Data Center Water Use Tracker

The core CWA machine: §301 makes any point-source discharge of a pollutant to a water of the United States unlawful unless authorized by a §402 NPDES permit with numeric effluent limits, monitoring (DMRs), and reporting. Most enforcement in this record — from Smithfield's $12.6M judgment to routine state NOVs — is §402/§301 enforcement.

Applies whenever a data center discharges cooling-tower blowdown, on-site treatment effluent, or process water directly to surface water rather than to a sewer. The AWS Lake Anna VPDES draft permit is the first direct cooling-discharge NPDES action at a hyperscaler campus; the Homer City PA draft permit covers cooling outfalls at a DC power plant. Once a campus holds its own NPDES permit, every exceedance is ECHO-visible and enforceable — including by citizen suit.

Land disturbance over one acre requires coverage under a construction general permit with a SWPPP, erosion/sediment BMPs, and inspection records. This is the single most-enforced CWA program against large construction projects.

The primary CWA touchpoint for data centers today: multi-hundred-acre campus build-outs generate sediment-laden runoff, and every documented DC stormwater enforcement (Google Stillwater OK, Microsoft Boydton VA, QTS Fayetteville GA, Johns Hopkins DSAI) arose here. Enforcement usually names the general contractor — the permittee — not the operator.

Operating industrial facilities with materials or equipment exposed to rain need MSGP coverage with benchmark monitoring for pollutants in site runoff.

Post-construction exposure for campuses with exposed fuel farms, chemical storage, switchyards, and equipment yards. Most operating data centers hold only this near-empty stormwater permit — which is why the operational record is thin — but benchmark exceedances are public and enforceable.

Placing fill in jurisdictional wetlands or streams requires a Corps permit — an individual permit for larger impacts or a nationwide permit (NWP) for smaller ones. Sackett (2023) narrowed which wetlands are jurisdictional; unpermitted fill draws cease-work orders and restoration liability.

Site-grading at greenfield campuses regularly crosses wetlands and headwater streams: Google's three pending §404 individual permit applications (Project Raspberry and Loch in VA, Port of Little Rock AR at 16.8 acres), the SDC ATLA Douglas County GA application with ~1 mile of stream impacts, and the Amazon New Carlisle IN cease-work order all sit here. The 2026 NWP 39 reissuance expressly enumerates data centers.

Any federal permit that may result in a discharge (a §404 fill permit, a FERC license) needs the state to certify the project meets state water-quality standards. States can condition or deny certification — an independent veto point — and courts have vacated deficient certifications.

Every data-center §404 permit needs a companion §401 certification, so state agencies (and commenters in the state proceeding) get leverage even where federal jurisdiction survives Sackett only barely. The SDC ATLA GA matter pairs §404 with a contested §401; the 4th Circuit's WV vacatur shows certifications are litigable.

Industrial users discharging to a municipal sewer must meet national prohibitions (no pass-through or interference) and locally issued industrial-user permits with local limits. The POTW enforces; EPA enforces against POTWs whose pretreatment programs fail.

The regime that most directly governs operating data centers: cooling-tower blowdown (concentrated dissolved solids, biocides, corrosion inhibitors) goes to the sewer under an industrial-user permit. High-volume DC loading can cause POTW pass-through — putting the plant in violation of its own NPDES permit, as at Port of Morrow OR. IU permits and local limits are FOIA-able.

Heat is a regulated pollutant: §316(a) governs thermal discharge variances and §316(b) requires cooling-water intake structures to reflect the best technology available for minimizing fish impingement and entrainment. Historically applied to power plants (Brayton Point's forced closed-cycle retrofit; Greenidge's bitcoin-era findings).

Bites hardest where a campus takes surface water for once-through or evaporative cooling, or pairs with on-site generation — the full power-plant profile. The AWS Lake Anna draft permit and the Homer City DC power plant put §316 review squarely into data-center permitting for the first time.

Prohibits harmful-quantity discharges of oil or listed hazardous substances to waters of the U.S., requires Spill Prevention, Control and Countermeasure plans for facilities storing oil above thresholds, and carries per-barrel civil (and negligence-based criminal) penalties — the Plains Pipeline and East Palestine lines.

