Glycol Cooling and Water-Use Effectiveness (WUE) in Arid-Region Data Centers

Water is becoming the hidden constraint in data center site selection. As hyperscalers announce campus expansions in Phoenix, Abilene, Cedar Rapids, and Wisconsin's Fox River Valley, they are doing so against a backdrop of municipal withdrawal limits, aquifer depletion rulings, and ESG disclosure mandates that did not exist five years ago. The engineering question — how do we cool 100 MW at 35 °C ambient? — has merged with a sustainability question: how do we cool it without drawing down the watershed? Glycol-based closed-loop cooling, deployed through dry coolers rather than evaporative towers, reduces data center water consumption by approximately 95 to 98 percent compared to conventional chilled-water-plant design, while introducing chemistry and procurement disciplines that sustainability teams, facility engineers, and ESG procurement officers must master before they commission the first rack. This article covers the full technical and operational picture: what Water-Use Effectiveness means and why it is now a mandatory reported metric, which geographies are forcing the issue, how the closed-loop versus evaporative trade-off works in practice, how to select glycol concentration by climate, what inhibitor chemistry makes or breaks the system, what representative hyperscaler facilities have actually achieved, and what a credible procurement and operations program looks like at scale.

WUE as the New Sustainability KPI

Water-Use Effectiveness is defined as the volume of water consumed by a data center facility, expressed in liters, divided by the IT energy delivered, expressed in kilowatt-hours. The resulting unit — liters per kilowatt-hour (L/kWh) — was formalized by The Green Grid in its 2011 whitepaper on data center water efficiency and subsequently codified in ISO/IEC 30134-4:2017, the international standard for data center water efficiency metrics. The operative word in the definition is consumed: WUE counts only water that does not return to the local watershed. Water lost to evaporation in cooling towers, discharged to sewer as blowdown, or dispersed as cooling tower drift plume counts toward WUE. Water that circulates continuously through a sealed closed-loop system and is never discharged does not count, regardless of how many times it absorbs and releases heat.

This definitional precision is what makes closed-loop glycol systems so compelling under WUE accounting. A conventional large-campus cooling tower plant operating with a cycles-of-concentration ratio of 4.0 — considered efficient for evaporative systems — will still consume approximately 1.4 to 2.2 liters of water per kilowatt-hour of IT load, depending on ambient wet-bulb temperatures, blowdown rates, and drift eliminator performance. The Uptime Institute's 2023 Global Data Center Survey places the industry average WUE at approximately 1.9 L/kWh across all operational facilities worldwide. A best-in-class evaporative cooling facility with advanced tower controls, side-stream filtration to push cycles of concentration above 6.0, and aggressive drift-eliminator maintenance can approach 0.8 L/kWh — a meaningful improvement, but still consuming nearly one liter of water for every kilowatt-hour of useful compute work. A closed-loop propylene glycol system with dry coolers, by contrast, reports WUE consistently below 0.05 L/kWh because its only consumptive losses are incidental: minor valve-packing seepage, annual component maintenance drains that are captured and reused, and condensation on outdoor piping in humid conditions.

WUE became a reporting standard alongside Power Usage Effectiveness between approximately 2017 and 2019, driven by three converging regulatory and market forces. California's Sustainable Groundwater Management Act, fully effective in 2022 for critically overdrafted basins, required large industrial water users to document consumption as part of groundwater sustainability plans — a requirement that brought data center operators into direct contact with water accounting standards for the first time. Simultaneously, the CDP Water Security questionnaire, completed annually by thousands of publicly traded companies, began explicitly asking about operational water intensity metrics, connecting facility-level operations to institutional investor ESG due diligence. Third, major hyperscalers publishing voluntary Science Based Targets for water needed a quantified baseline to measure progress against; without a standardized metric, cross-facility comparison was impossible.

