Center for Practical AI
10-guide series

“AI and the environment” is ten arguments wearing one coat.

Some of them are global. Most of them are not. Every one of them has two addresses: the place where the impact lands, and the room where the decision gets made. Which question you’re actually asking determines what the evidence says, and which of those rooms can answer you.

Built on federal energy reporting, peer-reviewed water, health, and lifecycle research, open hydrological data, state statute, live dockets, and court records. Each guide names where the evidence is strong, where it is contested, and where the honest answer is a documented blank.

The ten questions

Ten questions, two axes.

Arguments about AI and the environment go wrong in two ways: a claim true at one scale gets asserted at another, or a question gets carried to a venue that cannot decide it.

1How much electricity does this take?

The one most people start with, and the best measured.

National totalUtility commissionWell established

LBNL puts US data centers at 4.4% of national electricity in 2023, headed for 6.7–12% by 2028. Federal, and it says nothing about where the load lands.

Read the guide
2What does that electricity actually emit?

Not the same question as the first one, and the answer isn't in the first number.

GlobalUtility commissionUnresolved

Depends entirely on grid mix and on what the marginal plant is. The marginal-emissions question is unresolved and we say so.

Read the guide
3How much water, and whose?

Where the famous statistic lives, and where it was retired.

WatershedState agencyWell established

Real as a peak and local problem. The per-query figure everyone quotes was withdrawn by the researchers who produced it.

Read the guide
4What does it do to the air people breathe?

The least discussed and among the better substantiated.

AirshedState agencyWell established

Substantiated and scenario-dependent. For backup diesel, generation dominates the burden; the turn toward onsite primary generation is changing what sits behind the fence.

Read the guide
5Who pays for the buildout?

Being decided right now, in rooms most people don't know exist.

Utility territoryUtility commissionContested

Actively contested in live rate cases and dockets. Attribution of specific retail increases to data centers is disputed by serious people.

Read the guide
6Who gets a say before this gets built?

Where the fight actually happens, and whether you're allowed into it.

County / parcelLocal governmentWell established

Packed hearings where hearings exist; by-right approvals, state preemption, and NDAs where they don't. The constraint has shifted from apathy to access.

Read the guide
7What did it take to make the hardware?

The question this series once declined to answer. Now researched.

Supply chainNo working venueContested

Mostly copper and grid wire, not rare metals in chips. The famous e-waste projection was recalibrated roughly tenfold, and the biggest findings are the numbers nobody publishes.

Read the guide
8What happens when the rules are broken?

What a permit means when the equipment is already running.

AirshedCourtsIn litigation

Unpermitted generation is documented at multiple sites. A federal Clean Air Act case and a citizen appeal dismissed as moot are deciding what enforcement means now.

Read the guide
9What does a data center sound like next door?

The complaint that fills hearing rooms first.

NeighborhoodLocal governmentWell established

Measured at up to 90 dB at fence lines against 55 dB limits. Five of the six data center class actions filed in 2026 are about noise.

Read the guide
10What are we not told?

The question underneath all the others.

County / parcelNo working venueWell established

Facility water figures held as trade secrets, NDAs across most Virginia data center localities, and generators found by thermal drone rather than by any filing.

Read the guide

A global problem hands you guilt. There is nothing at the end of it you can actually do on a Tuesday. A watershed problem hands you something else entirely: a named municipal water system, a drought plan, a withdrawal permit, and a disclosure requirement that either exists or doesn’t.

Getting the scale right is not a way of caring less. It is the only way to do anything.

The tell

Two of these move in opposite directions.

If you need proof these are separate questions rather than one issue with several faces, it's this, and it is the least-quoted finding in the literature.

“Water use by data centres can be negatively coupled with CO2-equivalent emissions, with methods of reducing water consumption increasing carbon emissions in some cases.”

Chien, Gupta, Ren, Sriraman & Tomlinson (2026), Nature Reviews Clean Technology

WatershedGlobal

Closed-loop “zero water” cooling cuts water use and raises electricity demand, which raises emissions. The water question is answered in a basin; the carbon question is answered in the atmosphere. Anyone demanding both zero water and zero carbon is asking for something the engineering does not currently offer.

The second axis

Where a question is felt is not where it is decided.

New to this series as of August 2026, because the questions we added made the difference impossible to ignore.

A noise problem is felt in the houses along the fence line and decided in a county code hearing. A rate increase is felt across a utility territory and decided in a commission docket. And two of the ten questions carry the badge No working venue: no process currently exists where they get decided. Nobody convenes on hardware supply chains, and no venue orders the disclosure that most of these pages end up asking for.

