Dangerous Thoughts

Part 3: Water

Energy costs arrive on a bill. Water losses arrive in your tap, your well, and your watershed — and unlike money, an emptied aquifer doesn't come back when the company leaves. This chapter covers how data centers consume water, why operators fight so hard to hide the numbers, and the twelve demands that protect your community's supply.


3.0 Why water is different

Three things distinguish the water fight from every other chapter in this handbook:

It is irreversibly local. Ratepayer costs are spread across a region; water comes out of your basin. When a hyperscale facility draws from your aquifer or your municipal system, no one in the next county shares the burden. The Newton County, Georgia water authority director put it in one sentence: the data centers are taking up the community's wealth, and the county simply doesn't have the water (New York Times, July 2025).

Peak demand is cruelly timed. Evaporative cooling consumes the most water on the hottest days — exactly when your reservoirs are lowest, your residents' demand is highest, and the electric grid (and its own water-hungry generation) is most strained (ELI, Jan 2026). A water commitment that looks comfortable on an average day can be a crisis on the fifteen days a year that matter.

It is the most heavily concealed number in the industry. Operators routinely classify facility water use as a trade secret, wrap host governments in NDAs, and report only fleet-wide averages. Estimates of average consumption vary widely precisely because developers often won't disclose actual figures (E&E News, Dec 2025). You cannot negotiate caps on a number you're not allowed to see — which is why disclosure is Demand 1, and everything else depends on it.

The encouraging news mirrors Part 2: the policy landscape is moving fast in communities' favor. Water-usage legislation is gaining momentum across statehouses in 2026; Minnesota established a dedicated water-permitting requirement for data centers in 2025, Utah passed first-in-the-nation disclosure amendments in 2026, and more states are following (MultiState, May 2026). Your community doesn't need to invent these tools — it needs to demand them, and demand they be stronger than the watered-down versions industry lobbies for.


3.1 How a data center actually uses water

Direct use: cooling

Servers convert nearly all their electricity into heat, and that heat must go somewhere. The dominant technologies, in order of water intensity:

Cooling approach How it works Water profile Energy profile
Evaporative (cooling towers) Heat is rejected by evaporating water to the atmosphere High consumption — water leaves the local cycle permanently; the industry's cheap default Low energy
Once-through / open loop Water drawn from a source, warmed, discharged back High withdrawal, lower consumption; thermal discharge issues Low energy
Closed-loop liquid / direct-to-chip Coolant circulates to the chips in a sealed loop; one initial fill, minimal makeup water Minimal ongoing draw — Microsoft's latest designs advertise zero evaporated water Moderate
Air cooling / dry coolers Outside air carries heat away Near-zero water Higher energy — fans and chillers raise PUE
Immersion cooling Servers submerged in dielectric fluid Near-zero water Efficient at high densities

Two facts matter for negotiation. First, low-water designs are proven, commercial, and increasingly standard for AI workloads — closed-loop systems are spreading because high-density GPU racks favor liquid cooling anyway, and major operators (Microsoft's zero-evaporation direct-to-chip designs; Vantage's closed-loop campuses) market them publicly (Data Centre Magazine, Feb 2026; Microsoft, 2026). A developer who tells your drought-prone county that evaporative cooling is the only option is describing their cost preference, not a technical constraint.

Second, water and energy trade off. Air cooling and water-recycling systems (like RO treatment of cooling-tower blowdown) consume more electricity, worsening PUE (DOE FEMP). This is not a reason to accept evaporative cooling — it is the reason your water demands (this chapter) and energy demands (Part 2) must be negotiated together, so the developer can't play one against the other.

Indirect use: the electricity multiplier

The larger share of a data center's water footprint never touches the facility. Thermoelectric power plants — coal, gas, nuclear — evaporate enormous volumes of cooling water generating the electricity data centers consume. In 2023, U.S. data centers consumed roughly 17 billion gallons directly for cooling — and an estimated 211 billion gallons indirectly through electricity generation (LBNL data via The Network Installers, 2026). Wind and solar require essentially none. This is why Part 2's Demand 10 (clean-energy sourcing) is secretly the largest water demand in this handbook.

