Economics — regional cloud-consumer demand & the public benefits extended to it

A demand-side companion to HYDROLOGY.md. Unlike HYDROLOGY, this document is not watermark-generated — it is hand-assembled analysis over cited records. Every figure is tagged: [verified] (read from a committed extraction or cited record), [inference] (a labelled derivation), [assumption], or [open] (a question, not a finding). The spine is civil/land/hydrology; this is the second axis — what the campus consumes, and what the public gives it — not a claim about who benefits.

Localized baseline. For the quantitative ground beneath this — Allen County’s employment by industry, its export-orientation, and the employment trend — see the generated localized economic baseline (BLS QCEW), the place the ~50 promised jobs and the 75% abatement actually land.

1. The load — what the campus draws

The one hard, document-anchored magnitude is electrical:

A ~275 MW IT load is a large consumer — roughly the scale of a mid-size city’s electricity demand, sited on one corridor. The water consequence is already modeled in HYDROLOGY (net basin loss ≈ 24–30× the Ottawa 7Q10). This page is the power and tax-base half of that consumption story.

2. The public benefits extended to it

What the public side committed, from the county’s own production [verified: data/extracted/legal/prr-mandamus/bosc-prr-production-2026-06-05.response-index.yaml]:

3. The mismatch — benefit vs. jobs vs. consumption [inference]

Set the verified columns side by side:

• ~275 MW IT load and 3.1–3.84 MGD consumptive water, against
• ~50 permanent jobs and a 15-yr/75% abatement on a ~$500M build.

That is on the order of ~5–6 MW per job and a multi-MGD basin draw for a headcount a single big-box store would exceed. The economic argument the corpus substantiates is structural: the public subsidizes load and consumption, not employment — and does so for a counterparty it cannot name (the Delaware shell; see DOSSIER §2). [inference] This is the demand-side mirror of HYDROLOGY’s “burden already maxed” finding (the 1996 SSO consent decree, the $11.8M I/I backlog).

The MW-per-job figure is modeled, not transcribed here (issue #1665): it is the load_per_job line of the economics-scenarios feed, and the feed states it as a band — ~5.0–10.0 MW/job, with the ~5.5 above as its reference corner (the disclosed IT-load central over the agreement’s own stated headcount). The ~5–6 quoted here is that corner; the lean-ops end of the band is roughly twice it. Quote the band or quote the corner, but say which.

4. Why this load exists here — demand-side drivers [open]

These explain the incentive to site authorized cloud capacity in a low-cost, low-scrutiny jurisdiction. The magnitudes are now document-backed industry reference ranges — from the relator’s data appendix, with its cited sources — though whether each applies to this campus stays [inference]/[open].

These four are modeled, not asserted here (issue #1665). Each is an axis of the economics-scenarios feed, computed from the committed pooled priors (data/reference/datacenter-industry/priors.yaml) and carrying every published source behind its band. The numbers below are quoted from that feed; if the two ever disagree, the feed is right and this page is stale.

• Authorized-region premium. Government/sovereign cloud (GovCloud-class, FedRAMP / DoD IL2–IL6) runs ~20–30% above commercial (BCG: up to 30%; AWS GovCloud EC2/S3 examples) — a recurring premium per hour and per GB. That rewards building dedicated, hardened capacity. [verified: appendix §1] / application-to-campus [open]
• Tax-base forecasting risk. Ohio’s data-center sales-tax exemption (DCTE) is scored against an equipment-purchase forecast — but AI-class hardware breaks that forecast: GPU servers $200k–$515k, replaced on a short cycle, ~30–40% of cost annually in opex. The abated base may never materialize against the consumption. [verified: appendix §2] / fiscal outcome [open]
• Refresh / AI-rack cost curve. Rack power density jumps 5–15 kW → 40–140 kW (conventional → AI/GB200), with projections of 250–900 kW/rack by 2027 — i.e. MW/water per rack trend up, not flat, across the abatement window. [verified: appendix §2]
• Facility footprint. A single site is a community-scale draw: 25 MW (the Ohio tariff/amendment reference) to 100 MW–1 GW, WUE ~1.8–1.9 L/kWh, up to ~5M gal/day evaporative — and blowdown discharge ~20–40% of cooling water, the wastewater tie-in to the WWTP capacity in HYDROLOGY. [verified: appendix §3]

These drivers are the substance of the relator’s committee data appendix (reproduction; prepared but not submitted). The figures are industry reference ranges with cited sources — real, documented magnitudes — not facility-specific values for the Bistrozzi campus.

5. Document-backed vs. analysis — the discipline line

6. Consumer energy-price pressure — the demand spillover [inference]

The 2026-06-10 facility-design call asked to “bring in fuel costs at the consumer level due to macro pressures and data-center demand.” The watermark.economics.energy thread sizes that spillover against committed EIA consumer prices (watermark eia → data/reference/eia/): Ohio residential electricity (¢/kWh), residential natural gas ($/Mcf), and total state retail electricity sales.

The link is the facility’s first-class total facility_draw (§1 + the PUE model, issue #87 — IT load × PUE), not IT load alone. derive_demand_pressure persists this sensitivity to data/reference/eia/demand-pressure.yaml (per-site, facility-gated) and exposes it as the economics-demand-pressure bundle feed (issue #1105), so the frontend sources these figures rather than the docs hand-copying a console printout:

The demand share and households-equivalent are the robust, EIA-cited headline; the price-pressure band is a deliberately stylized screening sensitivity, not a forecast (retail price formation is far more complex than one coefficient, and the campus buys at wholesale/industrial rates, not the residential price shown). This is the consumer-cost mirror of the §3 “subsidizes load, not jobs” finding.

7. The scenario bands — where these numbers are computed

Everything in §3 and §4 that is not a straight record read is modeled, and since issue #1665 (epic #1659, cluster ME-F) it is modeled in code rather than in this page’s prose. watermark.economics.scenarios assembles it from three committed inputs and publishes the economics-scenarios bundle feed:

Run it with watermark scenarios. What the feed carries:

• profiles — the four what-if corners (building share × jobs), each priced for forgone property tax, un-abated tax kept, sales-tax exemption, net subsidy and per-job. These used to be a hardcoded array in the frontend and a table in the-economic-ledger.md; there is now one computation.
• lines — each ledger line as a band over those corners.
• load_per_job — the §3 MW-per-job ratio, as a band.
• withheld — the four figures the record does not fix, each naming the record that would collapse its band.
• constants — every modeling constant as a cited ProvenancedValue (including the effective millage, which announces itself as an [assumption]).
• axes — the §4 drivers, with every published source pooled into each band.

The discipline is enforced in the type system, not asked for in prose. A band refuses low == high, and every axis, line and withheld input refuses the verified tag and any confidence above low — so a scenario structurally cannot be published as an assertion. The feed is also instrument-gated: a watershed point with no abatement agreement on its record has no feed at all, and its report locks and asks for that agreement rather than being priced off Allen County’s mills.

Nothing on this page promotes a defense-intelligence thesis. Defense-ecosystem actors enter only as [open] context where the public record already names them (see COURSE §1.4); the load, the benefits, and the consumption are the findings. The GovCloud / defense-hardened scenario profile is a labeled counterfactual on two knobs — it prices what such a facility would cost the public and asserts nothing about what this one is. That question is open; see defense-nexus.md.