The surprising variable that correlates with data center opposition

What predicts data center opposition?

The answer to that question is worth millions. Brad Barton, Applied Digital’s SVP of Construction, told us in May that each month of delay to a data center loses its developer an average of $14 million. Opposition is rising across the country and shutting down projects left and right. Developers are losing battles.

If you know where opposition will be before you encounter it, you save capex and the headache.

So we tested it several variables — and one surprised us.

What doesn’t correlate

Many of the variables you might expect to correlate with opposition, don’t.

Partisan lean? 2024 vote count had no correlation with data center opposition across ~1,000 counties with significant opposition in the US.

Water usage? Similarly, opposition was the same across counties with more or less water usage.

Income inequality? No relationship; whether measured by the Gini coefficient or top 1%’s share of the wealth, neither showed any more or less opposition.

The demographics that mattered

Still, some demographics still matter.

Counties with significant data center construction before 2024 saw less opposition to new projects afterwards. This finding makes sense; these counties have seen data centers before and are familiar with what they mean. The legal paths to operation are well-defined, and locals have known about living near data centers for years. One more data center won’t impact their lives, whereas for a locality with no data centers, the first data center will make waves.

On the other hand, a county’s opposition to data centers rises with its unemployment rate. Each additional percentage point of unemployment is associated with about 8% higher odds of significant opposition to a new data-center project.

And interestingly, more educated counties also oppose data centers at a higher rate. A higher percentage of college graduates does correlate with more opposition to local data center construction. If one county’s education rate is 10 points higher than another’s (say, a county where 20% of residents have bachelor’s degrees, versus a neighboring county with 30%), the more educated county sees 7% higher odds of significant opposition for a data center project on average.

But, while these variables all matter, they only loosely fit the data. When looking for stronger correlations, one surprising variable stands out.

Manufacturers don’t want data centers nearby

The strongest correlation we found was a surprising one: the percentage of adults who work in manufacturing.

Among counties with <1% of adults working in manufacturing, only about 12% of projects saw major opposition. That rate increases by 5.6 percentage points (a 46% increase) when more than 6% of the population works those jobs.

So why do blue-collar workers oppose data centers nearby?

It’s not immediately obvious. New data centers create demand for local construction, HVAC, and manufacturing vendors. Developers argue that their projects provide money to communities and add jobs.

But manufacturing workers aren’t buying that case. In manufacturing-heavy locations, opposition consolidates around a few causes:

1. Noise pollution: In low-manufacturing counties, only 5.7% of data center projects saw opposition because of noise pollution. In counties with significant manufacturing, that rises to almost 14% — more than double. Manufacturing hubs are especially sensitive to additional noise.

2. Environmental impact: Many Americans are worried about how data centers use water and affect air quality. But in manufacturing hubs, the environmental concerns come from elsewhere. 12% of projects in manufacturing hubs run into concerns about ecological harm, and 7% see questions about habitat loss, both considerably more than the low-manufacturing baseline. Manufacturers may be well-informed about local pollution, but they may worry about other externalities in the natural environment.

And there’s another interesting trend: manufacturers know what they’re talking about. That may be why they’re far less concerned about the development process & openness. Only 1 in 20 projects in manufacturing-heavy locations ran into questions about secrecy, compared to almost 1 in 9 in low-manufacturing authorities.

Manufacturers have distinct, serious concerns about data centers. Their opposition tends to be more vocal and more specific. Like many Americans, they are mindful of development externalities. But they have unique reasons — ones that can’t be ignored.

If you want to know where opposition is building, and where you can avoid it, get in touch.

Methodology

This analysis draws on Spark's internal project-tracking database (5,045 data-center projects across 959 U.S. counties, current as of August 2026), matched at the county level to public demographic and economic data: unemployment rate and educational attainment (BLS/Census, 2023), poverty rate (Census, 2023), population and rural-urban classification (USDA, 2023), 2024 presidential election results, and manufacturing employment share (Census County Business Patterns, 2018 — the most recent year with detailed, unimputed county-level industry data).

"Significant opposition" and project counts were modeled two ways: the number of data-center projects per county via negative binomial regression, and the share of a county's projects with significant opposition via quasi-binomial logistic regression, with standard errors adjusted for state-level clustering where a predictor (e.g., income inequality) varied only at the state rather than county level. Reported effect sizes are odds ratios from these models; statistical significance was assessed via likelihood-ratio tests against a baseline model of unemployment, education, and poverty. All bucketed charts show pooled rates (total significant-opposition projects ÷ total projects in each bucket), not simple averages across counties, so larger counties carry proportionally more weight — consistent with the modeling approach.

Opposition reasons were classified from free-text event records (developer filings, hearings, news coverage) tied to each project, using a set of 19 standardized categories built from keyword patterns grounded in the actual language of those records. Each project was tagged with up to 4 reasons based on keyword frequency. About 30% of significant-opposition projects had no reason we could confidently classify from the available text (e.g., "residents opposed" with no stated cause) and are excluded from the reason percentages shown.