AIER Summer Research · Plan of Record

The Local Economic Effects of Data Centers

When a data center is built in a U.S. county, what happens to local investment, the tax base, and employment — and do the public incentives used to attract them justify their cost?

Author: Luke HillHillsdale College · AIER internR + StataStaggered diff-in-differences

The paper

Working Paper · Draft
Welcome the Plants, Question the Abatements: The Local Economic Effects of Data Centers and the Incentives Used to Attract Them
A Two-Ledger Evaluation of Data-Center Investment and Subsidy in U.S. Counties

When a data center opens in a U.S. county it brings an unusual bundle: hundreds of millions of dollars of long-lived capital, large electricity and water draws, and a thin permanent payroll. This paper asks what such a facility does to local investment, the tax base, and employment, and whether the public incentives used to attract it justify their cost. The organizing move is to separate two objects that public debate routinely conflates. Ledger A is the investment itself: its capital, output, low-service-burden tax base, modest workforce, and any agglomeration spillovers. Ledger B is the firm-specific subsidy used to attract it. We bring the modern place-based-policy and heterogeneity-robust difference-in-differences toolkit to bear on a county panel of staggered first openings, anchor the employment evidence to Bahar and Wright (2026), and evaluate subsidies through a cost-per-induced-job and marginal-value-of-public-funds accounting that applies the but-for rates of Bartik (2018a). The original county-panel estimates are forthcoming and are flagged as such throughout; the conclusions drawn here rest on the external causal literature. The evidence points in opposite directions on the two ledgers. The investment is plausibly net-positive locally, concentrated in hyperscale clusters; the firm-specific subsidy is usually value-destroying because most of its dollars are inframarginal. The policy that follows is to welcome the plants, tax their capital neutrally, and retire the discretionary abatements.

JEL: H25, H71, R11, R58, Q41  ·  Keywords: data centers, place-based policy, business tax incentives, but-for rate, marginal value of public funds, staggered difference-in-differences

At a glance

Question

Local investment, tax-base, and employment effects of data centers — and whether the incentives pay off.

Thesis

“Welcome the plants, question the abatements.” Separate the investment from the subsidy.

Method

Staggered DiD (Callaway–Sant'Anna) on a county panel of first openings + an incentive cost-benefit.

Data

56 pro-grade public datasets (8 must-have), merged on county FIPS.

Thesis & angle

Free-market-favorable, but rigorous — the case is stronger for engaging the counterarguments.

The organizing move is to separate two objects public debate routinely conflates. Ledger A is the data-center investment itself — long-lived capital, output, a low-service-burden tax base, a modest permanent workforce, and any agglomeration spillovers — which is plausibly net-positive locally. Ledger B is the firm-specific subsidy used to attract it, which is usually value-destroying because most subsidy dollars are inframarginal: paid for investment that would have arrived anyway. The policy that follows is to welcome the plants, tax their capital neutrally, and retire the discretionary abatements.

Methodology

Three escalating tiers; the headline estimator is heterogeneity-robust.

  1. Descriptive + case studies. Profile data-center counties (Loudoun, VA via synthetic control) using locations, electricity, employment, and subsidies.
  2. Difference-in-differences. Primary estimator: Callaway–Sant'Anna group-time ATT with not-yet-treated controls, aggregated to an event study. Robustness: Sun–Abraham, Borusyak–Jaravel–Spiess imputation, de Chaisemartin–D'Haultfœuille, a TWFE benchmark, the Goodman-Bacon decomposition, and Rambachan–Roth honest-DiD sensitivity.
  3. Incentive cost-benefit. Cost per promised vs. induced job (Bartik but-for rates), an MVPF account, and the break-even but-for rate.
Reproducible & bilingual. Data ingestion + the event study run in R (did, fixest::sunab, didimputation, HonestDiD, tidysynth); the headline econometrics also map to Stata (csdid, eventstudyinteract, did_imputation, honestdid) per AIER's toolchain.