Data centers keep multi-million-gallon diesel fuel farms for backup generators — squarely SPCC-regulated. A tank or transfer release reaching a stream is legally identical to the pipeline cases, including the duty-to-report trap. The Johns Hopkins DSAI matter already pairs construction stormwater with a §311 diesel element.

States set designated uses and criteria for each water; new or expanded discharges into high-quality (Tier 2) waters trigger an antidegradation alternatives analysis, and impaired waters get TMDL load caps that flow into permit limits.

A procedural lever at the siting stage: a proposed new DC outfall into high-quality water must survive an alternatives analysis (e.g., closed-loop or dry cooling), and a discharge into a TMDL-capped watershed inherits tighter limits. Raised by commenters in the Lake Anna proceeding.

County of Maui v. Hawaii Wildlife Fund held that a discharge that travels through groundwater but is the functional equivalent of a direct discharge to surface water still needs an NPDES permit — injection wells, leach fields, and land application can't launder a point source.

Closes the loophole for campuses that dispose of blowdown or treated effluent via injection wells, drywells, rapid-infiltration basins, or land application near streams: if it reaches surface water, Maui says permit it. Fact-intensive (distance, travel time), but the most potent novel theory for inland campuses over shallow aquifers.

Any citizen may sue a discharger for ongoing permit violations (or EPA for non-discretionary failures) after 60-day notice, with civil penalties payable to the Treasury and fee-shifting. Riverkeeper groups built much of the modern enforcement record this way; Port of Tacoma and CERF v. Naples set current scope/mootness limits.

The enforcement backstop when agencies stand down: DMR exceedances at a DC's own NPDES permit, at its contractor's CGP coverage, or at the receiving POTW are all citizen-suit predicates — and this tracker's ECHO SNC/DMR pull is exactly the evidence base a §505 notice letter needs. QTS Fayetteville and TransGas Adams Fork began as riverkeeper notices.

When a contaminant present in or likely to enter a public water system or underground source of drinking water may present an imminent and substantial endangerment, EPA can issue immediate administrative orders or sue — without waiting for state action or a permit violation. Used against DuPont for PFOA in Ohio Valley drinking water and in Flint (2016).

The fastest federal hook if a data-center-linked contaminant (fuel release, cooling-chemistry constituents, construction-driven nitrate or sediment) shows up in wells or a public system's source water. The Amazon Boardman nitrate fact pattern — contaminated domestic wells near DC-hosting farmland — is exactly the §1431 profile, and EPA has pledged investigation of the Meta Newton County GA well complaints.

Any subsurface fluid emplacement is an 'injection well.' Class V covers shallow, non-hazardous injection — stormwater drywells, aquifer recharge, cooling-water return-flow wells, large septic systems. Wells that may endanger an underground source of drinking water can be closed or permitted with conditions; EPA banned entire well subclasses (motor-vehicle waste wells, 1999) where the endangerment was categorical.

Directly regulates common campus infrastructure: stormwater drywells and infiltration galleries, geothermal/aquifer-thermal cooling return wells, and any on-site disposal of treated effluent to the subsurface are Class V wells requiring inventory, rule authorization, or permits — an underused federal hook over inland campuses that never touch surface water.

Where EPA has designated an aquifer as the sole or principal drinking-water source for an area (Edwards Aquifer, 1975, was the first), no federal financial assistance may go to a project that EPA determines may contaminate the aquifer. A review trigger, not a general permit program.

Reaches data-center projects only through a federal-funding hook — federally backed loans, grants, or infrastructure assistance for a campus or its supporting utilities over a designated aquifer trigger EPA review. Relevant to Texas Hill Country and San Antonio-area siting over the Edwards, and to any federally financed AI-infrastructure build-out.

Systems serving 25+ people or 15+ connections are public water systems bound by maximum contaminant levels, monitoring, and reporting; chronic violators face escalating enforcement up to court-appointed receivership (Alisal Water).

Two directions: (1) a large campus with its own wells serving hundreds of workers can itself become a regulated PWS; (2) the small municipal systems that serve mega-campuses take on PWS compliance risk when DC demand strains capacity or forces new sources — the Richland Parish LA and Bessemer AL supply fights are about systems absorbing that risk.

