Water supply & consumptive use
How much water the project consumes, against the receiving water's own cited low flows.
A real impact study states how much water Bowling Green Data Center (Project Accordion) consumes at buildout. It sets that figure against the North Branch Portage River's own cited design low flows, which are the drought floors a permit writer screens every discharge against. This chapter reports as much of that comparison as the record supports. It names what it cannot compute.
Cooling methodclosed-loop dry / air-cooled (“closed loop”) [reference] an operator claim — a claim is not an instrument
Claim vs record · the cooling-cycling reconciliation
Bowling Green Data Center (Project Accordion) claims closed-loop dry / air-cooled (“closed loop”) [reference] Reservation conflict pin kept: closed-loop dry / air-cooled (“closed loop”)
Reserved 0.6 MGD makeup contradicts the closed_loop_dry claim's ~0 MGD prediction — but a reservation is a ceiling, not a discharge/withdrawal instrument, so keep the archetype pin (no [verified] re-archetype) and sharpen the water lead (→ C2 records request #1688). It is also 12x the operator's OWN disclosed 0.05 MGD — a conflict between the two [reference] figures that stays unresolved here; the reservation classifies because it is independent of the claim's source, the self-report does not because it is not. The reservation survives as the finding precisely because the withdrawal/discharge instruments cannot reach this facility at all (makeup (purchased municipal supply — outside the A1 withdrawal registry)), so it is the only quantified figure on record that does not originate with the operator. The reserved figure is an upper-bound ceiling, NOT a headline consumptive. Neither corroborator is on record for Bowling Green Data Center (Project Accordion) (air permit + Tier II) — both are C2 records-request items, not read as confirming the claim.
Instrument reach[verified] The MAKEUP side is outside A1 by construction and the record shows it: the campus buys finished water from the Northwestern Water & Sewer District, which buys it wholesale from the City of Bowling Green, and the Ohio DNR WWFRP registers withdrawals FROM WATERS OF THE STATE (R.C. 1521.16, >100,000 gpd) — a purchased supply is the seller's withdrawal, not the buyer's. The Wood County registry (data/reference/ohio-water-withdrawal/wood.yaml, 36 facilities) accordingly carries NO registration under Meta, Liames LLC, 'Project Accordion', NWWSD, or any data-center name, while within the county the whole supply chain reduces to one registered withdrawal three transfers upstream: BOWLING GREEN CITY PWS (00251, 2,103.37 MG in 2024 ≈ 5.75 MGD, two Maumee intakes). That absence is READABLE rather than merely empty because the register is demonstrably live at this site: 'Apollo Power Generation Facility - TEMP' (03717) registered a 0.27 MGD surface intake on 2026-03-26 in the campus's own HUC-12 041000100703, with no annual report due yet. The DISCHARGE side is UNKNOWN rather than established-to-sewer: an ECHO CWA sweep of Wood County (FIPS 39173, 2026-08-01) returns 241 records including 50 effective individual NPDES permits, none of them the campus — every campus-linked record (PROJECT ACCORDION OHGC15219, APOLLO POWER GENERATION FACILITY OHGC17963, APOLLO LAYDOWN YARD OHGC18721, APOLLO NORTH PIPELINE OHGC19094, ACCORDION-DOWLING 138KV INTERCONNECT OHGC15929) sits under the CONSTRUCTION stormwater general permit, master OHC000000. So there is verifiably no facility-own outfall and no DMR, but the corpus does not establish where process/sanitary flow actually GOES — no sewer-use or pretreatment instrument is in hand — and the honest value for a route the record has not established is unknown, not sanitary_sewer. The municipal supply alone is enough to blind the account. NB every identifier above is an OHIO key (Wood County FIPS 39173; Ohio DNR registrations; Ohio EPA NPDES), and both instruments are Ohio-statutory, so neither can return a Bowling Green, KENTUCKY record; the KY collision reaches only the press-sourced ~50k/~600k figures, which is precisely where it must be watched.