Evidence base

Adversarially-verified facts anchoring the paper (every claim sourced).

~12% (median 12.7% across 34 estimates; range 2-25%)
Central but-for rate (share of location decisions actually tipped by incentives)
Bartik, 'But For Percentages', Upjohn WP 18-289 (2018)
~$200,000 per induced job; customized services cost ~1/3 as much
Typical cost per job ACTUALLY INDUCED by business tax incentives
Bartik Benefit-Cost Model, Upjohn Reports 287/288 (2018)
~$1.4 million per job ($5.4B invested, 339 jobs)
Illinois subsidy cost per data-center job
Good Jobs First, 'Cloudy with a Loss of Spending Control' (2025)
~$1.6 billion (state, FY2025); ~$1.9 billion with local
Virginia annual forgone sales/use tax revenue from data-center exemptions
Good Jobs First (2025)
+4-5% total private employment; +3-4% wages; no significant home-price effect
Causal effect of a county's first large data center on local employment and wages (synthetic control, ~770 facilities)
Bahar & Wright, Brookings (2026)
overstated by a factor of ~3
Degree to which naive (non-synthetic-control) estimates overstate data-center employment effects
Bahar & Wright, Brookings (2026)
176 TWh / 4.4% in 2023 (up from 58 TWh in 2014); projected 325-580 TWh / 6.7-12% by 2028
U.S. data-center electricity use and share, 2023 vs. projected 2028
Lawrence Berkeley National Laboratory, 2024 U.S. Data Center Energy Usage Report

Data pipeline

Raw public data → tidy clean files → county-FIPS panel → estimates. Reproducible end to end.

SourcesPNNL · BLS · BEA · Census · EIA · GJF
Ingest (R)01–11 scripts
Cleandata/clean/*.csv
Merge on FIPScounty_panel.csv
Event study + cost-benefitdid / fixest
Figures & tablesoutput/

Datasets

56 tried-and-true datasets professional economists use — 8 must-have. Full log in the repo's data/raw/SOURCES.md.

Data-center locations
5 datasets
DOE/PNNL IM3 Open Source Data Center Atlas · OpenStreetMap Building Edit Histories
Investment & GDP
3 datasets
BEA Regional Economic Accounts
Employment & wages
4 datasets
BLS Quarterly Census of Employment and Wages · BEA Regional Economic Accounts
Fiscal & tax base
6 datasets
Tax Break Tracker · IRS Statistics of Income
Subsidies & incentives
6 datasets
Good Jobs First Subsidy Tracker
Energy & grid
11 datasets
EIA Form-860 / 860M · EIA Form-861 / 861M
Firm microdata
7 datasets
Census County Business Patterns · EFSY Imputed County Business Patterns
Housing
2 datasets
Zillow Research Housing Data · FHFA House Price Index
Geo & controls
7 datasets
Census Substantial Changes to Counties + Geographic Boundary Change Notes · Census TIGER/Line & Cartographic Boundary Files
Water & environment
2 datasets
USGS Water Use in the United States · NOAA Monthly U.S. Climate Divisional Database
Demographics
2 datasets
American Community Survey 5-Year Estimates · IRS Statistics of Income
Broadband
1 datasets
FCC Broadband Data Collection / National Broadband Map

Paper outline

  1. 1. Introduction
  2. 2. Institutional and Industry Background
  3. 3. Conceptual Framework
  4. 4. Literature Review
  5. 5. Data
  6. 6. Empirical Strategy
  7. 7. Results: Investment, Tax Base, and Employment
  8. 8. Case Study: Loudoun County, Virginia
  9. 9. Incentive Cost-Benefit Analysis
  10. 10. Counterarguments and Externalities
  11. 11. Policy Implications
  12. 12. Conclusion

Status

Topic & angle ✓ Verified research brief ✓ Dataset catalog (56) ✓ Draft paper + PDF ✓ R pipeline (drafting) Original estimates (pending data + Stata)