Data Centers and the Dynamics of an Investment Boom: Jobs, Population Movement, and Real Estate Prices

A two-period study of U.S. counties with documented data-center development
Draft for Bill Igoe · September 27, 2026

The town that changes while the building rises

The crane arrives before the permanent jobs do. Electricians, earthmoving crews, concrete contractors, equipment vendors, engineers, and workers on temporary assignments all require time, space, and services. A town may see more construction payrolls, more restaurant customers, a tighter rental market, and a rush to prepare land and power connections. The new facility can remain for decades. The work of building it cannot.

That difference in duration is the starting point for understanding a data-center investment boom. Its construction phase can create a sharp demand shock. Its operating phase may support a much smaller direct workforce, alongside a potentially durable tax base and continuing demand for power, maintenance, and specialized services. A county must decide what to build for the peak and what it will still need when the peak passes. Housing, roads, schools, and utility infrastructure do not adjust at the speed of a construction schedule.

The question is familiar from older investment booms. Some places turned a temporary rush into a lasting productive base. Others built for a population that never stayed. “Ghost town” is an evocative warning, though a data-center county is not necessarily a mining camp: the operating facility, local institutions, and wider metropolitan economy may endure. The useful lesson is to distinguish a temporary flow of work from a permanent stock of productive assets, and to ask which costs and benefits stay with residents.

The national backdrop is substantial. The Census Bureau estimated private data-center construction put in place at a $75.2 billion seasonally adjusted annual rate in July 2026, up from $47.8 billion in July 2025. This is a national construction-spending rate, not a sum invested in our counties or a count of facilities. It nonetheless explains why a county-level labor and housing investigation is timely. [1]

We examine two windows. The first compares 2022 with 2025 annual-average county construction employment and electrical-contractor pay; population comes from July estimates, rents compare December with December, and annual county house-price indexes compare 2022 with 2025. The second compares Q1 2025 with Q1 2026 QCEW employment and weekly pay, alongside August 2025–August 2026 rents. Its population figure compares July 2024 with July 2025, the most recent county estimate available. These are deliberately separate windows: the old comparison describes cumulative change over three years; the newer one asks what happened in the latest reported year. They are not a continuous experiment or identical observations at different dates.

We began with thirteen counties in ten states selected for documented data-center activity and usable historical industry data. The recent sample adds El Paso County, Texas, where Meta announced a data-center groundbreaking in October 2025. Richland Parish, Louisiana, offers an instructive current example but lacks the detailed electrical-contractor wage cells required for the main regression. These places were selected because of activity, not randomly drawn, and some host other large industrial projects. Our analysis therefore describes patterns in a chosen cohort; it does not estimate the causal effect of building a data center. [2]

Demand meets supply, but supply has several clocks

The first economic intuition is straightforward. If construction demand rises suddenly while qualified local electricians remain scarce, competition for labor can raise pay. If more workers seek local apartments before new homes are ready, asking rents can rise. A higher land price or a higher wage is a signal that a scarce resource has become more valuable. But the adjustment depends on how readily that resource can be brought in from elsewhere.

Workers can commute across county lines, travel between projects, work longer hours, or train for a trade. Developers can delay work, redesign it, or import prefabricated components. Households can share units, move from nearby counties, or postpone a move. New housing requires land, permission, materials, financing, and time. Each margin has a different speed. A wage response may appear before a population response; a rent response may reflect expectations before a building opens; a new apartment project may deliver units after construction employment has already crested.

There is a second feedback loop. The wage gain is good for the worker receiving it, and the higher rent may benefit an incumbent property owner. For a developer budgeting future projects, both can raise costs. If projected revenue and financing terms do not improve, higher construction wages reduce the net present value of the next investment. Higher interest rates increase the discounting of future cash flows and can restrain residential construction precisely when a county wants additional supply. Thus investment can make its own next increment more expensive. That is a possible economic mechanism, not a coefficient identified by the county regressions below.

