Data centres are eating the American balance sheet — and nobody is asking who picks up the bill

There is a number doing rounds in the trade press this week, and the political class in Washington is studiously ignoring it. Per a 30 June 2026 brief from Unusual Whales, US data-centre construction spending has now surpassed what the country lays out on airports, marine terminals and mass transit systems — combined. That single line, more than any analyst note or earnings call, explains the shape of the next decade of American industrial policy. The boom is real. The bill is coming. The question is who gets handed the invoice.

The narrative sold to voters and to markets is comfortable. Hyperscalers — the cloud and AI platforms building out model training capacity — are spending their own cash, on their own land, hiring their own electricians. The state has merely stood aside, signed a few tax-incentive packages, and let private capital do what it does best. It is a tidy story. It is also incomplete in ways that should alarm anyone who pays a utility bill.

What the headline number actually measures

Construction spend is the easy part of the ledger. The harder part is what arrives the day the ribbon is cut. A 100-megawatt campus does not run on optimism. It runs on water for cooling, on substations and high-voltage transmission, on dedicated gas turbines for redundancy, and on the local road network that has to absorb a construction workforce and a permanent operations crew. None of those line items show up in the data-centre construction series — they show up in the budgets of utilities, water districts, county governments, and state transportation agencies that did not vote on the build-out and were not asked.

Unusual Whales' framing makes the comparison vivid because the comparator set is deliberately unglamorous. Airports, marine terminals, mass transit — the connective tissue that makes an economy physically function. Data centres are now eating a larger share of national construction capital than all of those combined. The ratio is the story.

The cheerleaders and the counters

The dominant read inside financial media is that this is a private-sector miracle and that the state should keep its hands off. A second, quieter read — surfaced in utility filings and rural county commission minutes — is that rate-base growth is being quietly socialised. When a hyperscaler negotiates a discounted industrial tariff with a regional utility, the spread is recovered from the residential and small-commercial class. The data centre gets a low bill. The neighbour gets the rate-case.

A third view, mostly absent from cable news, is the labour and materials one. The same crews, the same transformers, the same concrete — they cannot build out AI campuses and rebuild the country's bridges in the same cycle. Capacity is finite, and the AI bid is winning it.

Why the political class is quiet

Because nobody loses an election over a spreadsheet line. Data centres do not have angry grandmothers chaining themselves to bulldozers. Their host counties are mostly small, mostly rural, mostly represented by legislators whose senior committee posts are decided by leadership, not by protesting constituents. The capital is concentrated. The political geography is favourable. The lobby is professional, in-house, and patient. The incentive structure for Congress is to hold still.

This publication finds that the silence is itself the tell. When a category of construction is outgrowing the country's entire publicly-financed logistics stack and the public discourse is dominated by chip-export controls and quarterly earnings, the public-finance debate has not started. It is being deferred.

What remains uncertain

The short open questions are real. The Unusual Whales note does not specify which segment of the comparator basket is being measured — federal only, federal-plus-state, or total public-and-private capital formation in those categories. The construction-spend figure itself is a flow, not a stock; the right comparison for political purposes is accumulated infrastructure stock, which is a much slower-moving target. And the AI demand curve is not yet tested against a sustained power-cost shock. None of this makes the headline less striking. It does mean the policy response — when it eventually arrives — will be designed against a number that the public has not yet been invited to read.

The default American answer to a private boom with a public tail is to do nothing, then to subsidise the cleanup. That model worked for the early auto industry and arguably for the shale build-out. It will not work as neatly for a sector that consumes a quarter of a percentage point of national construction capital before it has produced a single consumer product the average household can name. The bills are coming. The politics have not caught up.

— Monexus Staff Writer