What AI Data Centers Mean for Local Power Bills, Air and Water
Grid data, a federal court ruling and a Minnesota council vote show what is known, what is modeled and what residents can still influence.
Data center electricity use has more than tripled since 2014, and in the PJM grid region record capacity prices coincide with that growth. Air and water effects are real but often modeled, and the Memphis and Hermantown cases show where the public can and cannot shift outcomes.
Large data centers, increasingly built to run artificial intelligence, can affect nearby residents in three measurable ways: the electricity bills that help pay for new supply, the air near on-site generators and distant power plants, and the water used for cooling and power generation. The strength of the evidence differs sharply across the three. Grid auction prices are official records, while many of the air and household-bill figures are modeled estimates or scenarios.
Two events in the final days of October 2026 put these questions in front of the public. A federal judge allowed a Clean Air Act lawsuit over gas turbines at xAI's Memphis-area data center operations to go forward, and a city council in Hermantown, Minnesota, approved a large Google data center after roughly four hours of mostly hostile public comment. Neither event settles the broader debate, but both show how permitting, litigation and public input work in practice.
Electricity demand and who pays for it
A fast-growing load
A 2024 report from Lawrence Berkeley National Laboratory, prepared for the U.S. Department of Energy, estimated that U.S. data centers used 58 terawatt-hours (TWh) of electricity in 2014 and 176 TWh in 2023. For 2028 the report projects a wide range of 325 TWh to 580 TWh. A summary of the report puts the 2023 figure at about 4.4% of total U.S. electricity use.
The estimates are built from the bottom up, by counting installed equipment and applying assumptions about how it is used. The report acknowledges that a lack of direct energy data limits the analysis. The history through 2023 is therefore a reasonable official estimate, while the 2028 range is a scenario spread, not a forecast with a single most-likely value.
The PJM record: prices, not yet bills
The best-documented case for cost effects is PJM, the grid operator covering 13 states and Washington, D.C. PJM runs an auction in which power suppliers are paid to guarantee capacity for a future year, and those costs are passed on to customers. According to the analysis group IEEFA, which advocates for an energy transition, clearing prices rose from $28.92 per megawatt-day for the 2024/25 delivery year to $329.17 for 2026/27. The auction results themselves are PJM's.
| Label | Clearing price, $ per MW-day (2025/26 applies to most zones; 2026/27 is the capped price) |
|---|---|
| 2024/25 | 28.92 |
| 2025/26 | 269.92 |
| 2026/27 | 329.17 |
| 2027/28 | 333.44 |
The latest auction, for 2027/28, set a record total capacity cost of $16.4 billion, as reported by Canary Media. PJM said it secured 134,479 megawatts at $333.44 per megawatt-day. It also reported that its peak load forecast was about 5,250 megawatts higher than in the previous auction, with nearly 5,100 megawatts of that increase attributable to data centers.
PJM's independent market monitor, Monitoring Analytics, estimated that data centers caused 63% of the price increase in the 2025/26 auction, equal to $9.3 billion in costs recovered from customers across the region. Many pages repeat that estimate, but the repetition does not amount to independent confirmation, because they all trace to the same analysis. For 2027/28, secondary sources put the data center share at roughly $6.3 billion to $6.5 billion of the $16.4 billion, and the two figures have not been reconciled against the monitor's own report.
From capacity prices to household bills
The step from auction prices to a monthly bill is where the evidence thins. Capacity costs are one component of a retail bill, and states pass them through in different ways. Coverage of the auctions said some PJM customers were already expected to see increases of 1.5% to 5% in the 12 months starting June 2026. The Natural Resources Defense Council, an advocacy group, has published a scenario in which unconstrained data center growth raises an average family's monthly bill by about $70 by 2028. That is a projection under stated assumptions, not a measurement.
Interpretations of the cause also differ. The market monitor attributes the high 2027/28 prices to existing and forecast data center load together with flaws in market design, according to a secondary summary that should be checked against the report. IEEFA itself notes that data center demand forecasts may not materialize, while markets prepare for worst cases. If forecasts overshoot, customers could pay for supply that is never needed. One quoted official argues that the underlying problem is that not enough electricity is being produced. The evidence does not support a claim that data centers alone raised any particular household's bill.