Supply also differs across industries. Our electrical series is nonresidential electrical contractors, NAICS 238212. The residential series is residential building employers, NAICS 2361. It does not cover the electricians, plumbers, and other specialty subcontractors who build homes. An increase in electrical-contractor jobs beside a decline in residential-building jobs is therefore consistent with competition for some workers, but it is not a direct headcount of workers switching projects. QCEW classifies payroll at the reporting establishment's location, which can differ from the job site. Weekly pay is total quarterly wages divided by average employment and thirteen weeks; overtime, bonuses, and worker mix can move it even when hourly rates do not. [3]

The rent measure is Zillow's Observed Rent Index, or ZORI, a repeat-rent measure of market asking rents weighted toward the rental stock. It is useful for the price of entering today's rental market; it is not every tenant's existing lease. For the earlier period we also use FHFA's all-transactions annual county house-price index, retrieved through RainbowStats/FRED. Its index change tracks single-family prices using repeat mortgage transactions; it is not the dollar price of a typical house, and its annual county observations currently end in 2025. Neither index directly measures the cost of constructing a house. [4]

Window one: an expanding county economy, 2022–25

The earlier comparison does show a broad construction and pay expansion. Among thirteen selected counties, electrical-contractor employment rose a median 36.4% from the 2022 annual average to 2025; eleven increased and two declined. Nominal weekly electrical pay increased in all thirteen, with a 23.9% median cumulative gain. Residential-building employment rose a median 5.9% among twelve counties with both observations. Total construction employment increased in all eleven counties with usable paired values. The two industries grew together in many places, but their rates and directions varied considerably.

Resident population rose a median 3.2% over 2022–25 in thirteen counties. Expressing each county's three-year change as a compound annual rate gives a median near 1.05% per year. Median December asking-rent growth was 6.5% over the three years in the twelve counties with both December rent observations. That is cumulative nominal rent growth; it cannot be set directly against one-year wage or rent inflation. The median county was growing, but this alone says nothing about how many new residents moved because of a data center. Births, deaths, broader regional migration, and other job creation all enter the population estimate.

Table 1. Equal-weighted medians of county percentage changes; different measures have different available samples. The recent population entry is a lagged one-year measure, not Q1 2025–Q1 2026 growth. The historical QCEW entries are annual averages; the recent entries compare the same quarter in successive years. Source calculations and county panels accompany this essay.

The home-price series adds a useful second housing clock. Annual FHFA index gains from 2022 to 2025 ranged from 1.5% in Washington County, Oregon, to 25.9% in DeKalb County, Illinois, with a 14.3% median across the thirteen selected counties. Prince William gained 18.4% by this measure, versus 11.9% growth in December asking rents over the same years. In Licking, the corresponding changes were 22.6% and 15.0%. The measures cover different homes and transactions, so those differences are not a simple measure of what it cost to buy instead of rent.

Figure 1. Changes in the FHFA annual all-transactions house-price index. The index does not provide a matched county 2025–26 outcome for this essay. These are price-index changes, not median sale prices or project effects.

The averages hide important local distinctions. Licking County, Ohio, recorded 176.8% growth in nonresidential electrical employment over the older window, 37.2% higher electrical weekly pay, 5.8% more residential-building jobs, 2.1% more residents, and 15.0% higher December asking rents. It would be tempting to call this the “data-center effect.” That would be too confident. Licking also sits in the area of Intel's semiconductor construction, which competes for trades and materials. A descriptive county panel cannot allocate its gains between those investments, other economic growth, and the initial size of each industry. [5]

Loudoun County, Virginia, tells another version of the story. Its electrical employment rose 80.5%, while residential-building jobs fell 24.3%. Population increased 3.9%, and December asking rents rose 11.7%. A market with a growing regional workforce may still have declining employment at establishments classified as residential builders. The latter is a narrower measure than total homebuilding capacity, and a worker may commute from another county. Prince William, a neighboring market, gained 36.4% in electrical jobs and 11.1% in residential-building jobs in the older window, while population grew 3.2% and asking rents rose 11.9%. Neighboring counties can face the same regional boom and respond differently.

Ellis County, Texas, adds a useful counterexample to the claim that population growth always relieves or always worsens housing pressure. Its estimated population rose 13.4% from 2022 to 2025, by far faster than the cohort median, and asking rents rose 6.3%. Residential-building employer jobs fell 14.8% while electrical jobs more than doubled. Demand, permitting, completion of earlier construction, commuter geography, and supply responses all matter. A growing headcount does not tell us how quickly new units became available or whether existing residents were displaced.