Every state maintains EPA-approved source-water assessments delineating the recharge and intake zones that feed public systems; many states and localities layer land-use restrictions (wellhead protection ordinances, recharge-zone rules) on top. Largely non-regulatory federally, but the assessments are the evidentiary backbone for local fights.

The mapping layer for siting fights: a campus, fuel farm, or drainfield inside a delineated source-water protection area gives residents and utilities a concrete, government-authored basis to object — the posture of the Occoquan Reservoir objections to the PW Digital Gateway and the aquifer-recharge zoning fights in Texas and Georgia.

Anyone who manufactures, processes, or distributes a chemical and obtains information reasonably supporting a conclusion of substantial risk to health or the environment must tell EPA immediately. DuPont's failure to report PFOA blood and drinking-water data drew what was then the largest civil administrative penalty in EPA history ($10.25M + $6.25M SEPs, 2005); 3M settled similar PFAS non-reporting in 2006.

Reaches the chemical suppliers behind data-center water chemistry — cooling-tower biocides, corrosion inhibitors, film-forming additives, and legacy AFFF fire suppressants. If a supplier or large processor learns its formulation is contaminating water near campuses and sits on it, that is the DuPont fact pattern.

The 2023 rule requires anyone who manufactured or imported PFAS (including in articles) since 2011 to report uses, volumes, byproducts, exposure, and disposal — no de minimis exemption. The resulting dataset maps who put which PFAS where.

Indirect but growing: imported electrical equipment, coolants (some immersion-cooling fluids are PFAS), and fire-suppression stocks can pull vendors into reporting, and the public dataset gives water advocates a way to trace PFAS found near a campus back to a supply chain as NPDES PFAS limits tighten (see the Chemours Washington Works effluent-limits consent agreement).

Where the handling or disposal of any solid or hazardous waste may present an imminent and substantial endangerment, EPA or any citizen can obtain injunctive cleanup orders — no permit violation needed and no CWA jurisdictional water required. Interfaith v. Honeywell forced a ~$400M chromium cleanup on this theory.

The gap-filler where Sackett cuts off CWA jurisdiction: contamination migrating through soil and groundwater from fuel releases, construction fill, sludge, or spent cooling-treatment chemicals at a campus can support a §7002(a)(1)(B) citizen suit even with no surface-water connection at all — the same post-Sackett role Maui plays for connected waters.

Building any structure in, or affecting the course/capacity of, traditionally navigable waters requires a Corps §10 permit — intakes, outfall structures, piers, pipeline crossings. Republic Steel (1960) read the Act to reach industrial deposits that reduce channel depth. Unlike CWA §404, §10 jurisdiction was untouched by Sackett.

Any data-center or DC-power-plant cooling intake or outfall structure placed in a navigable river or lake needs §10 authorization on top of NPDES/§404 — a second, Sackett-proof Corps permit gate (relevant to Lake Anna-style surface-water cooling and river-supplied campuses).

Hughes River Watershed Conservancy v. Glickman (4th Cir. 1996) held the Army Corps and Natural Resources Conservation Service violated NEPA on a West Virginia dam project by using an inflated recreational-benefit estimate that skewed the EIS's cost-benefit picture, and by failing to seriously analyze zebra-mussel infestation risk before declining to prepare a supplemental EIS. On remand the agencies recalculated benefits using net rather than gross figures and commissioned species-impact studies; the Fourth Circuit's 1999 sequel (Hughes River Watershed Conservancy v. Johnson) held the corrected analysis satisfied NEPA's hard-look standard.

A data center's own construction is rarely a 'major federal action,' but the federal actions clustered around one often are — a DOE loan guarantee, an NRC license amendment, a Corps permit, or federal financing for supporting transmission or water infrastructure. Hughes River sets what those reviews must actually do: real analysis (not inflated benefit figures or a boilerplate species screen), and a duty to go back and supplement if a materially different picture — a changed water-supply or grid-reliability forecast, for instance — emerges before the agency acts.

Restarting a shuttered nuclear plant runs through at least two independent federal actions, each carrying its own NEPA review. For the Crane Clean Energy Center (Three Mile Island Unit 1) restart, DOE's Loan Programs Office issued a final EIS in August 2025 covering its loan-guarantee decision, while NRC separately released a draft environmental assessment and finding of no significant impact in June 2026 for the license-reauthorization decision — NRC staff determined a full EIS was unnecessary for that action, a threshold call that is itself reviewable.