That badge is not a shrug. A question with no working venue is a question whose venue is still available to be built, and two of the guides below are about exactly that.

The series

Ten guides, one question at a time.

Each guide names the overstated version and the dismissive version of its claim before saying what the evidence supports.

1All ten

What Are We Arguing About?

The ten separate questions hiding inside “AI and the environment” — where each impact lands, and where each is actually decided. Start here.

Read the guide →
2Questions 1 & 2Grid region

Electricity and Emissions

US data centers used 4.4% of national electricity in 2023, headed for 6.7–12% by 2028. What that does to emissions depends on something the number doesn’t contain.

Read the guide →
3Question 3Watershed

The Water Question

The most-quoted statistic in this debate has been retired by its own authors. What replaced it is more useful — and it points at your watershed, not the planet.

Read the guide →
4Questions 4 & 5Airshed

Who Pays

The health and cost burden of the AI buildout lands on specific counties and specific ratepayers — and the algorithms optimizing for aggregate efficiency make that worse, not better.

Read the guide →
5Case studyCounty / parcel

The North Carolina Case

A repealed carbon target, 37 unpublished tax determinations, a large-load queue the state’s own Commissioner calls 70% data centers, and an Attorney General who wouldn’t sign.

Read the guide →
6Question 6County / parcel

Who Gets a Say

Packed hearings, state preemption, NDAs, and $130B in blocked projects in one quarter. The constraint on participation has shifted from apathy to access.

Read the guide →
7Question 8Airshed

When the Rules Are Broken

Forty-five generators with no permits, a citizen appeal dismissed as moot, and fines smaller than the delay they avoid. What enforcement does when a facility builds first.

Read the guide →
8Question 7Supply chain

The Hardware Question

The question this series once declined to answer, now researched: the AI mineral story is mostly copper and grid wire, the famous e-waste figure was recalibrated tenfold, and the biggest findings are the numbers nobody publishes.

Read the guide →
9Question 9Neighborhood

The Sound Next Door

Five of the six data center class actions filed in 2026 are about noise. What the hum is, why an ordinance can be met while a bedroom still vibrates, and which fixes have actually worked.

Read the guide →
10Question 10County / parcel

What Are We Not Told

Water use held as a trade secret, projects code-named through approval, generators found by drone. What nobody is required to report, and the routes facts take around the gaps.

Read the guide →

A gap, closed

Question 7 — mining, materials, and e-waste — sat on this hub as a declared blank from the day the series launched, because we hadn’t done the research and wouldn’t pretend to an answer. The research is done. The biggest findings turned out to be the numbers nobody publishes, and the guide says which ones.

Read the hardware guide →
The method, on the phrase itself

Both of the usual answers are wrong in the same way.

AI is bad for the environment.

The version that goes too far

Treats ten claims of wildly different evidence quality as one settled verdict, and leans on figures their own authors have since retired or recalibrated. The strongest version of this argument is weakened, not helped, by the numbers most often used to make it.

The version that waves it away

Answers a county-level rate increase with a global denominator — “it's only about 3% of world electricity” — which is true, and beside the point for the person holding the bill, breathing the plume, or hearing the hum.

What the evidence supports

Electricity demand is real and well measured. The carbon consequence depends on grid mix and the marginal plant, which is unresolved. Water, health, and noise burdens are real, local, and concentrated in specific basins, counties, and fence-line neighborhoods. Unpermitted generation is documented and in litigation. And the cost, consent, and disclosure questions are being decided right now, in dockets and hearings that most people never learn exist, or that, in a growing set of places, no longer exist to attend.

Sources for this split: lbnl2024 · healthComputing · floodlightDataOne · dcWatchTracker — full citations below.

Where this leads

Knowing which question you're asking is a skill.

This series applies it to one contested topic. The Applied AI Certification builds the underlying capability (telling a claim's scale from its rhetoric, and knowing what evidence would change your mind) across every domain where AI shows up, not just this one.

Key research

What this series is built on.

Every empirical claim across the ten guides traces to a source registry entry. Each carries three things: where it was published, how the number was produced, and the scale at which its claims hold. A federal report and a preprint are not the same kind of thing, and a modeled estimate is not a measurement.

