The neighbourhood's share

This ledger exists because the book made claims that data would eventually test, and in July the data arrived. Daniel Yue of Georgia Tech's Scheller College of Business and co-author Yiyang Zeng measured what happens to an American county when a data centre switches on, using facility-level records merged with county economic outcomes and a difference-in-differences design built for staggered openings. With more than 2,500 data centres active or under construction in the United States and hyperscale facilities costing over $1 billion each, by the university's own framing of the stakes, somebody finally asked the question this blog keeps asking, with a method instead of a press release: who benefits?

The answer deserves to be reported straight. A data centre's activation raises local employment by 3.5 per cent, total wages by 5 per cent, business establishments by nearly 5 per cent and median household income by about 2 per cent. Those are real spillovers, credibly identified, and they cut against any lazy reading in which the build-out benefits nobody outside the fence. The ledger claim's own refuter test is capex that demonstrably raises labour income where it lands, and in the average treated county, it demonstrably does.

The two conditions

Then come the study's two conditions, and they are where the claim earns its keep. First, location decides everything. The gains concentrate in metropolitan counties; in non-metropolitan counties the researchers find no measurable growth in employment or business formation at all, only a small dip in unemployment. "Location, not facility size, determines whether the local benefits show up," Yue says. Read that carefully: the benefit does not scale with the machine. It scales with the humans already nearby. What a data centre pays its neighbourhood is an agglomeration effect, the ordinary economics of construction crews and service demand, not a share of what the machine inside earns. Build the same billion-dollar facility in a rural county, the kind with cheap land and willing zoning boards where so many of them go, and the measurable local benefit rounds to zero.

Second, the one effect that arrives everywhere is the bill. Electricity prices in affected areas rise by roughly 5 per cent, and a single large facility draws power equivalent to about 80,000 homes. The spillover is conditional; the externality is universal. A retired household in a non-metro county near a hyperscale site gets, on the study's numbers, no measurable income gain and a 5 per cent dearer kilowatt-hour.

Verdict

Complicates. The measured spillovers are genuine, and the entry credits them without squinting: metro counties near data centres are measurably better off, and the claim cannot pretend otherwise. But the study's own architecture confirms the distinction the ledger keeps returning to. The neighbourhood's share is a side effect, contingent on geography and paid out in wages and shop traffic; the machine's earnings are a property right, contingent on nothing and paid out wherever the shareholders live. A 2 per cent lift in median household income, set against the output of a billion-dollar facility, is not a partnership. It is proximity. The researchers measured the neighbourhood's side of the question: gains sometimes, a dearer bill always. The ledger adds the side their data does not need to measure, because it is written into the ownership structure before the first shovel: the owner, always.

Who owns the machine?