The live, current version of a federal-financing NEPA trigger in the AI build-out: hyperscalers are underwriting nuclear and gas-plant restarts and new generation to meet data-center power demand, and every dollar of federal loan support or every NRC license action on those plants is an independently NEPA-reviewable action — even though the resulting EIS or EA typically analyzes only the plant's own radiological and environmental footprint, not the data-center deal driving the restart. Whether an EA-plus-FONSI is legally adequate, versus requiring a full EIS, is exactly the adequacy question Hughes River's hard-look standard was built to answer.

United States v. Bestfoods (1998) held CERCLA §107(a)(2) reaches whoever actually operated a contaminated facility. A parent corporation is not automatically liable just because its subsidiary owned or ran the polluting site — ordinary corporate separateness survives — but a parent that itself actively managed, directed, or made compliance decisions specifically about the facility's pollution-generating operations can be directly liable as an operator in its own right, independent of any veil-piercing theory.

Relevant to how hyperscalers structure campus ownership — sites are frequently held in special-purpose subsidiaries or leased from developers, while the parent's facilities or sustainability group makes centralized decisions about cooling-system chemistry, generator fuel storage, or remediation at a site with legacy contamination. Bestfoods says centralized parent-level control over the environmentally sensitive decisions, not the corporate org chart, is what determines CERCLA exposure if a plume later turns up under or near the campus.

CTS Corp. v. Waldburger (2014) held that CERCLA §9658 — which overrides state statutes of limitations so a contamination claim's clock starts at discovery rather than at the polluting act — does not also override state statutes of repose, which cut off claims after a fixed number of years regardless of when the harm was discovered. North Carolina property owners who learned of TCE groundwater contamination from EPA in 2009, decades after the polluting plant closed, had their state nuisance claims barred by the state's 10-year repose period.

A data center that later discovers its water supply well or campus soil is contaminated by a decades-old industrial release nearby may still have CERCLA's own federal cleanup remedies available, but a separate state-law tort claim against the original polluter (nuisance, trespass, diminished property value) can be time-barred by a state repose statute no matter how recently the contamination was actually discovered — a real limit on what an operator, or a neighboring landowner, can recover once EPA's own process is not in play.

Metro Atlanta's decades-long 'water wars' defined this ceiling. Southeastern Federal Power Customers v. Geren (D.C. Cir. 2008) held that a 2003 Corps settlement letting water utilities draw far more municipal/industrial supply from Lake Lanier's storage — at hydropower's expense — was a 'major operational change' that §301(d) does not let the Corps make unilaterally; it needs specific congressional authorization. In re Tri-State Water Rights Litigation (11th Cir. 2011) then held, on a related piece of the same fight, that water supply was in fact one of the Buford Project's originally authorized purposes under its 1946 authorization — giving the Corps more room than the district court had allowed, but still bounded by that authorized-purposes ceiling. A physical alteration of a Corps project — an intake, an outfall, a pipeline crossing a levee — needs its own, separate permission under 33 U.S.C. §408, evaluated against the same question of whether it impairs the project's authorized purposes.

A hyperscale-driven metro utility asking the Corps for more storage out of a reservoir built for flood control, hydropower and navigation runs directly into this ceiling — exactly what the 2026 Water Resources Development Act (H.R. 9497, ordered reported July 2026) responds to: it would direct the Corps to report within a year on how 'new commercial and industrial water users' — a category the accompanying Congressional Research Service analysis (R49057, July 2026) says data centers fit — are affecting reservoirs authorized for water supply. Separately, a campus's own cooling-water intake structure built into or near a Corps flood-control or navigation project needs §408 permission independent of any NPDES or §404 permit it also holds.

Section 7 bars FERC from licensing a dam, water conduit, reservoir, or other project works 'on or directly affecting' a designated Wild and Scenic river, and bars other federal agencies from assisting a water-resources project with a direct and adverse effect on the values the designation protects. The clearest historical demonstration: after a decade of fighting Appalachian Power's proposed Blue Ridge Project — a two-reservoir pumped-storage hydro complex the Federal Power Commission had licensed in 1974 on the upper New River — Congress placed 26.5 miles of the river into the National Wild and Scenic Rivers System in September 1976, and the dam was never built.