Federal government reportModeled estimate · not metered measurementNational total
Shehabi et al., Lawrence Berkeley National Laboratory / DOE (Dec 2024)2024 United States Data Center Energy Usage ReportUS data centers used 4.4% of national electricity in 2023 — 58 TWh (2014) → 176 TWh (2023) → a projected 325–580 TWh by 2028, or 6.7–12% of US electricity. Cite the range, never the top of it. Published December 2024, so it predates the 2025–26 capital expenditure surge. The report's own limitations section notes it says nothing about where the load lands, which is why its figures are badged national rather than grid.
International agency reportModel projection · scenario-dependentGlobal
International Energy Agency (2025)Energy and AIGlobal data centre electricity demand roughly doubling by 2030 — about 485 TWh to about 950 TWh, near 3% of world electricity. Also the source of the finding that renewables are growing about 22% a year and meeting nearly half of that demand growth, which belongs on the page for the same reason the alarming figures do. The hardware guide additionally cites this report's critical-minerals estimate: roughly 512 kt of copper demand from data centres by 2030, about 2% of global copper demand, the first institutional attribution of mineral demand to data centres with a defensible denominator.Citation still being verified against our research files.
Peer-reviewed studyModeled estimate · not metered measurementWatershed
Li, Yang, Islam & Ren (2023/2025)Making AI Less "Thirsty"Source of the retired 500 ml figure and of its replacement: one GPT-3 output of 150–300 words consumed 16.9 mL total in an average US data center — 2.2 mL onsite cooling plus 14.7 mL at the power plant. Most of the water is not in the building. Ren notes later models are likely more efficient.
Preprint · not yet peer-reviewedModeled estimate · not metered measurementWatershed
Han, Li, Wierman & Ren (2026)Small Bottle, Big PipeUS data centers could require 697–1,451 million gallons per day of new peak water capacity through 2030 — New York City's entire daily supply is about 1,000 MGD — at a build cost of roughly $10B to $58B; or 227–604 MGD if water intensity falls 10% a year. Ren: "Only comparing the annual totals can obscure the real water challenge." The constraint is peak capacity, not annual volume.
Peer-reviewed studyGlobal
Chien, Gupta, Ren, Sriraman & Tomlinson (2026)Strategies and design for increasing AI sustainabilityNature Reviews Clean Technology. Quote in full, never truncated: "Water use by data centres can be negatively coupled with CO2-equivalent emissions, with methods of reducing water consumption increasing carbon emissions in some cases." The dropped second clause is both the mechanism and the paper's own hedge. Closed-loop "zero water" designs raise electricity demand.
Peer-reviewed studyModeled estimate · not metered measurementAirshed
Han, Wu, Li, Wierman & Ren (2026)Health-Informed ComputingCommunications of the ACM. $20.9B in public health costs is the high-growth 2028 scenario; low-growth is $11.7B and the 2023 baseline is $6.67B — always state which. About 1,300 premature deaths and 600,000 asthma symptom cases in the 2028 high scenario. Scope 2 (generation) dominates Scope 1 (onsite backup diesel) by roughly 12 to 1. The worst-affected counties carry about 7× the national average per household.
Open datasetHydrological model · CMIP6 scenariosWatershed
World Resources Institute — Aqueduct 4.0 (2023)Updated Decision-Relevant Global Water Risk IndicatorsThe projected water-stress layer. Hydrological output from PCR-GLOBWB 2 translated into water risk indicators and aggregated to HydroBASINS level 6 sub-basins — which is why the map is basin-shaped rather than county-shaped, and is itself the argument that water is a watershed variable. Projections center on 2030, 2050, and 2080 under three scenarios: optimistic (SSP1 RCP 2.6), business-as-usual (SSP3 RCP 7.0), and pessimistic (SSP5 RCP 8.5). The 2080 milestone is built from the 2065–2095 window. Creative Commons; attribution required.
Federal government reportModel projection · scenario-dependentCounty / parcel
Mongird, Thurber, Vernon, Burleyson, Akdemir & Rice — Pacific Northwest National Laboratory (2025)IM3 Projected US Data Center Locations (v1.1)Model projections of new data center facilities across the contiguous US through 2035, produced with the CERF-Data Centers model by the IM3 project at PNNL, supported by the DOE Office of Science. These are modeled expectations, not announced projects — a distinction that matters, because a map of where a model puts facilities and a map of where developers have filed answer two different questions. CC BY 4.0.
Independent policy analysisCompiled from public filings and reportingCounty / parcel
Edward Kubiak — Compute Atlas (2026)An open, source-cited map of US data centersThe announced-reality layer: proposed, permitted, under-construction, and operational facilities with coordinates, each traced to public sources and human-reviewed before publication. Independently maintained rather than institutional, and it is a living record — figures move between refreshes, so anything drawn from it carries a retrieval date. Data CC BY 4.0.
Government agency memo or determinationCounty / parcel