Chart F: In 2023, U.S. data centers consumed about 17 billion gallons directly for cooling versus roughly 211 billion gallons indirectly through electricity generation.

Definitions that win or lose your cap

  • Withdrawal vs. consumption. Withdrawal is water taken in; consumption is water not returned (mostly evaporation). EESI defines data center consumption as withdrawals minus discharge. Operators prefer to report whichever number flatters them. Your agreement must cap and meter both.
  • Potable vs. non-potable. Most facilities still run on drinking water; alternative sources average under 5% of typical facility supply (Project Finance Law, 2025). That gap is your negotiating space (Demand 4).
  • Chemical treatment. Cooling water is dosed with anti-corrosion and biocide chemicals, making blowdown discharge an industrial wastewater stream — and making the evaporated share unavailable to people or agriculture (Indiana University OneWater).
  • Fleet average vs. facility actual. Corporate sustainability pages cite global WUE averages and "water positive by 2030" pledges. None of it binds the facility in your watershed unless your agreement says so.

3.2 The numbers

The scale, as established in Part 1 and extended here:

  • U.S. data centers directly consumed about 17 billion gallons for cooling in 2023; LBNL projects this could double or even quadruple by 2028 (The Conversation, Aug 2025).
  • An average facility uses roughly 300,000 gallons per day; a large hyperscale site can draw up to 5 million gallons daily — about 1.8 billion gallons a year, the usage of a town of 10,000–50,000 people (EESI; Indiana University OneWater).
  • Utah's Rivers Council put it in terms any council member understands: two data centers can use as much water as a city of 100,000 people (Wyoming Public Media, April 2026).
  • Concentration in real places: Google's facility in The Dalles, Oregon consumed 29% of the entire town's water supply in 2022; Meta's Newton County, Georgia campus draws about 500,000 gallons daily — 10% of the county's total water use — in a county now projecting a water deficit by 2030 (PPC Land; NYT, 2025).

Chart E: U.S. data center direct cooling water consumption of 17 billion gallons in 2023 is projected to double or quadruple by 2028.

Chart G: Documented cases — Google's facility consumed 29% of The Dalles' town water supply in 2022; Meta's facility draws 10% of Newton County's daily use.


3.3 The secrecy machine

No other impact in this handbook is hidden as aggressively as water. The documented playbook:

Trade-secret claims. In Wisconsin, Microsoft argued to state officials that documents covering its use of Lake Michigan water should be treated as trade secrets — a status normally reserved for things like the Coca-Cola formula. The claim failed only after Midwest Environmental Advocates sued the city of Racine for withholding water records (The Progressive, Oct 2025). In South Carolina, a county redacted Google's projected water and sewer usage from a Freedom of Information Act response, claiming trade-secret protection — and got sued by a resident (Post and Courier, 2024). The industry's stated rationale is that water data lets competitors infer facility design (E&E News).

NDAs binding your own officials. At least four Michigan communities signed non-disclosure agreements covering data center projects — including one where the city providing the water signed an NDA a year before Microsoft revealed itself as the developer (Detroit News, Feb 2026). In Virginia, the densest data center market on earth, NDAs are the tool of choice for concealing water and energy use (The Progressive; WVTF, Feb 2026). The Dalles itself went to court trying to keep Google's water use secret from a newspaper — while 98% of the surrounding county was in extreme drought (The Register, 2021).

A regulatory patchwork with holes. The federal Clean Water Act governs pollution discharge, not water quantity — there is no federal backstop on consumption (MultiState, April 2026). In Louisiana, home to Meta's largest-ever data center, groundwater pumping requires no permit at all (E&E News). Virginia's governor vetoed a data center water-use bill in 2025. Even Utah's pioneering 2026 Data Center Water Transparency Amendments require only an estimate of future use — actual consumption records can stay secret (Wyoming Public Media).