On policy, the American Action Forum reports that the Trump administration and a bipartisan group of governors agreed to direct PJM to hold an emergency auction intended to make technology companies pay for new power plants. The current status of that auction could not be confirmed here, since the only report of a regulatory rejection and postponement came from a single weak source. No systematic study of residential bill effects outside PJM turned up, so what applies in Texas, the Southeast or the West remains unresolved.
Air quality: on-site combustion and the wider grid
The Memphis ruling
The Southern Environmental Law Center and Earthjustice filed the suit in April 2026 on behalf of the NAACP. According to Mezha's report of the complaint, the plaintiffs allege xAI "installed and operated at least 35 combustion turbines and other sources of air pollution" at the Colossus site without preconstruction or operating air permits. The suit also alleges that unpermitted turbines have run in Southaven, Mississippi, since mid-2024, and reports describe Southaven and the downwind South Memphis neighborhoods as majority Black. These are allegations, not findings.
In an opinion released late on October 30, 2026, Judge Debra Brown of the Northern District of Mississippi denied xAI's motion to dismiss, according to news reports. She wrote that the plaintiffs have statutory standing, that the turbines are not "mobile sources" under Mississippi's State Implementation Plan, and that the Clean Air Act's citizen-suit provision is constitutional. xAI had argued the turbines should be treated as mobile and so exempt from permits, and that citizen suits are unconstitutional. A Justice Department bid to dismiss also failed, with the judge writing that the EPA's ability to intervene in citizen suits is not the same as controlling their outcome.
The ruling is procedural. It allows the claims to proceed and says nothing yet about whether permits were required in fact, whether the law was violated or whether anyone's health was harmed. Reports also differ on the number of turbines: one cites 27 turbines at Southaven operated by xAI and MZX Tech, another repeats the complaint's "at least 35." The opinion itself was not available for this article, so no single count is stated as fact. xAI says its permanent installation will run 41 turbines, that it received a Clean Air Act permit in March 2026, and that temporary generators will be phased out in connection with a 1.2-gigawatt facility. That is the company's account, and the issuing agency and permit terms are not confirmed.
Modeled health costs from the wider grid
Memphis concerns combustion at the site. A different line of research looks at pollution from power plants and backup generators serving data centers broadly. Researchers at UC Riverside and Caltech estimate that air pollution tied to AI could cause as many as 1,300 premature deaths a year in the United States by 2030, with health costs of nearly $20 billion a year, according to the university's press release. The work circulated as a preprint, and its peer-review status is unconfirmed here.
These are modeled projections, not measured health outcomes. The team took average emissions from power plants in different regions and ran them through established air-dispersion and health-impact models. The researchers estimate that pollution from backup generators in Northern Virginia drifts into several other states and creates regional health costs of about $190 million to $260 million a year, falling disproportionately on some low-income communities. In some areas, they say, the health costs exceed what tech companies pay for electricity. They also note that no safe exposure level is known for fine particles, so meeting air standards does not guarantee healthy air.
Caltech's Adam Wierman, an author of the study, describes backup mobile generators as loud and about the dirtiest pollution source a community can host. He says residents feel the health effects without setting the regulations or sharing in the tax and job benefits. A UC Riverside report for the think tank Next 10, relayed by a newsletter, found that health impacts from California data center pollution tripled from 2019 to 2023 and could rise another 72% by 2028 without mitigation, while electricity use nearly doubled over the same period. Most of the quantitative claims here come from one research group, and independent replication was not found.
Water: small nationally, potentially large locally
The 2024 Berkeley Lab report estimated that U.S. data centers directly consumed 17 billion gallons of water for cooling in 2023, and indirectly consumed 211 billion gallons through the electricity that powers them, as relayed by The Conversation. The indirect figure is about 12 times the direct one. The report projects that direct use could double or quadruple by 2028. A statistics site gives that range as 38 billion to 73 billion gallons a year, which should be checked against the report.
IEEE Spectrum puts the direct figure in perspective. At a 50% consumption ratio, 17.5 billion gallons consumed implies about 35 billion gallons withdrawn, roughly 0.3% of contiguous U.S. public water supply. The magazine adds that many facilities draw from the same municipal systems that serve homes and businesses, so a small national share can still be large for one town. Both statements can be true at once.