North Carolina Department of Commerce (April 6, 2026)Memo to the Governor's Energy Policy Task ForceAbout 800 MW operational in North Carolina as of December 2025, with roughly 6,300 MW in the pipeline. Commerce issued 37 written data center tax-exemption determinations between 2015 and 2025; companies are not required to report actual investment or exemption value, and the list of 37 is not published. Commerce states openly that its MW figures come from Baxtel, a commercial tracker — a disclosure worth noticing, because it means the state is reading its own buildout off a private database.
Statute / session lawUtility territory
North Carolina Senate Bill 266 / Session Law 2025-78The Power Bill Reduction ActVetoed by Governor Stein on July 2, 2025; veto overridden July 29, 2025 (Senate 30-18, House 74-46). Eliminates from G.S. 62-110.9 the goal of reducing utility CO2 emissions 70% from 2005 levels by 2030 — the baseline year is not optional, because a percentage without a baseline is not a citable target. The 2050 carbon-neutrality requirement survives. Also broadens construction-work-in-progress recovery, so customers begin paying financing costs during construction rather than after it.
Journalism · secondary reportingCounty / parcel
Floodlight / The Guardian (August 27, 2026)One of East Coast's largest data centers accused of violating federal lawAt least 45 of 62 tractor-trailer-sized generators running at the partly built DataOne facility in Vineland, NJ, documented by thermal drone in mid-August 2026. NJDEP: no permits issued and none under review. Generators operating since October 2025; the state noticed during a July 29 inspection. Two schools within a mile, two municipal stop-work orders, a March noise citation, and a planned 300 MW fuel-cell system. The Guardian co-published; cite the Floodlight version, which is not paywalled.
Independent policy analysisCompiled from public filings and reportingNational total
Data Center Watch (2025–2026)Tracking blocked and delayed projects and organized oppositionThe opposition ledger: $18B in projects blocked and $46B delayed across the tracker's first two years, with Q1 2026 alone matching all of 2025 at roughly 75 projects and $130B. Active opposition groups more than doubled from 396 at the end of 2025 to 833 across 49 states by March 2026. Run by 10a Labs, an AI research firm — a provenance worth stating, since a tracker of opposition maintained by an industry-adjacent firm is not a neutral instrument, and its direction of drift would flatter the industry's alarm, not the opposition's.Citation still being verified against our research files.
Advocacy or industry position paperCompiled from public filings and reportingCounty / parcel
Public Citizen (July 8, 2026)The Secret Data Center BuildoutThe anchor compilation on nondisclosure agreements: roughly 80% of Virginia localities with proposed or existing data centers are under NDAs (a figure the report credits to Virginia Mercury's April 2025 survey); Shreveport-area residents learned of Amazon's $12B campuses and an accompanying 5 million-gallon-a-day water sale only at the governor's announcement; Memphis officials were NDA-bound while the turbines ran. Advocacy tier: the compilation is theirs, the underlying instances are documented reporting.Citation still being verified against our research files.
Peer-reviewed studyModel projection · scenario-dependentSupply chain
de Vries-Gao, Resources, Conservation and Recycling (2026)Recalibrating global AI e-waste estimatesThe peer-reviewed correction: modeling AI server e-waste bottom-up from TSMC's actual chip-packaging capacity, rather than extrapolating demand, lands at 131–225 kt a year by 2030 — seven to fourteen times below the widely quoted projection — with realistic server lives of 4–6 years rather than 3. The paper's stated limitation is itself citable: basic information about AI server deployments, configurations, and replacement cycles is not publicly available. The publisher page blocks automated access.Citation still being verified against our research files.
Journalism · secondary reportingNeighborhood
Wisconsin Public Radio (July 2026)The class action against Microsoft's Mount Pleasant facilityA federal class action filed July 1, 2026 over the $7.3B Fairwater facility Microsoft calls its most powerful AI data center: cooling and generator sound allegedly audible a mile and a half beyond the property line, plus construction noise and light, on behalf of a proposed class of more than a thousand households. Microsoft's own April 2026 website update acknowledged a tonal humming sound reaching residents to the north. One of six data center class actions counted in 2026, five of them about noise.Citation still being verified against our research files.
Last reviewed: August 2026We review this page quarterly. Statistics in this category change rapidly.Several threads in this section are actively moving: a North Carolina Utilities Commission decision is due in September 2026, a federal Clean Air Act injunction ruling was pending when this hub was last verified, and the participation and noise dockets change monthly. Questions badged “in litigation” carry a shorter shelf life than the quarterly cycle. Sources still being confirmed against our research files are marked individually above.

Want CPAI to teach this in your community or classroom?

We deliver this material as workshops and sessions for schools, libraries, local government, and community organizations, including the version built for county-level decisions.