Narrow but absolute where it applies: a data-center campus, or the power or water infrastructure built to serve one, that would touch a designated Wild and Scenic river corridor cannot get a FERC hydropower license or most other federally assisted water-resources authorization for that reach — no matter how far along the project already is, since a designation can moot a license that has already been granted. Due diligence on any western or Appalachian site with river-adjacent power or cooling infrastructure should check the designated and study-river lists before assuming a federal water permit is available.

California v. FERC (1990) held the Federal Power Act preempts a state's attempt to impose its own, different minimum stream-flow requirements on a FERC-licensed hydroelectric project — FERC's license terms occupy that field, because letting a state layer on stricter or different flow rules would hand it a veto over federally licensed projects that the FPA does not give it.

Matters wherever a data center, or the power feeding one, sits at or draws from a FERC-licensed hydropower reservoir — increasingly real, since cheap hydro basins in the Pacific Northwest have already drawn large computing loads (crypto-mining operations at Chelan and Grant County PUD dams in Washington are the existing precedent for large computing co-locating at hydro facilities). The reservoir's water levels and release schedule are set by the FERC license, not by state water law, so a data center's cooling-water draw or a utility's request to hold water back for it has to work within license terms that only FERC, through relicensing or a license amendment, can change.

United States v. Klamath Drainage District (9th Cir., decided Jan. 2025, No. 23-3404) held that a 1946 repayment contract let Reclamation control how much water the Klamath Drainage District could divert through the federal irrigation project — and that Reclamation's authority reached even a separate, later-acquired 1977 state water right the district tried to exercise through its own canal, because the contract's 'reasonable rules and regulations' language gave Reclamation that scope. The Ninth Circuit affirmed a 2023 injunction against the district's unauthorized diversions.

Any western data-center campus — or the agricultural water right it buys, leases, or offsets against — that ultimately traces back to a federal Reclamation project takes that water subject to the delivery contract's terms, not just the nominal water right. The Central Arizona Project water that Tucson-area operators are relying on for offset credits is the live example. Reclamation can enforce the contract directly, by injunction, against a water user who diverts more than allocated, even where that user also holds an independent state-law right to the same water.

Steel Co. v. Citizens for a Better Environment (1998) held a citizen group lacked Article III standing to sue over a company's years of missed Tier II hazardous-chemical inventory filings once the company filed everything late but before the suit was brought — because none of the relief EPCRA's citizen-suit provision offers (civil penalties payable to the U.S. Treasury, cost recovery) would redress a purely past, already-cured violation.

A data center, or its fuel-farm or chemical-storage contractor, that misses a Tier II filing deadline for backup diesel or cooling-treatment chemicals faces real exposure to a citizen-suit notice letter — but Steel Co. means filing the overdue report before litigation is filed can moot the suit entirely, leaving state regulators or EPA's own enforcement discretion, not a community group, as the only route to a penalty for the historical lapse.

EPCRA §313 requires facilities that manufacture, process, or otherwise use a listed toxic chemical above threshold quantities to file an annual public Form R disclosing releases and waste management — separate from, and in addition to, Tier II's hazardous-chemical-storage inventory. EPA's July 2024 enforcement sweep against several Georgia manufacturers, including a $34,730 penalty against Nidec Elesys Americas for failing to file Form R for lead compounds processed at its Suwanee facility in 2021–2022, shows the agency still actively enforces routine TRI non-filing with real penalties, not just Tier II lapses.

TRI reporting reaches chemical processing that Tier II's storage-threshold framing can miss — a data-center campus's on-site water-treatment system, biocide dosing, or backup-power battery operations could cross a TRI processing threshold for a listed chemical even where storage volumes alone would not trigger Tier II. The resulting public Form R database is one more FOIA-free source, alongside this tracker's own EPA ECHO pulls, for tracing a campus's chemical footprint without a records request.

Disputes between states over a shared water resource. Historically applied to rivers; in 2021 the Supreme Court held for the first time that groundwater in an interstate aquifer is equally subject to equitable apportionment — no state owns the water beneath it outright, and a state claiming injury must sue for an apportionment rather than for conversion.