These cases are reasons to inspect the cross section, not anecdotes from which to generalize. If a boom predictably moved workers out of homebuilding, the counties with faster electrical employment growth should tend to have systematically weaker residential-building growth. In the 2022–25 sample, that correlation is −0.25 among twelve counties; the simple fitted slope is −0.109 percentage points of residential job growth for each additional point of electrical job growth. That is a weak descriptive association. It could reflect project competition; it could also reflect unrelated local housing cycles. It does not establish that a data center removed even one worker from housing construction.

Window two: an active boom against a weaker residential market

The most recent labor comparison looks different. From Q1 2025 to Q1 2026, nonresidential electrical employment grew a median 16.3% among fourteen counties and increased in twelve. Residential-building employer jobs fell a median 7.0% across thirteen counties, declining in ten. Total construction jobs still grew a median 7.3% among twelve counties. One possible interpretation is that nonresidential projects remained busy as the housing-building sector weakened. That is a sharper statement than “all construction is booming,” and it matches the different industries recorded in the data.

How unusual was the residential slowdown? Among 1,443 U.S. counties with published residential-building employment cells in both first quarters, the median change was about −0.5%, and 741 showed a decline. The selected counties' −7.0% median was substantially weaker. This national set is a broad published-cell benchmark, not a matched control group. Large and small counties, different housing markets, and disclosure-related missing cells all enter it. Interest rates, land markets, and general demand for new homes offer plausible competing explanations for the selected counties' residential weakness.

Figure 2. Same-quarter change in county QCEW jobs. The scale makes Licking's large percentage change visible; it also shows why the cohort median is more stable than its mean. DeKalb lacks a paired residential observation. No bar is a count of jobs at a particular data-center site.

Licking again stands out: electrical-contractor jobs grew 135.9% in a single year, while residential-building jobs grew 8.1%. This is difficult to square with a universal claim that electrical growth necessarily displaces residential employers. Polk County, Iowa, had 62.2% electrical growth and a 2.2% residential decline. Loudoun registered 36.6% electrical growth and a 3.7% residential decline. In Prince William, electrical jobs rose 23.3% while residential-building jobs fell 7.7% and electrical weekly pay fell 9.0%. Rising hiring and falling average weekly pay can coexist when hours or composition changes.

Douglas County, Georgia, points in the other direction: electrical jobs fell 25.7%, and residential-building jobs fell 12.9%. Valencia County, New Mexico, showed a 35.7% increase in electrical jobs but a 34.2% fall in the available residential-building series. El Paso's electrical jobs rose 23.3% and residential-building jobs fell 9.5%; Meta had recently broken ground there. These observations do not tell us which firms worked on which project. They show substantial heterogeneity, which any serious boom narrative must preserve.

The operating phase deserves equal attention. Meta's July 2026 announcement for El Paso projects more than 4,000 construction jobs at peak and 300 operational jobs once complete; it also reported more than 2,300 workers already on site at the time. These are company projections and an on-site report for one large project, not QCEW county employment and not a universal conversion ratio. They do make the timing problem tangible. A locality that plans permanent housing or public services around the peak construction workforce could overshoot if much of that workforce commutes or leaves. A locality that assumes no enduring benefit could underprovide for a lasting operation, new tax receipts, or a broader supplier base. [6]

The apparent rent contradiction

If electrical demand is still increasing, why is asking-rent inflation so much lower? The answer begins with the calendar. The initial rental inflation surge happened earlier. In thirteen counties with a complete monthly Zillow history, the median of county year-over-year ZORI increases reached 13.3% in March 2022. By August 2026 it was 1.5% on the same thirteen-county basis. The fourteen-county median for August 2025–26 also rounds to 1.5%. The peak in this figure was computed from monthly data, not inferred by dividing a three-year rent change into annual pieces.

Figure 3. Median year-over-year asking-rent change in a balanced set of thirteen counties. This is a median of county growth rates, not the change in a pooled rent level. The 2022 peak precedes much of the currently visible construction wave in some places and cannot be assigned to data centers.

Slower rent inflation does not mean rents returned to their old level. A lease quoted at $1,000 that increases 13% and then increases another 1.5% costs about $1,147, ignoring other changes. A decline in the inflation rate is a change in slope, not necessarily a reversal in the price. Owners, prospective tenants, and developers can all experience that difference differently. The FHFA history shows higher house-price indexes by 2025; it cannot tell us whether county house-price growth continued, cooled, or reversed in 2026.