Why this matters strategically: every hour your community spends fighting for basic numbers is an hour not spent negotiating terms — which is the point. The counter is structural: make disclosure a precondition of every approval (Demand 1), and treat any invocation of secrecy as itself disqualifying. Note the asymmetry the Great Lakes Compact demonstrates: where strong regional water law exists, the trade-secret gambit fails; where it doesn't, the gambit becomes the norm.


3.4 What's at stake on the ground: the Newton County file

One case study, because it contains nearly every lesson in this chapter. After Meta broke ground on a $750 million data center in Newton County, Georgia in 2018:

  • Beverly and Jeff Morris, living 1,000 feet from the site on well water, watched their taps slow to a trickle and fill with sediment; appliances failed repeatedly; they spent roughly $5,000 and replaced equipment in 2019, 2021, and 2024, and still haul water — afraid to drink from the tap (NYT via San Juan Daily Star; BBC via Yahoo, 2025).
  • No baseline existed. The Joint Development Authority conducted no well-water study before construction, so causation can be argued forever — the timing "could be a coincidence," its spokesman offered (Moneywise, 2026). Meta later commissioned its own study, which — unsurprisingly — found its operations were "unlikely" to be responsible.
  • Hydrologists note that construction at this scale often involves "dewatering" — pumping out the surrounding groundwater — with effects on neighboring wells that are entirely foreseeable (NYT, via Moneywise).
  • The county's water rates are set to rise 33% within two years (annual increases were typically ~2%), and the county projects a water deficit by 2030 (Moneywise; PPC Land).
  • By mid-2026, residents in adjacent Morgan County were bringing jars of brown well water to an EPA hearing; the EPA noted that private wells are not regulated under the Safe Drinking Water Act, and Georgia doesn't regulate private well quality either — there was no agency whose job it was to protect these families (CBS Atlanta, May 2026).

The lessons, demand by demand: no baseline study means no accountability (→ Demand 6); residents bore the burden of proving harm against a $1.8-trillion company's hired experts (→ Demand 7); construction impacts arrived years before "operations" did (→ Demand 8); and the costs landed on ratepayers and well owners while the facility's own supply was contractually secured (→ Demands 5 and 10). Every one of those failures was preventable — in the agreement, before approval.


3.5 The demands: what to ask for, and why

Same format as Part 2: the ask, the justification, and the benchmark.

Demand 1 — Full, metered, public disclosure — actual use, no trade-secret shield

The ask: (a) Dedicated metering of all water sources (municipal, wells, surface, reclaimed); (b) monthly public reporting of withdrawals, consumption, and discharge, by source; (c) a contractual waiver of trade-secret and confidentiality claims over water quantity data; (d) no NDAs binding public officials that survive the application filing.

Justification: Every other demand in this chapter is unenforceable without measured, public numbers. The industry's own conduct proves the point: trade-secret claims in Wisconsin and South Carolina, NDAs in Michigan and Virginia, a city suing to keep usage secret in Oregon. A 2016 survey found fewer than a third of operators even tracked water consumption (EESI). Quantity data reveals no protectable technology — what it reveals is impact.

Benchmark: The Racine trade-secret claim failed when challenged in court — public water records are public. Utah's 2026 Transparency Amendments establish disclosure as state policy (demand actuals, not just the estimates Utah settled for). Minnesota requires a dedicated water permit. ELI's January 2026 fact sheet names public reporting of water use and cooling technology as essential to planning.

Demand 2 — A hard cap at full buildout, with penalties and a re-opener

The ask: Enforceable maximum daily and annual consumption and withdrawal limits at full buildout, with per-gallon liquidated damages for exceedance and a mandatory renegotiation trigger for any expansion, change of cooling technology, or change of operator.

