Erin Brockovich, whose fight with Pacific Gas & Electric was dramatized in the 2000 film starring Julia Roberts, recently launched a data center complaint tracker at brockovichdatacenter.com. Within a week, more than 1,600 residents living near new and proposed data centers had filed complaints: noise that never lets up, water tables they are watching drop, and bills climbing as fast as the substations going up next door. The pattern is already visible at scale. In South Memphis, xAI ran its Colossus supercomputer on dozens of gas turbines without federal permits, claiming they were temporary, in a city already among the nation’s asthma capitals. This month Anthropic agreed to take over that facility’s entire output to serve its paid Claude subscribers. Corporations promise jobs and tax revenue, local agencies wave the projects through with minimal review, and the people who live there are left holding the bag. A recent Gallup poll found that seven in ten Americans oppose a data center in their community; many said they would rather live next to a nuclear plant.
Against that backdrop, a reassuring counter-message has gained traction. In May, the consulting firm E3 released a 40-page whitepaper concluding there is “no quantitative evidence” of historical cost-shifting from data centers to residential ratepayers. The report is funded by the Data Center Coalition, and its strongest supporting case study is an Amazon-funded E3 analysis of Amazon’s own facilities. The historical claim is defensible. The problem is what gets built on top of it.
The case for data centers as a local good is not weak. In Loudoun County, Virginia, some 200 data centers pay roughly half the county’s tax revenue, and residents’ property tax rate has fallen about 30 percent over a decade. A Brookings analysis of 93 counties found that those ending up with four or more facilities saw private employment rise 23 percent over five to six years. But that record was built on a class of facility that no longer resembles what is being proposed. A median large campus today spans roughly 64 football fields of building and is provisioned past 1 GW. “Megacampus” is a misnomer; these are gigacampuses, and they are being sited by the hundreds.
The whitepaper’s evidence base has two weaknesses. First, it reviews a handful of studies, most not peer-reviewed and selected with no stated criteria, then treats the no-cost-shift conclusion as foregone. The one study with a credible causal design, a difference-in-differences analysis by Georgia Tech economists finding residential rates rise five percent where a data center arrives, is set aside as a “measurement problem,” even though the conclusion runs directly counter to it. Meanwhile an earlier Amazon-funded E3 study is cited as evidence of consumer surplus, despite resting on a period when equipment costs and incremental revenue were far lower than today’s.
Second, on tariffs, E3 reports a selected 38 of at least 77 SEPA/NCCETC filings and offers no analysis showing they actually protect customers. Scored against a customer protection rubric, the full universe earns a median of one point out of five, and half score zero or one. A few tariffs are strong. Most are not.
A more fundamental issue is the whitepaper’s reliance on state-level averages. Cost recovery happens within a utility service territory and lands on a specific customer population, not across a whole state, so a state average is the wrong tool for the hard causal question of whether data centers are raising costs. At the county level the picture changes. Data centers collapse into a handful of service territories, with the top 30 counties holding half of all announced megawatts. The stability E3 documents is partly a selection effect: data centers arrive in counties that already had below-average rates and the infrastructure that drew them there. Virginia and North Dakota, where demand surged and rates still fell, are the most pre-conditioned cases in the country, not a template others can copy.
My county level analysis shows host counties also already carry heavier pollution, higher emissions, more freshwater withdrawal, and a larger share of Black and lower-income residents than non-host counties. Where the energy mix is dirty, the marginal data center makes it worse, as the gas turbines at xAI’s Colossus campus in Memphis have shown. Where the mix is clean, the math runs the other way. The same facility lands very differently depending on where it is built, which a state-level analysis cannot see by design.
This is where Brockovich’s transparency demand cuts to the core. “If you’re using public resources,” she says, “the public has a right to know how much.” Yet the binding facility-level data, metered electricity, water use, generator runtime, and demand-response performance, sits behind non-disclosure agreements at the utility-tariff level. Even E3 cannot fully reproduce its own analysis outside that perimeter.
So what should regulators do? Start with three things. Mandate facility-level disclosure of metered energy, water, emissions, and demand-response. Commission the next major cost study independently, through neutral parties using a causal design. And create a distinct data center customer class, with cost-allocation and curtailment terms calibrated to how these loads actually behave, paired with a present-value benefit-cost test on each proposed facility so siting reflects where a project pencils out for the host community rather than who offers the deepest tax break.
“Not raising bills” is an extremely low bar. We should expect more from companies of this scale and footprint. And if the economics only work when the public absorbs the water, the emissions, and the grid, the operators are always free to test whether siting these things in orbit makes more sense.
Jalal Awan, Ph.D., is a volunteer member of the Alliance for Policy Research (APR). Opinions are his own.
His full re-examination of the E3 study, including code, datasets, and a statistical-methods appendix, is at github.com/jalalawan-sudo/e3-causal-bar-county-reexamination.