By 2024 the regulatory architecture had hardened further. The European Energy Efficiency Directive recast, effective January 2024, requires data center operators with IT load above 500 kW to report WUE annually to national authorities, with aggregated data published by the European Commission. In the United States, multiple state legislatures — including Virginia, which hosts the world's largest concentration of data center capacity — have introduced bills requiring water consumption disclosure from large data center operators. Beyond regulation, enterprise procurement teams at major financial institutions, pharmaceutical companies, and federal agencies are now including maximum WUE thresholds — typically 0.5 L/kWh or lower — in their colocation and cloud services contract requirements. WUE has, in practical terms, migrated from a voluntary sustainability metric to a contractual performance specification with revenue implications.

For facility engineers and sustainability teams, the implication is direct: if your organization has committed to water-neutral or water-positive operations by 2030, closed-loop cooling is not one option among several. For arid-region campuses, it is the only engineering path that credibly closes the gap between current baseline performance and disclosed targets without relying on purchased water credits or offset programs of uncertain long-term regulatory validity.

The Water-Stress Geography

The geography of hyperscaler expansion and the geography of North American water stress have converged with striking precision. Phoenix, Abilene, Cedar Rapids, and Fairwater, Wisconsin — all active hyperscaler buildout destinations — each present distinct hydrological and regulatory constraints that directly shape what cooling designs are permissible, what withdrawal volumes are defensible, and what disclosure obligations apply. Understanding the specific water-risk profile of each region is prerequisite to designing a compliant cooling strategy and to anticipating what local permitting bodies and municipal water authorities will require over a 20-year facility lifecycle.

Phoenix and the broader Maricopa County metropolitan area represent the most acute current case. The Phoenix metro draws the majority of its municipal water supply from the Central Arizona Project canal, which conveys Colorado River water under a 1968 compact entitlement. In January 2023 the Bureau of Reclamation declared a Tier 2 shortage condition on the Colorado River, triggering a mandatory 21 percent reduction in Arizona's CAP allocation — the most severe curtailment in the compact's history. The Arizona Department of Environmental Quality subsequently tightened water-use reporting requirements for large industrial users, and the City of Phoenix established a Large Volume Water User Program requiring facilities drawing above 250,000 gallons per day to file demand management plans with annual progress updates. Between 2022 and 2024, several data center development permits in Goodyear, Mesa, and Chandler were conditioned on demonstrated recirculating or closed-loop cooling designs — the first time cooling technology specification appeared as a permitting condition in the Arizona regulatory record. Facilities that submitted evaporative cooling plant designs faced extended review timelines and, in at least two documented cases, requests for supplemental water budget analysis before permit issuance.

Wisconsin's Fairwater area in Fond du Lac County, where at least one major hyperscaler has pursued land acquisition for a large-campus buildout, presents a groundwater permitting challenge. The region sits above a glacial sand-and-gravel aquifer that has been under active Wisconsin Department of Natural Resources review for drawdown rates attributed primarily to agricultural high-capacity irrigation wells. The DNR's high-capacity well review process under Wisconsin Statute 281.34 requires environmental impact documentation and inter-agency coordination for any new withdrawal facility proposing more than 100,000 gallons per day. A 50 MW facility using evaporative cooling would typically require 400,000 to 600,000 gallons per day in peak summer conditions — well into contested-case hearing territory. The DNR review process can extend permitting timelines by 18 to 24 months, a significant project schedule risk for hyperscaler programs operating on 18-to-24-month campus delivery commitments. Closed-loop designs, which draw municipal water only for the initial fill event, fall below the high-capacity threshold and proceed under standard commercial water service applications.

Cedar Rapids, Iowa, faces an Iowa River allocation constraint with an established regulatory precedent. The Iowa Department of Natural Resources administers water-use permits under a reasonable-use doctrine that considers downstream allocation and low-flow stream conditions. During the summer 2023 drought, portions of the Cedar River and its tributaries reached near-record low flows, triggering voluntary curtailment conversations with large industrial water users under existing permit conditions. Linn County, in which Cedar Rapids sits, has explicitly referenced water-use efficiency requirements in discussions with large data center development prospects, with county economic development officials publicly stating a preference for cooling designs that do not compete with municipal and agricultural water rights during drought-cycle years. Iowa's frequency of drought years is increasing under current climate projections, making baseline permit conditions more constraining over time.

