domestic policy response has been two-pronged: federal incentives for siting behind-the-meter generation, including gas turbines and small modular reactors, and state-level moratoria on new connections in jurisdictions where the grid is already saturated. The first response privatises the cost of the build-out. The second response externalises it.

Stakes and what to watch

If the trajectory continues at the rate Unusual Whales is flagging, three things become likely by the end of the decade. The first is that data-centre construction will be the dominant non-residential line in the United States's gross fixed-capital formation, surpassing commercial real estate, healthcare construction, and possibly the combined value of all manufacturing plant. The second is that the federal-state bargain on infrastructure will be rewritten around the assumption that the largest loads on the grid are no longer manufacturing. The third is that the country's dependence on a small number of foreign-domiciled tenants for a strategically decisive asset will look, from the vantage point of 2028 or 2030, like the kind of single-buyer risk that earlier generations of American industrial policy tried to design out.

The decisions that determine whether any of that comes to pass are not being made in Washington. They are being made by county boards of supervisors in Loudoun and Prince William, by city councils in Quincy and West Texas, by the rate-payers who will fund the substations, and by the project-finance teams at the banks that underwrite the bond issues for the buildings. The capital is global. The cost is local. The narrative is still being written.

Desk note: this article is built around a single framing line from Unusual Whales on 30 June 2026 — that US data-centre construction outlay is now larger than the combined spend on airports, marine terminals and mass-transit. The wire coverage this week is dominated by separate threads; Monexus has elevated the construction-spend thread into a structural piece because the cross-asset comparison is a more durable signal than any single project announcement.