Justification: Without a cap, every disclosed number is trivia. The phasing trap (Part 1, §1.1) applies with full force to water: a Phase 1 commitment means nothing if Phases 2–4 arrive under administrative approvals. Newton County's deficit projection shows what uncapped growth does to a county's water math.

Benchmark: Water-service agreements are ordinary contracts — municipal utilities cap large industrial users routinely. ELI: governments must plan for average and peak demand and incorporate data center cooling profiles into long-term supply planning.

Demand 3 — Cooling technology and a contractual WUE

The ask: (a) In water-stressed basins, prohibit evaporative cooling as the primary heat-rejection method — require closed-loop, direct-to-chip, air, or immersion designs; (b) everywhere, a binding facility WUE ceiling at full buildout, reported against meter data.

Justification: Low-water cooling is commercially standard — the AI hardware transition is pushing the industry to liquid closed-loop designs anyway (Data Centre Magazine, Feb 2026), and Microsoft now advertises zero-evaporation direct-to-chip systems. A developer claiming evaporative towers are unavoidable is negotiating their capex, not stating physics. Pair this demand with Part 2's efficiency demand so the energy penalty of dry cooling is managed rather than weaponized.

Benchmark: Amazon's own fleet WUE of 0.15 L/kWh (vs. an industry average several times higher) proves what's achievable — hold every operator to the standard their marketing departments already claim.

Demand 4 — Reclaimed and non-potable water first

The ask: Maximum feasible use of reclaimed/recycled water, with a contractual floor (e.g., 50%+ of cooling demand where infrastructure exists or can be built at developer expense), and developer funding of purple-pipe extensions where it doesn't.

Justification: Most data centers still run on drinking water — alternative sources average under 5% of typical supply (Project Finance Law) — yet the model is proven at scale: Loudoun Water operates a reclaimed-water service for data center customers in the densest data center market on earth; Google runs over 25% of its campuses on reclaimed or non-potable water, including Douglas County, Georgia entirely on recycled municipal wastewater; AWS reports 20+ facilities cooling with purified wastewater; Phoenix operates a dedicated recycled-water distribution system serving multiple data centers (FWPCOA; Genesis Water Technologies, 2025). Treated wastewater even costs operators 30–50% less than potable supply — you are demanding something that saves them money.

Benchmark: Loudoun County, VA; Douglas County, GA; Phoenix, AZ; Santa Clara, CA (Vantage). If those markets can, yours can.

Demand 5 — Drought priority: the facility curtails before residents do

The ask: Mandatory consumption reductions tied to your jurisdiction's official drought stages — with the data center cutting first and deepest, junior in priority to residential, agricultural, and ecological uses; verified by the Demand 1 meters.

Justification: Evaporative demand peaks exactly when supply is scarcest. The Dalles was asked for more Google water while 98% of its county sat in extreme drought. Without contractual priority, the entity with the long-term supply contract — the data center — is functionally senior to the residents whose council signed it. Operators can engineer for this: on-site storage, pre-cooling, workload shifting, and temporary dry-mode operation are the water analogues of the demand-response capability they already acknowledge on the energy side (Part 2, Demand 11).

Benchmark: Drought-stage curtailment schedules are standard in municipal codes for car washes and golf courses; extending them to the largest user in the system is not radical — exempting it is.

Demand 6 — Baseline hydrogeology and independent monitoring, at developer expense

The ask: Before any ground is broken: an independent (community-selected, developer-funded) baseline study of the aquifer, neighboring wells, surface waters, and recharge areas — then permanent monitoring wells with public, continuous data, for the life of the facility.

Justification: This is the Newton County lesson in one line: no baseline, no accountability. Because the JDA did no pre-construction well study, every harmed family argues causation against the company's hired hydrologists, forever. A $200,000 study is rounding error on a $750 million project and the difference between enforceable rights and a decade of he-said-she-said.

Benchmark: Project-finance counsel already advise developers that water-availability assessments and drought-mitigation plans are standard diligence (Project Finance Law, 2025) — your community should simply require that the study be independent, public, and pre-construction.

