Dossier mode
Capital Without Labor
The same investigation, restaged one beat at a time. Drive it with the arrow keys, space, or autoplay. Nothing is cut from the piece — long runs are split across frames. Read the full investigation or open the Return on Investment hub.
Capital without labor.
Towns hand data centers decades of tax breaks for the promise of a local boom. A new study says the boom doesn't come and the public bill does — but the evidence is split, and that's the real story.
Local governments trade away decades of revenue for a data-center boom the best evidence describes, at most, as contested — and at least one serious 2026 study says turns negative once you count borrowing costs, schools, and housing.
Not 'data centers are bad.' The new 'Capital without Labor' paper's negative findings are FACT (as circulated); the split literature (NBER, SSRN find positive effects) is FACT; the tiny-permanent-workforce core is PROBABLY TRUE. Honest verdict: unproven bet, not proven disaster.
the structural fact nobody disputes: a data center is one of the least labor-intensive large investments there is — which is exactly why its county-wide payoff is contested, and why the tax deals to land one are a gamble.
Across the literature
A new 2026 study reports data-center counties get the costs without the boom.
'Capital without Labor: Data Centers and the Local Economy' (Chia, Cornaggia, Haushalter, Wang; Sept 1, 2026) reviewed counties with data centers and reported no gain in overall financial health, no rise in local employment or new business formation, higher local-government borrowing costs, slower housing-price growth, and schools drawing more from property tax. Conclusion: 'investment without labor strains public infrastructure without generating widespread agglomeration gains.' We grade that the paper makes these findings — cited as circulated; primary not independently opened at build time.
The academic evidence is genuinely split — other 2026 studies find positive local effects.
The part a one-paper thread leaves out. NBER Working Paper 35194 (May 2026, 'Data Centers and Local Economies in the Age of AI') used instrumental variables and found positive effects on total employment, construction, establishments, house prices, income, and wages. A separate SSRN study found clustering amplifies local benefits, and a county study reported private employment up 4–5% over five to six years. Reasonable economists reading county data reach opposite conclusions.
The one thing nobody disputes: data centers employ almost no one permanently.
Whichever study you trust on ripple effects, a billion-dollar data center can run on a few dozen permanent technicians. That's why the payoff is contested — there isn't much ongoing payroll to spread around — and why the tax deals used to land them are a bet on ripple effects that may not exist. Our own reporting: Georgia's state evaluator put the forgone revenue at $474.2M and estimated ~70% of the activity would have happened anyway.
The sales pitch has outrun the proof — and the costs are more certain than the benefits.
Electricity and water demand, grid strain, and (per the new paper) higher borrowing costs and squeezed school budgets are concrete. The promised agglomeration boom is the uncertain part. Trading guaranteed decades of revenue for a contested payoff is a bad use of public money before you even decide which paper is right. Graded PROBABLY TRUE: the reasonable reading of a split literature plus the undisputed jobs math.
The subsidy is guaranteed. The payoff is a maybe.
The AI build-out is pouring hundreds of billions into data centers, each shopping for the best tax deal — giving a few of the largest companies on earth enormous leverage over small county governments. The subsidy locks in up front; the payoff is left to faith. It's the core of The Data-Center Giveaway (the money) and Data Centers, Cows, and Water (the power and water), and it sits in The Austerity Myth: towns that can't fund schools somehow find room to forgo decades of revenue for a server farm.