The rent chart also challenges an easy causal story. Much of the 2022 peak was part of a broad housing-market episode; several cohort facilities began, expanded, or were announced at different times. A median across counties with various project dates is not an event study. The cohort's median rent inflation subsequently eased even as electrical employment continued growing. Housing supply may have adjusted; demand may have cooled; broader macroeconomic conditions may have changed. The chart establishes the timing and magnitude of the rent slowdown, but it cannot select among those explanations.

Recent county relationships are weak. Across fourteen counties, the correlation between electrical job growth and August asking-rent growth is +0.12; the corresponding one-predictor regression explains only 1.4% of county rent-growth variation. In the older window the correlation was +0.28 among twelve counties with both measures. A positive sign in each window is compatible with a demand story, yet the scatter is too wide and the sample too small to justify a forecast such as “each data-center job raises rents by a fixed amount.” County prices also respond to metropolitan migration, new apartment completions, incomes, mortgage rates, and local policy.

Cross sections: a result that changes with the clock

The most provocative regression is not the rent regression. It relates the growth of electrical-contractor average weekly pay to residential-building employment growth across counties. During 2022–25 the simple slope is +0.397: a county with one percentage point more residential-building job growth had, in this fitted line, 0.397 point more electrical weekly-pay growth. The correlation is +0.63 and R² is 0.40 among twelve counties. The positive sign argues against the simplistic story that strong homebuilding necessarily depressed electrical pay or that weaker homebuilding reliably raised it. More broadly prosperous construction markets can expand several sectors together.

But shorten the clock to Q1 2025–Q1 2026 and the relationship largely vanishes. Among thirteen counties with recent residential cells, the slope is +0.122, the correlation +0.09, and R² 0.008. Restricting the new regression to the same twelve counties used in the older fit yields correlation +0.10 and slope +0.133. Thus the change cannot simply be attributed to adding El Paso. The outcome is a different horizon and a different wage observation: 2025 versus 2022 annual weekly pay in the first model; one first quarter versus the next in the second.

Figure 4. Each point is one of the same twelve counties. Lines are ordinary least-squares descriptions of the points. The recent panel's relatively flat fit is not a statistically identified change in a structural labor-demand parameter.

Table 2. Separate bivariate OLS cross sections. Each coefficient is outcome percentage points per one percentage point of the explanatory variable. Parentheses contain heteroskedasticity-robust HC3 standard errors. The recent population change is July 2024–July 2025, earlier than part of the rent outcome; its slope is descriptive, not a contemporaneous demographic response. Sample size differs by row because some county-industry cells are unavailable.

The direct house-price check reinforces the need for caution. Historical home-price appreciation has correlation +0.06 with electrical job growth and −0.32 with population growth across thirteen counties. These two simple fits explain almost none or little of the variation and omit differences in housing supply and metro conditions. They cannot be read as evidence that adding residents lowers house prices or that data-center construction has no effect on a particular neighborhood.

For completeness, add electrical job growth and population growth to the wage–residential regression. In the older twelve-county cross section, the residential coefficient is +0.413 with HC3 standard error 0.150, while the electrical coefficient is +0.103 with standard error 0.116. The three-predictor fit has R² 0.56. In the recent thirteen-county cross section, the residential coefficient is +0.027 with standard error 0.644, the electrical coefficient +0.072 with standard error 0.231, and R² 0.057. This is a particularly fragile use of a small sample: the recent model has only thirteen observations for three slopes and an intercept, and its population control predates the labor outcome.

One should also resist treating a good older fit as a proven economic law. We explored multiple specifications and windows, so a conventional post-selection p-value would flatter the evidence. In the simple historical fit, leaving out one county at a time moves the residential slope between +0.343 and +0.550; it remains positive in that exercise, but the exercise cannot repair selection or confounding. In the recent thirteen-county fit, the leave-one-out slope ranges from −0.095 to +0.682, indicating substantial instability. Even the wage series is an average of different workers and hours, rather than an observed price for a fixed unit of skilled labor.

The study's preferred empirical conclusion is therefore modest: the earlier positive wage–residential-building association does not persist in the recent window, while the recent sample shows continued electrical hiring and much weaker residential-building employment. The data describe two changing county cross sections. They do not measure project-level exposure, separate local residents from commuters, or establish a county-wide wage premium caused by a particular facility.