Demand 7 — The well-protection presumption and compensation fund

The ask: A rebuttable presumption that degradation of wells within a defined radius (commonly 1–2 miles) occurring after construction begins is attributable to the project — entitling owners to immediate interim water supply, and permanent remedy (well deepening, replacement, or municipal hookup) at developer expense, administered through a pre-funded escrow with a neighborhood claims process.

Justification: Beverly Morris spent $5,000, replaced her appliances three times, hauls buckets to flush her toilet, and cannot afford a new well — while the company's commissioned study found itself not responsible and the EPA explained that no agency regulates her well at all. The presumption flips the burden of proof from the retiree to the trillion-dollar company, which is where engineering knowledge, hydrological data (Demand 6), and financial capacity actually reside. Oil, gas, and mining law uses exactly this structure for the same reason.

Benchmark: Pre-drilling water-supply replacement presumptions in Pennsylvania's oil and gas law (within a distance/time window of drilling) are the established template; apply it to dewatering and blasting.

Demand 8 — Construction-phase water rules

The ask: Permits and limits for construction dewatering (volumes, discharge points, duration), blasting plans with seismic and well-impact monitoring, and enforceable sediment/stormwater controls — all subject to the Demand 6 monitoring network and Demand 7 presumption.

Justification: The Morris family's taps failed during construction, years before "operations" began — the phase most agreements ignore entirely. Hydrologists confirm large-site dewatering foreseeably affects neighboring wells (NYT). If your agreement's water provisions begin at commissioning, the worst damage may already be done.

Benchmark: Dewatering permits are routine in heavy construction; the demand is simply that they be conditioned, monitored, and tied to the compensation mechanism.

Demand 9 — Discharge: chemistry, temperature, and pretreatment

The ask: Full disclosure of cooling-water treatment chemicals; a pretreatment agreement with your wastewater utility sized to actual blowdown volumes; temperature and chemistry limits on any surface discharge; and developer-funded monitoring of receiving waters — with data public.

Justification: Cooling water is dosed with corrosion inhibitors and biocides; blowdown is industrial wastewater. St. Charles, Missouri adopted a moratorium amid resident concern over potential links between data center wastewater discharge and drinking-water contaminants — an open scientific question that your utility should not answer by experiment (TechPolicy.Press, Nov 2025). Open-loop designs add thermal pollution to the list.

Benchmark: Clean Water Act pretreatment programs exist precisely for this; the demand is to apply them at the right scale and make the data public.

Demand 10 — Full-cost pricing and infrastructure: no subsidized water, no rate shock

The ask: (a) The facility pays the full marginal cost of supplying it — including its share of new treatment, storage, and conveyance capacity — via a large-user rate class or capacity charges; (b) a contractual commitment, with indemnity, that no system expansion costs attributable to the facility appear in residential rates; (c) take-or-pay minimums mirroring Part 2's Demand 4, so reserved water capacity is paid for whether used or not.

Justification: This is Part 2's cost-causation principle in water form. Mansfield's mayor reports water rates rising 33% in two years against a historical norm of ~2% annually — residents subsidizing the infrastructure strain a single industrial user created. Newton County's authority is candid that new data center requests exceed available supply; whoever funds the expansion owns the politics of the next rate case.

Benchmark: Large-industrial water rate classes and capacity charges are standard utility practice; the West Des Moines/Microsoft agreements show infrastructure cost-sharing can be negotiated explicitly rather than absorbed silently.

Demand 11 — Groundwater accounting and no water grabs

The ask: (a) All facility groundwater pumping is metered, permitted, and counted against basin-level budgets even where state law doesn't require it; (b) prohibition (or community right of first refusal) on the developer acquiring agricultural water rights, neighboring wells, or land primarily for its water; (c) cumulative-impact review covering all proposed facilities in the basin, not each in isolation.