The records that would test it [open]
- executed water & wastewater service agreement (the instrument text, not a summary)
- metered water-service use (actual makeup withdrawal vs the reserved ceiling)
- cooling-tower blowdown / low-volume-wastewater discharge record — a facility-own INDIVIDUAL NPDES permit + DMR on a direct-discharge path, or the industrial-user (IU) / pretreatment permit + sewer-use agreement on a sanitary-sewer route (OHD000001 was WITHDRAWN 2026-07-21; no general-permit coverage will ever exist)
- industrial pretreatment / indirect-discharge (IU) permit + sewer-use agreement
- facility air permit (PTI/PTIO) — cooling-tower emission-unit list + PM drift limits
- Tier II / EPCRA-312 chemical inventory — cooling-water treatment (biocide, scale / corrosion inhibitor)
Holder: Northwestern Water & Sewer District (the campus's water/sewer provider — the service agreement and the campus meter) AND the City of Bowling Green (the wholesale supplier — the August 2024 wholesale contract whose ceiling was raised to 1.5 MGD, and the WWFRP registrant of record). Neither alone holds both instruments: a request filed only with the City reaches the wholesale contract but not the campus's metered use, and one filed only with the District reaches the reverse; Ohio EPA (INDIVIDUAL NPDES — OHD000001 withdrawn 2026-07-21, no general permit to seek); Ohio EPA / regional air agency (DAPC); SERC / LEPC
Corroborators · secondary, never the outcome air permit · silent · Tier II chemistry · silentNeither corroborator is on record for Bowling Green Data Center (Project Accordion) (air permit + Tier II) — both are C2 records-request items, not read as confirming the claim.
[reference] company claim (NOT instrument-confirmed): Meta describes closed-loop, liquid-cooled with dry coolers — 'no operational water', with domestic/cleaning/fire use only. In tension with the NWWSD wholesaling BG water to Meta (contract ceiling 1.5 MGD, Aug 2024; conflicting ~50k vs ~600k GPD figures; a Meta-funded 2 MG tank + 16-in main) — that reconciliation is tracked at the water sub-issue #1439, not decided here. B5 (#1685) RAN that tension through the A3 cooling harness and the pin SURVIVES, but as a reservation_conflict rather than an untested claim: the district-linked ~600,000 gpd design commitment is a demand signal independent of Meta's own account of its cooling, and is disproportionate to 'no operational water'. It cannot re-archetype the pin — a negotiated ceiling is not a withdrawal or discharge instrument — and the harness confirmed there is no instrument to be had: the campus buys finished water (so the Ohio DNR withdrawal registry, which records withdrawals from waters of the state, carries no Meta/Liames/Accordion/NWWSD registration in Wood County at all), and a full ECHO sweep of the county finds every campus-linked record under the CONSTRUCTION stormwater general permit with no process outfall and no DMR. With OHD000001 withdrawn 2026-07-21, an INDIVIDUAL NPDES permit is the only remaining instrument on the DIRECT-DISCHARGE path — but that is not the only path, and here it may well be the wrong one: this campus's discharge route is not established, and a facility blowing down to a sanitary sewer files no DMR at all and is disclosed instead by the City/district industrial-pretreatment (IU) permit and sewer-use agreement, which is what the records ask actually seeks. See data/reference/oepa/cooling-reconciliation.yaml. Replace with a documented cooling design when an NPDES/water instrument lands.
- The harness recommends; it never mutates the pinned cooling model — re-archetyping is a reviewed edit with the instrument cited.
- A back-solved cycles-of-concentration is an [inference] bracket, never a headline scalar.
- A gap (no documented makeup or blowdown) is an [open] records-request lead — never read as 'confirmed dry'.
- A reservation ceiling (a will-serve / water-agreement figure) is not a discharge/withdrawal instrument — it keeps the archetype pin and is never collapsed into a headline consumptive figure.
- A reservation conflict survives a blind route: a negotiated ceiling is not something the withdrawal or discharge instruments could ever have metered, so blinding them cannot erase it — and where the operator's own figure conflicts with that reservation, the two are separated by provenance and not by size, with the conflict reported rather than resolved.
- An operator self-report lands on its own disclosed_* slot, never on documented_*, and cannot upgrade the claim's source.
- An instrument that cannot reach a facility returns an absence of jurisdiction, not a measurement — a municipally-supplied, sewer-discharging campus reads ~0 in the withdrawal registry and the discharge record by construction, and that ~0 never corroborates a claim.
