An evidence guide for local communities
A concise guide to what recent research can tell us about jobs, firms, housing, public revenue, and electricity prices - and what remains unresolved.
This early search tool retrieves short, prewritten summaries from the studies included below. It does not generate new claims.
Try a plain-language question. Each answer identifies disagreements and links back to the underlying research.
The label covers different ownership models and generations of technology. Those distinctions matter when applying historical evidence to a proposed project.
A large cloud provider builds or controls the campus for its own network. These facilities can draw hundreds of megawatts, but ongoing on-site employment is modest.
An operator leases space, power, and connectivity to many customers. The tenants' engineers and business teams need not work in the host community.
GPU-heavy campuses are larger and more power-dense than much of the cloud-era sample. Most are too recent for credible estimates of long-run local effects.
Power, fiber, land, permitting, and tax terms shape where projects go.
Land clears, lights rise, and specialized construction activity begins.
Servers run continuously, while the permanent on-site workforce remains small.
Taxes, utility investment, housing, and nearby business activity may adjust.
Bahar and Wright begin with a registry of 341 hyperscale facilities and combine satellite-based construction timing with within-site rings, 84 announced projects that were not built, and county panels. Outcome samples vary with data availability, but a common-site check produces the same qualitative pattern. The clearest findings are a sharply localized physical footprint and narrow bounds on advertised salaries and firm entry.
The change fades rapidly with distance and is near zero by five kilometers. Daytime imagery independently confirms land clearing at the same time.
The 95% interval is -2.7% to +5.5%. The evidence rules out a large salary response in posted, mostly white-collar jobs.
The upper confidence bound is +1.6%. A national business-applications analysis produces the same qualitative result.
The 95% interval is -5.3% to +16.8%. This rules out increases above roughly 17% but allows modest gains and cannot isolate the facility's own hiring.
The 95% interval is -15.0% to +32.0%. The estimate does not establish whether local demand for supplier industries rises.
Synthetic-control rent estimates are positive but imprecise. School property-tax revenue has a +16.6% point estimate, but it is not statistically distinguishable from zero after adjustment across six fiscal outcomes.
Average estimates range from -0.9% to -1.5% across comparisons and are imprecise. The sample does not cover the larger post-2022 AI build-out.
The figure shows preferred estimates and 95% confidence intervals. Salary and firm-entry intervals are narrow. Hiring and supplier intervals are much wider.
A result is always an answer to a particular question. Changing the geography, treatment, period, estimator, or outcome can change the estimand.
A one-kilometer ring measures something different from a county, commuting zone, utility territory, or wholesale power market.
First opening, facility count, capacity, revenue growth, and proposed AI load are not interchangeable shocks.
Cloud-era facilities provide a longer post-period. New AI campuses are larger but generally too recent for long-run evaluation.
Within-site comparisons, shift-share instruments, utility event studies, and dispatch models rely on different assumptions.
Data-processing employment and establishments can rise because the facility itself enters, even when related industries show no consistent response.
These studies are complementary rather than directly comparable. The summaries below separate historical estimates from forward-looking power-market simulations.
Construction transforms the parcel. Satellite measures make that timing and spatial concentration visible.
Direct data-processing activity can increase without creating a technology ecosystem around the site.
Wholesale markets, utility ownership, rate regulation, investment timing, and spare capacity all matter.
The newest campuses are larger and more power-intensive. Historical evidence should inform the debate, not end it.