Another widely cited number measures something different. Ceres, in modeling covering seven states, estimated that data centers' electricity use drew about 3.4 trillion gallons of freshwater in 2024, according to Latitude Media. That is a withdrawal figure, not consumption, and it should not be set beside the 211 billion gallons consumed. Ceres' analyst acknowledged the analysis is not a precise science.
Two practical points follow. First, the larger water footprint sits at power plants, so a facility drawing from a water-heavy or gas-fired grid has a different local water profile than one that is air-cooled or on a low-water grid. Second, company disclosure is voluntary and inconsistent, and reports often leave out indirect use, which makes totals hard to compare or verify. In Hermantown, residents' water worries ran into a gap of this kind: the coverage reviewed here contained no quantified water-use figure for the project.
Hermantown and the question of when public input counts
What happened on October 30
The Hermantown City Council unanimously approved Google's proposed data center at a special meeting on October 30, passing all six resolutions, according to KARE 11 and KQDS. Sixty people spoke over about four hours, with a three-minute limit each. KQDS counted 55 opposed and 5 in favor; other outlets said only that most or nearly all opposed, so the exact tally rests on one station. Hundreds gathered outside, but only 85 people were allowed into the chamber, and three were removed after the comment limit ended.
A first vote planned for October 26 had been postponed after two hours of chanting and protests. The approved development agreement covers a 1.7 million-square-foot project on more than 200 acres in a rural residential area just outside Duluth. Minnesota Power told the Minnesota Public Utilities Commission this month that the facility is expected to draw up to 500 megawatts, according to the Star Tribune as relayed by Blandin on Broadband. Other approvals included a special use permit and a decommissioning plan, and the project is estimated to bring in about $3 million a year in property taxes. That tax figure comes from news coverage and has not been independently verified.
Many speakers pointed to a lack of transparency, citing the city's non-disclosure agreements with Google. The fight is not over: Google still needs permits from other agencies, one lawsuit concerns ownership of 80 acres near the center of the site, and the group Stop the Hermantown Data Center has asked a court to block permits or construction.
Where earlier leverage may lie
The final vote followed earlier review stages. KAXE reported in June 2026 that the city had opened a second public comment period on a draft Alternative Urban Areawide Review (AUAR) and its mitigation plan, a multi-step Minnesota environmental review with several comment windows. A 30-day period closed July 16. The council vote, the most-covered moment, did not change the outcome.
From this one case, it is reasonable to infer that residents may have more influence in environmental review comment periods, utility regulatory dockets such as the one where Minnesota Power disclosed the 500-megawatt estimate, and state permitting than at a final council hearing. That is an inference from a single example, not a general finding.
Ways to follow a proposal
The Hermantown and Memphis cases suggest several places where a proposal leaves a public record:
- Local approvals, such as development agreements and special use permits, which in Hermantown were decided at council meetings.
- Environmental review documents and their comment periods, such as the AUAR process in Minnesota.
- Utility filings, such as Minnesota Power's disclosure of expected demand to the state utility commission.
- Air permits, which were the central issue in the Memphis suit, where the plaintiffs say permits were never obtained.
- Court dockets, where residents' groups have asked a judge to block permits or construction.
What to watch next
The strongest evidence points one way on demand: data center electricity use has risen quickly, and grid operators are building forecasts around further growth. It points less clearly on who bears the cost. PJM's capacity prices are a matter of record, but the share caused by data centers rests on one monitor's analysis, and the market's design is part of the story. Whether those auction costs reach a given household depends on state rules.
The air and water questions turn on data that mostly do not yet exist: metered water use by facility, independent air monitoring near gas-fired sites, retail-bill analyses outside PJM, and disclosure of how much load is AI versus other computing. The Memphis case may eventually test whether on-site turbines needed permits, and Hermantown's remaining permits and lawsuits will show whether leverage persists after a council vote. For residents near a proposal, the record so far suggests that specific questions asked early, about power source, backup generators, cooling design and cost allocation, are likelier to matter than a statement at the last hearing.
Frequently asked questions
Do AI data centers raise residential electricity bills?
