WHY DATA CENTER PROJECTS DIE

82 projects. 68 resolved outcomes. The first structured analysis of what kills data center developments and what saves them.

1. THE SIGNAL

That gap is modest. Earlier versions of our model inflated the separation by accidentally encoding outcome into the score. We fixed it. Six points is the honest signal.

But the risk score is not the story. The strongest discriminator in the dataset is not any opposition characteristic. It is whether the developer responded with structural concessions.

2. WHAT KILLS PROJECTS

Four factors distinguish blocked projects from those that survived. All are DC-specific. None are ported from the transmission literature.

Water Consumption

The number one opposition argument in blocked DC projects. No transmission parallel exists. In arid regions, water approaches veto-level concern. Tucson's city council voted unanimously to reject Amazon's Project Blue. Aquifer depletion was the primary vector.

Noise and Residential Proximity

Not a standalone kill factor. Noise amplifies residential proximity concerns. People cite it last on their list, but it makes the abstract real: 24/7 industrial humming behind a house. The anticipation may matter more than actual decibel levels. Nobody opposing a project has lived next to one yet.

The Scale Mismatch Ratio

Using Census population data, we measured project footprint relative to community size.

We do not know the exact threshold. The data suggests a gradient, not a bright line. But if your footprint exceeds 20 acres per 1,000 local residents, history says you need significant structural concessions to survive.

Developer Opacity

Big Tech identity is not a risk factor. Google, Amazon, Microsoft, and OpenAI each have both blocked and approved projects in the database. The risk comes from secrecy, not from the brand on the building.

• Project Tango (Palm Beach): end user kept secret under NDA. Mayor said she will not vote for it.
• Project Cumulus (St. Charles, MO): NDA secrecy cited as the primary driver of community distrust.
• Project Blue (Tucson): Amazon identity hidden until the community discovered it independently.

3. WHAT SAVES PROJECTS

Not all counter-strategies are equal.

"Economic development" was cited by blocked and survived projects alike. Everyone says it. It distinguishes nothing. The PACE of Trust roundtable (ACEG/DNV, January 2025) found that communities see unearned benefits as "trinkets, not investments."

The PACE of Trust report, developed through a multi-stakeholder roundtable including Invenergy, Earthjustice, IBEW, and the Nature Conservancy, found the same pattern: the speed of project development is commensurate with the level of trust built with affected communities. Process quality matters more than the dollar figure on the benefit package.

4. THE GEOGRAPHIC PICTURE

Documented events. Not model outputs.

Indiana Wave

Nine counties enacted moratoriums or bans between November 2025 and March 2026. The largest coordinated state-level wave in the database. Agricultural counties watched neighbors get developed and mobilized preemptively. Marshall County is pursuing a permanent ban. Clinton County unanimously denied a 600 MW, $10B+ project.

Northeast Tennessee

BrightRidge, a publicly owned utility serving 85,000 customers, imposed the first utility-level moratorium on serving new data center customers in May 2025. City and county moratoriums stacked on top. No zoning workaround exists when the power company will not connect you.

North Carolina

25 cancellations in 2025. That is four times the 2024 rate. Tarboro, Apex, Gates County, Chatham County, Canton.

Virginia Litigation Belt

King George County officials said they are "ready to go to war" against Amazon's $6B proposal. Warrenton has three parallel lawsuits. PW Digital Gateway: QTS and Compass face litigation from the American Battlefield Trust, now before the VA Court of Appeals.

Texas Preemption Dynamic

Hood County voted to reject its own moratorium because the state might not let counties impose them. Not actual preemption, but the threat of it changed local government behavior.

5. THE CROSSOVER

Data center opposition uses the same playbook as transmission opposition. The arguments transfer. The kill mechanisms do not.

What Transfers

• The same arguments: "benefits leave the community," ratepayer cost, land use and character, democratic process exclusion
• The same tactics: moratoriums, lawsuits, organized groups, developer withdrawal
• The same core pattern: opposition wins when it converts local impacts into a legitimacy problem

What Does Not Transfer

Susskind 2022, Vajjhala 2007, and Scott 2023 analyze generation and transmission opposition. Data Center Watch and trade press cover DC opposition as news. Nobody has connected the two literatures. This project is the first structured attempt to apply opposition risk analysis across both domains.

The transmission model has 38 projects with 11 years of data. The DC model has 82 projects with roughly 2 years. The DC findings are preliminary.

6. WHAT WE DON'T KNOW

1. The Intensity Coding Problem

Our opposition intensity ratings were researcher-assigned and may be influenced by outcome knowledge. We are recoding all projects using an observable-criteria rubric. Results forthcoming.

2. Counter-Strategy Cause vs. Correlation

Survived projects have more counter-strategies. But we cannot prove CBAs cause survival. It could be that well-resourced developers (who happen to survive for other reasons) are also more likely to implement CBAs. The sample is too small for causal inference.

3. The Dark Matter of Opposition

We only track publicized fights. Projects that quietly die before making the news, or projects that succeed with no opposition, are not in the database. This is selection bias toward documented conflict.

4. Process Quality Is Invisible

The most important variable (according to both our data and the PACE of Trust report) is how the developer engages with the community. We can measure structural outputs (CBA exists or does not exist) but not process quality (was the developer respectful, accessible, responsive?). This is an inherent limitation of any database approach.

5. Small N, Big Claims

82 projects, 68 resolved. Any finding from this dataset should be treated as a hypothesis, not a conclusion.

7. METHODOLOGY

Each site-specific project is scored on 11 factors. A separate counter-strategy response score tracks developer behavior. The two scores are intentionally independent: risk measures the environment, response measures the reaction.

Observable criteria are prioritized over subjective ratings. Population-based scale mismatch replaced the original absolute-size metric. Factor weights are hypothesis-based, not empirically calibrated.

The database tracks 82 site-specific projects and 51 legislative/regulatory actions (moratoriums, bills, regulations). These are separate categories. A moratorium is not a project.

8. ABOUT THIS RESEARCH

Three years on a $7B HVDC transmission project across four states. Senior Associate, Transmission Public Affairs at Invenergy. $2M+ public affairs budget. Four years as a registered lobbyist in energy and tech. I watched opposition kill projects that had every technical and economic argument in their favor.

This database exists because I wanted to understand why. When the same patterns showed up in data center development, I started tracking those too.

This is not advocacy. It is pattern recognition applied to infrastructure siting. Developers, policymakers, and communities all benefit from the same honest data.

This analysis is an independent editorial resource compiled from public sources including news reports, court documents, regulatory filings, and public records. It is not exhaustive and does not constitute legal, investment, or professional advice. Data reflects available information as of March 2026.