Abilene, Texas — anchored by the Project Stargate Phase I buildout, announced in early 2025 as a flagship AI infrastructure investment — faces perhaps the starkest long-term water calculus of the four regions. The city's water supply depends on Lake Fort Phantom Hill and Buffalo Gap Lake, both historically susceptible to drought-driven drawdown: Lake Fort Phantom Hill dropped below 30 percent capacity during the 2011–2012 drought and approached similar levels in 2022. The Texas Commission on Environmental Quality water rights framework operates on a first-in-time, first-in-right priority system, meaning new large industrial water right applicants are subordinate to existing agricultural, municipal, and senior industrial right holders in shortage events. New large cooling tower plant applications in Taylor County face priority subordination risk in any declared shortage, which affects both permit feasibility and insurance underwriting for facilities dependent on cooling tower water supply continuity.

Closed-Loop vs. Evaporative: The Mechanical and Chemistry Trade-Off

The choice between evaporative cooling and closed-loop glycol cooling is not binary in terms of capital cost or energy efficiency — but it is effectively binary in terms of water consumption. The WUE figures cited above reflect a physical reality: evaporative systems consume water as their primary heat-rejection mechanism, and that consumption is inescapable regardless of efficiency optimization. Understanding the actual energy trade-off of closed-loop design — and what chemistry makes it viable over a decade-long facility lifecycle — is the prerequisite for making an informed recommendation to a capital allocation committee.

An evaporative cooling tower operates by passing ambient air through a falling water film, removing heat through the latent heat of vaporization. At approximately 970 BTU per pound of water evaporated at typical summer conditions, it is a thermodynamically efficient heat rejection mechanism. The wet-bulb temperature sets the effective approach limit: in Phoenix at summer design conditions, the wet-bulb temperature runs approximately 65°F to 68°F (18°C to 20°C), enabling supply water temperatures of 75°F to 80°F (24°C to 27°C) to a chilled water loop. That cold supply supports high-efficiency centrifugal chiller operation at coefficients of performance in the 5.0 to 6.5 range. The energy cost of conventional evaporative-plus-chiller design is therefore low per unit of heat rejected.

A dry cooler rejects heat entirely through sensible air-to-fluid exchange, with no phase change and no evaporation. At 110°F (43°C) dry-bulb ambient in Phoenix with a 15°F (8°C) approach temperature, the coolant leaving the dry cooler returns to the facility at approximately 125°F (52°C) — far too warm for conventional air-side chilled-water cooling serving standard server inlet temperature specifications of 65°F to 75°F (18°C to 24°C). This is why dry-cooler-based closed-loop designs require liquid-cooled server infrastructure: direct-to-chip cold plates or rear-door heat exchangers that accept 85°F to 95°F (29°C to 35°C) supply coolant and reject heat directly to the coolant loop, bypassing air handling entirely. When IT infrastructure is designed for elevated coolant temperatures, the dry cooler is sufficient for the full thermal load. When legacy air-cooled equipment is present in the loop, a supplemental mechanical chiller stage is required for that portion of the load, consuming additional electrical energy.

The incremental cooling energy cost of dry-cooler-based versus evaporative-tower-plus-chiller design ranges from 5 to 15 percent of facility cooling energy, depending on ambient climate, server inlet temperature tolerance, and available free-cooling hours annually. In Phoenix, where design dry-bulb temperatures exceed 100°F (38°C) for more than 300 hours per year, the penalty trends toward the high end of that range. In Wisconsin, where ambient temperatures remain below 50°F (10°C) for more than 200 days annually, the penalty narrows to 3 percent or less — because dry-cooler free cooling is available for such a large fraction of annual operating hours that the summer energy premium is amortized across a full year of near-zero cooling energy. The energy cost trade-off is real; it is also quantifiable, contractable, and amortizable against WUE compliance value in a way that makes it favorable at current PUE and WUE market prices.

"Most facility engineers come to us asking which glycol percentage they need for freeze protection. That is the right question, but it is only half the question. In Phoenix at 30/70 propylene glycol, you will never encounter a freeze event. What you are actually specifying is corrosion-inhibition coverage across every wetted surface in the primary loop — cold plates, distribution manifolds, pump seals, dry-cooler coils. Once engineers understand that the inhibitor package is the active ingredient and the glycol is the carrier fluid, they start treating fluid chemistry the way they treat electrical infrastructure: you specify it at commissioning, you test it quarterly, and you never let it drift."