Justification: In Louisiana, groundwater pumping for the largest data center ever built requires no permit at all; in much of the country, the facility next door legally drinks from the same straw as your town with no accounting (E&E News). Meanwhile site-shopping means your basin may be evaluating one project while five are quietly queued — the water version of Part 2's phantom-load problem. ELI flags water-rights acquisition by data centers as a rising legal-equity issue in constrained regions.

Benchmark: Minnesota's dedicated permitting requirement; the Great Lakes Compact's demonstration that strong regional water law defeats secrecy and overdraft alike.

Demand 12 — Standing, enforcement, and the water authority's seat

The ask: (a) The water-service agreement names residents as third-party beneficiaries with direct enforcement rights; (b) service is contractually conditioned on compliance with Demands 1–11, with suspension as the remedy of last resort; (c) your water authority — not just the city attorney — sits at the negotiation table from day one; (d) annual public true-up of all metered data against caps, with the burden on the operator.

Justification: Part 1's enforcement rule applies doubly here: benefits without teeth are PR, and water promises without meters, penalties, and standing are marketing. Lancaster's CBA was faulted for limiting residents' ability to sue; your water agreement must do the opposite. The utility that controls the valve holds the most concrete enforcement power in this entire handbook — write the agreement so it may lawfully use it.

Benchmark: Third-party beneficiary clauses are standard contract drafting; the NAACP CBA template's independent-monitoring and community-oversight provisions supply the governance architecture.


3.6 Where each fight happens: the water jurisdiction map

Venue What's decided there Your tools
Municipal water utility / authority Service agreements, caps, rates, drought rules, pretreatment Demands 1, 2, 5, 9, 10, 12 — the contract is the law here
City / county board Rezoning & CUP conditions, construction permits, CBA Demands 3, 6, 7, 8 as conditions of approval
Groundwater / basin authority Well permits, basin budgets, cumulative impacts Demand 11; intervene in permit proceedings
State legislature & agencies Disclosure laws, dedicated permitting (MN model), drought law Push actual-use disclosure (beyond Utah's estimates); oppose preemption
State environmental agency / EPA Discharge permits (NPDES), pretreatment oversight Demand 9; comment periods and permit appeals
Courts / FOIA Trade-secret and records fights The Racine precedent: sue, and win

Sequencing matters here too. The water utility's leverage is greatest before the service agreement is signed — and your board's leverage is greatest before rezoning. Demands 6–8 (baseline, presumption, construction rules) are worthless retroactively; they exist only if they're in writing before the first shovel.


3.7 The asks at a glance

# Demand Benchmark Primary venue
1 Metered public disclosure, no trade secrets Racine FOIA win; Utah & MN statutes Utility + agreement
2 Hard caps + penalties + re-opener Standard large-user contracting Utility agreement
3 Cooling tech mandate + binding WUE Microsoft zero-evaporation; 0.15 L/kWh fleet WUE CUP + agreement
4 Reclaimed water first Loudoun, Douglas Co., Phoenix, Santa Clara Utility + agreement
5 Drought curtailment priority Municipal drought-stage codes Utility agreement
6 Baseline study + permanent monitoring Standard developer diligence, made public CUP condition
7 Well-impact presumption + fund PA oil & gas replacement presumption CUP + agreement
8 Construction dewatering/blasting rules Routine heavy-construction permitting CUP condition
9 Discharge chemistry + pretreatment CWA pretreatment; St. Charles caution Utility + state agency
10 Full-cost pricing; no rate shock Newton Co. 33% hike as the cautionary tale Utility rate class
11 Groundwater accounting; no water grabs MN permitting; Great Lakes Compact Basin authority
12 Standing + valve-backed enforcement NAACP template governance Agreement

3.8 References

Consumption data and technology

Secrecy and disclosure law

Community impacts

Figures current as of June 2026. Basin conditions, drought stages, and state disclosure law change rapidly — verify your state's current statute and your basin's status before testimony.