Population movement is the missing hinge

Population is what turns an investment boom into a town-changing event. A project can hire thousands of workers without attracting thousands of permanent new residents. Workers may live in adjacent counties; construction may be staffed by traveling crews; young adults may move in temporarily; existing residents may change jobs. A resident population count will miss a temporary daytime influx and may lag a rapid construction expansion. Conversely, rising population could reflect broader regional economic opportunities rather than the construction site that drew our attention.

Our data offer a limited first check. The historical median county population gain, converted to an annual rate, was about 1.05%. The latest available single-year median, July 2024–July 2025 for fourteen counties, was about 1.07%. Those figures are strikingly similar after matching the units of time, despite the very different labor windows. The more recent population interval overlaps the beginning of the Q1 2025–Q1 2026 employment comparison and ends well before August 2026 rents. It cannot answer whether the 2026 building wave caused additional migration.

The population–rent cross sections are also essentially uninformative about pressure from migration. In the historical sample, rent-growth correlation with population growth is −0.17 across twelve counties; in the recent sample it is +0.05 across fourteen, with the timing caveat above. Neither result refutes supply and demand. A county can add housing alongside residents, a metro can absorb workers across several jurisdictions, and population change is not an exogenous shock. Housing demand is a flow of households seeking a particular kind of unit at a particular time, not merely a percentage change in county residents.

The next empirical step should use Census components of population change, which separate natural increase from domestic and international migration, together with building permits, housing completions, vacancies, and measured home-sale prices. Compare those series with dated project starts and construction intensity. Census Vintage 2025 provides county totals and components through July 2025; a 2026 county population outcome is not yet part of our panel. For stronger identification, one would match project counties to similar places without a contemporaneous large project, test whether their trends differed before the start, and check the commuting zone rather than relying only on the site's county border. [7]

The investment decision after the first wave

Consider the developer's calculation. A prospective project produces a stream of expected future operating benefits. Its initial construction requires land, grid connections, specialized labor, materials, and finance. With an investment outlay today and revenues later, the project's net present value is the present value of expected net future cash flows minus the outlay. If a boom bids up electrician wages or the cost of serviced land, the outlay rises. If interest rates rise, the discount applied to distant cash flows also rises. Either effect can make the marginal project less attractive without making the original project a mistake.

Housing developers confront a related calculation. Higher asking rents can make new apartments profitable, but higher rates, construction wages, land prices, and permitting delays can raise the cost and postpone delivery. If a project employs many construction workers only briefly, the housing developer risks finishing units after demand has passed. If the county grows a durable cluster of employers, underbuilding could instead keep rents elevated. The same initial signal—higher rents—can therefore call for different decisions depending on expected duration and the flexibility of supply.

There are fiscal questions too. A large capital asset can contribute to local revenue, yet the locality may have to expand roads, water, power interfaces, and public services. Some investments may be privately funded; others involve ratepayers or public infrastructure. Benefits to a county treasury are not identical to benefits to renters, workers, or electricity customers. Nor does a large value of announced investment guarantee a comparable number of permanent jobs. Meta's El Paso construction and operating projections make that distinction concrete, but each site's contracts and financing arrangement need their own accounting.

Imagine three endings. In the first, the data center anchors a durable regional cluster. Training, suppliers, tax revenue, and complementary investment outlast construction, and housing supply adjusts. In the second, the facility operates successfully but has a smaller permanent local labor footprint; the county retains a valuable asset while temporary workers depart. In the third, projects are delayed or cancelled after the region has committed to infrastructure or speculative development, leaving excess capacity and financial losses. A “ghost town” is only an extreme version of that third scenario. Our county numbers do not assign probabilities to these endings.

What the evidence warrants

The case for studying this boom is strong; the case for declaring a single local outcome is not. National data-center construction spending has grown sharply. In a chosen group of counties, electrical-contractor employment continued expanding into Q1 2026, while residential-building employment recently contracted in most. Asking-rent inflation peaked much earlier in the available history and has slowed markedly. The longer-window positive cross-county association between residential-building employment and electrical weekly pay did not survive the shorter window, including when we kept exactly the same twelve counties.