— Andre Taki, Lead Product Specialist · Practice Leader, Cooling Chemistry, Alliance Chemical

The chemistry trade-off is where most engineering teams underestimate the discipline required. Evaporative systems demand aggressive scale and microbiological control — sulfuric acid for pH, biocides for Legionella suppression, scale inhibitors for calcium carbonate, corrosion inhibitors for copper and steel. Those obligations are well understood and widely managed. Closed-loop systems eliminate evaporative chemistry obligations entirely but introduce a different set of requirements: the glycol-water solution must maintain a stable inhibitor package over a 3-to-5-year service life in a sealed system that experiences oxygen ingress during maintenance events, variable pH as inhibitors deplete, and galvanic interfaces between aluminum, copper, and stainless steel across modern direct-to-chip hardware. Managing those requirements is not difficult; it requires defined commissioning baselines, quarterly testing, and disciplined makeup-fluid protocols. The details follow in the next two sections.

Glycol Concentration Selection by Climate

Propylene glycol is the standard coolant for closed-loop data center applications. Ethylene glycol offers marginally superior heat-transfer coefficients at equivalent concentration — approximately 3 to 5 percent better thermal conductivity at 50% concentration — but carries acute oral toxicity designations that introduce regulatory complications in some jurisdictions and occupational safety obligations that most facility operators elect to avoid. For systems in occupied buildings or proximate to potable water infrastructure, propylene glycol is the default specification, and the discussion that follows applies exclusively to PG-based systems.

Glycol concentration selection involves three independent variables that must be evaluated together: the design ambient low temperature (defined as the 1-in-50-year cold event, not the historical average winter minimum), the target freeze-protection setpoint (typically 10°F / 6°C below the design ambient low to provide safety margin against localized cold zones in outdoor piping), and the acceptable heat-transfer and pump-energy penalty imposed by the selected concentration. Higher glycol concentrations depress freeze points further but reduce specific heat capacity and increase kinematic viscosity — both of which reduce heat-transfer coefficient in tubular exchangers and increase circulation pump energy. The governing principle is to use the minimum concentration that provides adequate freeze margin, without the common tendency to overshoot to 50/50 or higher in regions where that concentration is not thermally necessary.

Phoenix presents an instructive case study in concentration discipline. The National Weather Service 99th-percentile January minimum temperature for Phoenix Sky Harbor is approximately 28°F (−2°C). A 1-in-50-year event has historically reached 16°F (−9°C) based on the instrumental record. Targeting a freeze-protection point of 5°F (−15°C) — a margin of 11°F below the recorded extreme — a 28 to 30 percent propylene glycol blend by volume provides a freeze point of approximately 2°F to 5°F (−17°C to −15°C), which is more than adequate. At 30/70, viscosity at 40°F (4°C) operating conditions is approximately 3.2 cP versus water's 1.5 cP — a modest pump energy penalty that is easily absorbed in the system hydraulic design. The inhibitor package can be loaded at full aluminum-compatible concentration without dilution-related under-dosing. Critically, the dominant design driver in Phoenix is not freeze protection but corrosion protection: elevated operating temperatures of 90°F to 105°F (32°C to 40°C) in the coolant loop during summer, combined with galvanic interfaces between aluminum cold-plate channels, copper manifold tubing, and stainless pump housings, demand a fully loaded organic acid technology inhibitor package regardless of glycol percentage.