A data center pumping heavily from an aquifer that crosses a state line creates precisely the fact pattern this doctrine governs, and the injured party is the neighbouring *state*, not the neighbouring landowner. The practical barrier is evidentiary: an apportionment claim needs clear-and-convincing proof of real injury, which means utility-level aggregate withdrawal data — the class of data the HB 496-style reporting laws tracked in this dataset are starting to generate.

Limits on a state's power to restrict the export of water across its borders. Groundwater is an article of commerce, so a state law blocking or conditioning its transfer out of state faces Commerce Clause scrutiny and must be genuinely tailored to conservation rather than to hoarding.

Runs in the opposite direction from most doctrines here: it is the constraint on a *state* that tries to keep its water away from out-of-state data-center demand. A statute written to stop water leaving for a neighbouring state's campus is more vulnerable than one that limits large withdrawals regardless of where the user sits — which is why the durable instruments in this dataset regulate by volume and use, not by destination.

The 2008 Compact bans new or increased diversion of Great Lakes water outside the basin, with narrow exceptions including supply to a 'straddling community' that sits astride the basin line. Wisconsin approved exactly that exception in 2018 to let Racine Water Utility pipe up to 7 million gallons per day of Lake Michigan water to the Foxconn campus in Mount Pleasant, most of it (5.8 MGD) for that single industrial customer. Environmental groups argued the diversion was not really for 'public water supply' and violated the Compact's terms; a Wisconsin administrative law judge rejected that argument in June 2019 and upheld the approval, and the challengers chose not to appeal further.

The Foxconn site is now Microsoft's Mount Pleasant data-center campus, so this is not hypothetical: any further increase in the Racine diversion to serve additional data-center buildings runs through the same straddling-community approval and the same 'is this really public water supply' argument that already lost once. Any other Great Lakes-basin hyperscale campus that is not itself inside the basin faces the same Compact gate before a single gallon can leave the lake.

Unlike most of the country, any withdrawal above a threshold in these two basins needs the commission's own approval, independent of state permitting. SRBC has used that authority both to approve large industrial withdrawals with conditions — its June 4, 2026 approval letting the Crane Clean Energy Center (the restarting Three Mile Island Unit 1, explicitly tied to Constellation's data-center power agreement with Microsoft) draw up to 73.2 million gallons a day, conditioned on curtailing withdrawals during low-flow periods — and to fine violators, as in its 2016 proposed penalties (roughly $97,000, including $44,250 for the Patriot plant in Lycoming County and $22,750 for Hummel Station in Snyder County) against Panda Power Funds for drawing from unapproved sources and exceeding limits while commissioning two Pennsylvania gas plants. DRBC has gone further still, voting in February 2021 to permanently ban high-volume hydraulic fracturing basin-wide — an outright prohibition on an entire water-intensive industrial activity, not just a permit condition.

A data center, or a power plant built to supply one, sitting inside either basin does not go through a normal state-only withdrawal permit — it goes through a public commission docket that can approve with drought conditions attached (the SRBC/Crane model), fine for taking more than authorized (the Panda Power model), or, in the most basin-protective posture, decide the activity is not compatible with the basin at all (the DRBC fracking-ban model). That full range of outcomes is on the table for any large new water user in either basin, and the dockets themselves are public.

The state holds navigable waters in trust and retains a continuing duty of supervision. A previously granted water right is not settled forever: it remains subject to reconsideration and adjustment where the diversion harms trust interests and a feasible alternative exists.

The doctrine that makes an already-issued water allocation reviewable. A data center supplied under a long-standing municipal or appropriative right is not insulated by that right if the resulting draw damages a trust resource — which matters because most data-center supply arrives through existing utility entitlements rather than new permits, and those are exactly what this doctrine reopens.

Groundwater pumping that measurably reduces flow in a navigable waterway falls within the public trust, and a comprehensive statutory groundwater regime does not displace that duty. The permitting authority — often a county — must consider trust resources when issuing well permits.

Closes the gap a data center's hydrology usually falls into. Campus wells are rarely regulated as a surface-water diversion, but where drawdown depletes a connected stream the trust duty attaches to the well permit itself. It also means a state groundwater-management statute is a floor, not a ceiling: complying with the statute does not answer the trust claim.