- A documented withdrawal that is not the cooling account (construction-phase water) is kept on its own slot, and a prediction the harness could not derive is shown as refused, never as zero.
- Where the facility is invisible to the withdrawal registry but its municipal supplier is not, the system's reported withdrawal is the supplier's account — a total across every customer, carried as the denominator the claim has to be read against, never as the facility's own water.
- The corroborators (air-permit PM, Tier II chemistry) are secondary — recorded and reconciled against the claim, never the sole basis for a re-archetype and never changing the outcome.
data/reference/oepa/cooling-reconciliation.yaml — regenerate: watermark cooling-reconcile --write · reconciled as of 2026-07-21
baseline
Current municipal loop, no data-center cooling draw.
buildout
Data-center campus cooling draw (closed_loop_dry) on the municipal supply.
- The draw is set against the receiving water's cited design low flow as a worst-case, basin-scale bound — a screening comparison, not a withdrawal claim.
Bowling Green sits on a divide. It drinks the Maumee, from an intake at Waterville into a 170-million-gallon reservoir. It discharges to the Portage, into the North Branch by way of Poe Ditch. It has built a municipal identity on staying out of Toledo’s regional system and wholesaling water outward to the communities around it.
Find the denominator first
The Water Pollution Control plant is
designed for 10 MGD — 15.47 cfsrec . The regulatory low flow at the outfall, the 7Q10, is
0.364 cfs. [verified]
The plant is about forty-two times its river. Every question about this receiving water begins there — and the discipline is to find that ratio before reading a single limit, while resisting the urge to reconcile a regulatory denominator against a differently-derived one just because you hold both.
Two refinements that are easy to blur and change what the number means. The low flow is computed
for Poe Ditch at River Mile 2.5, where this plant discharges — not for the North Branch Portage
it enters a short distance downstream, which is a larger reach with a larger denominator. And Poe
Ditch is designated a Limited Resource Water, yet Ohio EPA applies the North Branch’s Warmwater
Habitat criteria to this discharge, in its own words “to be protective of this higher quality
stream.” So the reach that sets the quantity and the reach that sets the standard are different
waters. The screen reads 0.024:1 chronic and 0.018:1 acute — both violation [verified].
The conflict the instruments cannot settle
Two figures circulate for the campus’s operational water. Meta’s own announced number is about
50,000 gallons a day. The water district’s linked design commitment is about 600,000. They
differ by twelve times, and both are attributed rather than preferred. [open]
Now watch what happens when you reach for the environmental record to settle it. Ohio’s withdrawal register meters withdrawals from waters of the state; the NPDES record covers discharges to them. This campus buys finished municipal water and holds no discharge permit of its own. Both instruments are structurally blind to it — they return approximately zero by construction, and a zero from a blind instrument corroborates nothing.
Before reading any record as evidence of absence, ask whether it was ever capable of seeing the
thing you are looking for. That is the difference between “there is no withdrawal” and “there is no
withdrawal of the kind this register records.” The two documents that would settle it are a
service agreement and a wholesale contract, held by two different bodies, neither of which holds
both. [open]
Two things that are true and get misread
The drinking-water side is clean and must be narrated as such: no violation, lead at zero across five rounds, a consumer confidence report that reports none. The stress here is supply expansion under drought, not tap quality — after the November 2025 flash drought, the city that sells water to its neighbours began studying an emergency interconnection to buy some from Toledo.
A numeric trap worth the name: the city’s public water system reports about 220.84 million
gallons of returns in 2024 — roughly 0.605 MGD, which looks exactly like the disputed
600,000 gpd. It is not. It is filter backwash, reported in a stable band since 2016, a decade
before the campus existed. [verified] When two numbers match, ask what each one counts before
letting them be the same number.
The record behind this chapter
What this chapter stands on: the records it reads, the inputs its modeled figures rest
on, and the reference data behind its baselines — the same pages the record screens
serve, not a second copy. A figure the record does not support stays [open] and links nothing.
- Bowling Green Water Pollution Control (WPC)Permits
- 2PD00009.fs.pdfOhio EPA
- Permits2 records