In the PJM grid region, record capacity auction prices coincide with data center load growth, and the market monitor attributes a large share of the increases to data centers. Household effects are less certain: they depend on how states pass costs through, and specific dollar figures are scenarios. Evidence outside PJM was not found.
Did the court find that xAI's Memphis turbines violated the Clean Air Act?
No. The judge denied a motion to dismiss, so the suit continues. She found the plaintiffs have standing, that the turbines are not mobile sources under Mississippi's plan, and that the citizen-suit provision is constitutional. Whether permits were required and whether the law was violated remain undecided.
How much water do data centers use?
Berkeley Lab estimated 17 billion gallons consumed directly for cooling in 2023 and 211 billion gallons indirectly through electricity generation. Directly, that is a small share of national supply, but facilities often draw on municipal systems, so local effects can be large. Other estimates measure withdrawals, which is a different quantity.
Are the health-cost estimates for data center pollution reliable?
They are modeled projections, not measured outcomes. The UC Riverside and Caltech estimate of up to 1,300 premature deaths a year by 2030 came from one research group and circulated as a preprint, and independent replication was not found. They are best read as a range of possible costs under stated assumptions.
When can residents comment on a data center proposal?
In Hermantown, comment opportunities included an environmental review with multiple comment periods, utility commission filings and council meetings. The final vote did not change the outcome, though that is a single case. Official guidance on procedures was not available for this article, so city, state agency and utility commission pages are the places to check.
Sources
- DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers , U.S. Department of Energy (2024-12-20)
- 2024 United States Data Center Energy Usage Report , Lawrence Berkeley National Laboratory (2024-12-19)
- Green Risks: 2024 U.S. Data Center Energy Usage Report from LBL , Green Risks (2026-04-01)
- Projected data center growth spurs PJM capacity prices by factor of 10 , IEEFA
- PJM's capacity costs hit record as grid falls short on… , Canary Media
- Don't count on lower electricity prices in 2026 , Yahoo News
- Emergency Energy Auction to Prevent Data Center-driven Rate Increases , American Action Forum
- Data centers, electricity bills and PJM capacity , Soleil Energy Corp
- SpaceXAI loses bid to dismiss NAACP Clean Air Act lawsuit , Beinsure (2026-11-01)
- Judge allows NAACP lawsuit over xAI data center pollution to proceed , Mezha (2026-11-01)
- Judge Lets NAACP's Pollution Lawsuit Against Elon Musk's xAI Proceed , Startup Fortune (2026-11-01)
- SpaceX AI unit loses dismissal bid in Mississippi Clean Air Act case , Traders Union (2026-11-01)
- Judge Allows NAACP Clean Air Act Suit vs SpaceXAI , Briefs (2026-11-01)
- Elon Musk's SpaceX Fails To Dismiss NAACP Lawsuit Over Memphis Data Center Gas Turbines , Benzinga (2026-11-01)
- AI's deadly air pollution toll , UC Riverside News (2024-12-09)
- Ask a Caltech Expert: Adam Wierman on the Pros and Cons of Data Centers , Caltech Science Exchange
- UC RIVERSIDE: California data center health impacts tripled in 4 years , Maven's Notebook (2025-11-22)
- Data centers consume massive amounts of water, companies rarely tell the public exactly how much , The Conversation
- The Real Story on AI Water Usage at Data Centers , IEEE Spectrum
- Data centers' hidden water footprint is linked to the grid , Latitude Media
- How to make data centers less thirsty , Grist
- 55 opposed, 5 in favor: Hermantown council unanimously approves Google's Data Center , KQDS-TV (Fox 21) (2026-10-31)
- Hermantown City Council greenlights Google data center project , KARE 11 (2026-10-31)
- Hermantown data center vote: Council moves project forward , FOX 9 (2026-10-31)
- Hermantown moves ahead with proposed Google data center, despite opposition (St Louis County) , Blandin on Broadband (2026-10-31)
- Hermantown approves controversial Google Data Center during late night meeting , News Beep (2026-10-31)
- Hermantown accepting more public comment on proposed data center , KAXE (2026-06-22)
This article was drafted with the help of AI and reviewed by Jaroha editors before publication.