Each result addresses a different part of the mechanism. The job counts reveal demand for particular types of establishments, not project staffing. The average weekly pay is a payroll measure, not a fixed hourly wage. Asking-rent inflation indicates current rental-market price pressure, not incumbent leases; the FHFA index adds house-price evidence through 2025. County population counts cannot measure traveling crews, commuters, or the 2026 migration response. A regression fitted to twelve or fourteen selected counties is a way to organize the observations; it cannot create a randomized comparison.

The strongest conclusion is a question for investors and local governments to keep asking through the full project life: how much of today's demand is temporary, who bears the adjustment cost, and what durable capacity will be left? That question joins workers' gains to the developer's NPV, housing demand to construction timing, and a county's tax base to the experience of its residents. A boom can change a town for the better. Whether this one does depends less on a single headline construction number than on the path from building to operation—and on how well housing and infrastructure are timed along the way.

Data, method, and replication notes

County roster. Historical documented cohort: Maricopa AZ; Douglas GA; Ada ID; DeKalb IL; Polk IA; Valencia NM; Licking OH; Washington OR; Bell and Ellis TX; Henrico, Loudoun, and Prince William VA. The recent labor panel adds El Paso TX. Historical comparison rows labelled “regional context” in the source panel are excluded from the main medians and regressions. Richland LA is discussed separately because the industry wage detail is unavailable. A county's inclusion establishes documented activity, not a measured dollar amount of data-center capital spending in that county.

Construction jobs and pay. BLS QCEW NAICS 238212, nonresidential electrical contractors; NAICS 2361, residential building employers; and NAICS 23, all construction. Older endpoints are annual average 2022 and 2025. Recent endpoints are quarter 1 of 2025 and 2026; quarter employment is the average of January, February, and March. Growth is 100 × (later / earlier − 1). Only published paired county-industry observations enter a measure. Cells withheld for confidentiality are missing, not zero. Nominal pay is not adjusted for consumer inflation. The national 1,443-county comparison uses the subset with both published residential-building cells, not a census of every U.S. county.

Population, rent, and home prices. Census Vintage 2025 county resident population estimates. Historical endpoints are July 2022 and July 2025; recent context is July 2024–July 2025, rather than unavailable 2026 county population. Zillow county ZORI all homes plus multifamily series; historical endpoints are December 2022 and December 2025, recent endpoints August 2025 and August 2026. The monthly historical rent chart uses thirteen counties observed in all required months, 2020–August 2026; Valencia is omitted from that balanced history. FHFA all-transactions county annual index series ATNHPIUS{FIPS}A, 2022 and 2025, retrieved through RainbowStats/FRED for all thirteen historical counties. The median is the median of county growth rates, not growth in an aggregate index.

Regression. Ordinary least squares on county percentage changes, with an intercept. Tables report heteroskedasticity-robust HC3 standard errors and in-sample R². No county weights are used, and simple models use available complete cases independently. The controlled wage model adds electrical-employment change and county population change; in the recent model the latter is lagged. The matched twelve-county wage comparison omits DeKalb because its historical residential-building observation is missing and omits El Paso because it is not part of the historical documented cohort. Small samples, selection, simultaneity, measurement error, and multiple explored specifications prevent a causal interpretation. essay_analysis.py regenerates the model summary, tables, and three publication SVGs from the accompanying CSVs.

Linked RainbowStats run. The user-provided RainbowStats analysis motivated this two-window essay. Its live result could not be inspected through the available public retrieval interface while drafting. The numerical tables and graphics here were recomputed independently from the saved county panels and source subsets; they should be compared with the linked run rather than assumed to be a transcription of it. The bundled .rstats files show the RainbowStats expressions and alignment used in the earlier analysis; the latest JSON must be hosted at the URL specified on the script's first line before the latest script can be run end-to-end in RainbowStats.

Primary sources. [1] Census, private construction put in place, July 2026. [2] Meta, El Paso groundbreaking; Meta, Richland project. [3] BLS QCEW overview, QCEW weekly-pay definition, and QCEW confidentiality. [4] Zillow ZORI methodology and county data; FHFA HPI methodology and county annual-index definition. [5] Ohio House, Licking County Intel road infrastructure. [6] Meta, El Paso investment and jobs announcement. [7] Census Vintage 2025 county population and components.