The trust establishes a presumption in favour of its purposes. The burden sits on the party seeking water for private gain to justify the use and its impacts, not on the public to prove harm; and where the science is unresolved, that uncertainty counts in favour of protection rather than against it.

Directly inverts the usual posture of a data-center water fight. The recurring complaint in this dataset is that opponents cannot prove harm because consumption figures are confidential — under this reading the applicant carries the burden, and unresolved hydrology argues for the resource. Its force is jurisdiction-specific: Hawai'i's constitutional water provisions are unusually strong, so this is the ceiling of the doctrine rather than its typical reach.

In states following ownership-in-place, a landowner holds a constitutionally protected interest in the groundwater beneath the surface before it is pumped. Regulation that restricts pumping can therefore support a takings claim requiring compensation.

Cuts both ways, which is why it belongs in the registry. It is the doctrine a data center would invoke against a withdrawal cap imposed after it acquired the land — and equally the one a neighbouring landowner invokes when a regulator permits the campus to draw down the property interest under their own acreage. It also explains why states in this family regulate through districts and permits rather than by prohibition: outright restriction carries a compensation risk.

Whether an agency issuing a high-capacity well permit must look beyond the applicant's own parcel. Where the state holds navigable waters in trust and the permitting statute protects waters of the state generally, the agency has both authority and a general duty to consider whether the proposed well may harm those waters — but the duty is triggered by concrete scientific evidence of potential harm, not by objection alone.

The most practically usable doctrine in the registry for an opponent, because it operates at the permit stage rather than after the damage. It also sets the price of admission: objections to a data-center well must arrive as hydrologic evidence, since generalized concern does not trigger the agency's duty. That is precisely the bind sites in this dataset describe — the evidence needs consumption data the operator holds.

Inside an Active Management Area, groundwater is governed by statute rather than by common-law pumping rights: withdrawals are permitted and metered, and new development must demonstrate an assured supply — water physically and legally available for a century — before it can proceed.

This, and not the common-law reasonable-use rule, is what an Arizona data center actually answers to, which is why Arizona fights look different from Georgia or Texas ones: the question is decided in a statutory supply-designation process rather than in a nuisance suit. It cuts both ways — the regime is a real constraint on new demand, and it also gives a project that clears it a defensible answer to critics.

Requires local agencies in high- and medium-priority basins to adopt plans achieving sustainable yield and avoiding undesirable results including chronic lowering of groundwater levels. Critically, the statute is a floor rather than a ceiling: enacting it did not occupy the field or extinguish the public-trust duty that runs alongside it.

Sets the numeric constraint a large new California groundwater user has to fit inside, and answers the compliance defence in advance. An operator pointing to its sustainability-plan allocation has not thereby answered a public-trust objection, because the two obligations coexist.

No one may divert state water from one river basin to another without a water right or permit amendment authorizing the transfer, and any right transferred out of basin becomes junior in priority to every right granted in the basin of origin before the transfer application was filed. In a shortage the transferred water is cut first.

The sharpest and least-used tool against a campus supplied by out-of-basin piping. Opposition to these projects is usually political — hearings, petitions, county votes — when the statute already imposes a structural penalty: the transferred supply sits at the back of the priority queue exactly when drought makes it matter. Worth checking wherever a project's water arrives from a reservoir in another basin.

Whether a state agency's issuance of upstream withdrawal permits can be the proximate cause of downstream harm to a listed species. The Fifth Circuit held it cannot on the record before it: with multiple natural, independent and unpredictable forces acting on the habitat, the causal chain from permit to death of the animals was not reasonably foreseeable, and the injunction against further permitting was an abuse of discretion.

Included as a *negative* reading, because knowing where a theory fails is worth as much as knowing where it works. ESA is the intuitive federal hook for a data center drying a river reach, and in the Fifth Circuit this is binding law that the hook will usually miss when the defendant is the state permitting authority. A §9 claim would need to run at the withdrawer directly, with a much shorter causal chain than 'permit issued, species died'.

Setting land aside for a federal purpose impliedly reserves enough water to serve that purpose. The priority date is the date the reservation was created, not the date use began — so reserved rights are typically senior to nearly every state-law right in the basin, and are unaffected by a state's later admission to the Union or its own allocation scheme.

Matters wherever a campus draws from a basin that also serves a reservation, which describes much of the arid Southwest. The exposure is not that a tribe sues the data center — it is that quantifying a senior reserved right shrinks the pool the state had been allocating, and junior users get cut in order. A supply contract resting on a junior state-law right is worth less than it appears in such a basin.

The Winters doctrine draws no line between surface water and groundwater. Where the United States reserved land in an arid region, it impliedly reserved the appurtenant water sources including groundwater; the tribe's not having historically pumped does not forfeit the right, and state water law is preempted where the two conflict.

Closes the escape route. A project can be told a basin's surface water is fully claimed and turn to wells instead — this says the senior federal right follows it underground. In arid basins with a reservation overlying or adjoining the aquifer, groundwater is not the unencumbered alternative it is usually treated as.

An environmental impact document approving a large project must identify and analyse the water supply for the project's full life, not merely its first phase. Deferring the analysis to a future regional planning document, or relying on a mitigation measure that would curtail development if the water never materializes, is procedurally inadequate — as is leaving unexplained discrepancies between claimed supply and regional estimates.

The most transferable procedural attack on a data-center approval, because it does not require proving hydrologic harm — only that the paperwork assumed water it could not demonstrate. Campus approvals routinely analyse phase one and defer the rest, which is the precise defect this reading names. It is also the doctrine that most often produces the outcome opponents actually want: the approval is vacated and redone, rather than damages awarded years later.

Where a challenger shows a state environmental review failed to disclose a project's water impacts adequately, courts have treated proceeding on that record as irreparable harm and granted interim relief rather than leaving the challenge to be resolved after construction.

The timing half of the supply-adequacy reading, and the reason state environmental review outranks most federal theories in practice for these projects: it can halt work while the dispute is live. Data-center construction moves faster than litigation, so a remedy that only arrives after the campus is built is not much of a remedy.

In rule-of-capture states a landowner may generally pump without liability for draining a neighbour. That immunity is not total: where the *manner* of withdrawal is negligent, willfully wasteful or malicious and proximately causes damage to another's land — the classic case being subsidence from closely spaced high-volume wells — the pumper is liable.

The opening in an otherwise closed door. A neighbour in a capture state cannot complain that a data center took the water, but can complain about how: well spacing, siting near a boundary, and pumping rates chosen despite engineering forecasts of subsidence. Coastal and alluvial settings are where this bites, because that is where measurable subsidence follows drawdown.

How far a statutory 'any person may sue' provision actually reaches. A citizens' group may sue over a withdrawal that injures its members' own recreational, aesthetic, economic or property interests, but not over harm to the environment at large — the injury must be concrete and particularized to identifiable members.

Explains a pattern visible across this dataset: why data-center opposition succeeds through zoning votes, moratoriums and records requests far more often than through lawsuits. A group organized around a watershed rather than around injured members can be dismissed before the merits. It also says which plaintiff wins — the household whose well failed, not the coalition, which is why the Newton County fact pattern is legally stronger than a general-harm campaign.

A prospective new water user holds no absolute right to service and need not be treated like established customers. Where a utility faces a genuine imbalance between supply and demand it has a rational basis for a moratorium on new connections, and a court will disturb that decision only for fraud, arbitrariness or caprice.

Runs the opposite way from how these fights are usually argued. The tracked conflicts assume the utility is under pressure to say yes; this reading says a utility facing real supply constraint is on strong legal ground saying no, and the data center has no entitlement to be served. Where a utility has already conceded on the record that it cannot serve the load without major upgrades — Bessemer being the clearest instance here — that admission is the rational basis a refusal would rest on.

Statements a company makes to the public about its environmental performance, where those statements are alleged to be materially misleading. The subject of the claim is the company's own published figures rather than any discharge, withdrawal or permit, and the remedy sought is typically corrective disclosure and an accounting.

The one reading in this registry that does not require a permit, a discharge or an aquifer. It reaches the water-positive pledges and efficiency percentages that operators publish — which is significant because those figures are the industry's main public answer on water, and until 2026 nothing in the tracker connected a claim to a legal consequence. Its evidentiary base is utility billing records obtained by FOIA, so it is unusually available to outsiders compared with the hydrologic proof most other readings demand.