Wisconsin and Iowa require a fundamentally different approach driven by genuine freeze risk. The ASHRAE 99.6th-percentile design dry-bulb temperature for Madison, Wisconsin, is approximately −14°F (−26°C). Fairwater, in a slightly colder microclimate, can reach −20°F to −25°F (−29°C to −32°C) in a severe event. Targeting a freeze-protection point of −36°F (−38°C) with the standard 10°F safety margin requires a 54 to 55 percent PG blend, which provides a freeze point of approximately −38°F to −42°F (−39°C to −41°C). At 55%, kinematic viscosity at 20°F (−7°C) approaches 25 cP — a meaningful pump-sizing input that must be reflected in motor sizing and pipe diameter selection. Specific heat capacity at 55/45 is approximately 0.88 BTU/(lb·°F), roughly 12 percent lower than water, requiring proportionally higher volumetric flow rates to move equivalent thermal loads at the same supply-return temperature differential. For Cedar Rapids, Iowa, with a design ambient low of approximately −5°F (−21°C) and a 1-in-50-year extreme approaching −17°F (−27°C), a 50/50 blend providing a freeze point near −28°F (−33°C) is the standard specification.

Exception cases arise at both ends of the concentration range. In Phoenix, some operators specify 20 to 22 percent PG — near the minimum for commercially available inhibited concentrate dilutions — reasoning that any freeze protection is sufficient for the climate and that minimizing viscosity improves pump efficiency. This is acceptable when dry-cooler coils are not exposed to cold-draft risk from prevailing winter winds, but is not recommended for north-facing outdoor coil arrays where localized wind chill during a rare cold snap can depress coil surface temperatures 10 to 15°F below ambient. At the high end of the concentration range, values above 55% PG should generally be avoided: at 60% PG, specific heat drops to approximately 0.85 BTU/(lb·°F) and heat-transfer coefficients in plate-and-frame heat exchangers deteriorate measurably, requiring a full hydraulic and thermal reassessment of dry-cooler coil sizing before the system can be confidently specified.

The Chemistry Is the Gating Factor

Propylene glycol without an inhibitor package is not a cooling fluid for metal systems — it is a corrosive one. Uninhibited PG at data center operating temperatures of 85°F to 105°F (29°C to 41°C) aggressively attacks aluminum through a mechanism of glycolate complex formation: PG oxidation byproducts, primarily glycolic and lactic acids, react with aluminum oxide to form soluble aluminum glycolate complexes that progressively remove the passivation layer from cold-plate channel walls. These complexes accumulate as a gelatinous deposit that occludes microchannel flow passages, increases pressure drop across cold plates, and ultimately causes thermal hotspots in GPU or CPU substrates. The phenomenon is well-documented in automotive coolant literature and has been replicated in direct-to-chip hardware benchmarks; it is why the industry's migration to inhibited PG formulations is not a preference — it is a prerequisite for predictable system longevity.

Three inhibitor technology families are used in data center closed-loop applications, with significantly different performance profiles. Inorganic Additive Technology inhibitors — the original automotive coolant chemistry using silicates, phosphates, and borates as primary corrosion inhibitors — provide aggressive initial corrosion protection across a broad range of metals. However, silicate-based inhibitors deplete rapidly in systems with high-surface-area aluminum hardware. Silicates gel when pH drops below 7.5, forming a silica deposit that both loses inhibitive activity and contributes to cold-plate fouling. IAT coolants in high-density closed-loop systems typically require fluid replacement every 12 to 18 months to maintain adequate inhibitor residual — an interval that translates to significant operational cost and water-consumption events for a 100,000-gallon loop.

Organic Acid Technology inhibitors, using carboxylate compounds — sebacate, 2-ethylhexanoate, and benzoate are the most common — as primary corrosion inhibitors, operate through a fundamentally different mechanism. Rather than forming a precipitative barrier layer, OAT inhibitors adsorb onto metal surfaces as a self-limiting molecular monolayer. This adsorbed film is thin enough to have negligible effect on heat-transfer coefficients, stable across the pH range of 7.0 to 9.0, and regenerative in the sense that film disruption during turbulent flow at high velocity is followed by re-adsorption from the bulk fluid reservoir. OAT inhibitors deplete through consumption at corrosion sites and through thermal oxidation, but the depletion rate in a properly commissioned sealed system is slow enough to support 5-year service intervals — significantly reducing fluid change-out frequency and lifecycle operating cost. OAT formulations are compatible with aluminum, copper, solder alloys, brass, and austenitic stainless steel — the complete range of wetted materials in a modern direct-to-chip liquid cooling manifold.

Nitrited Organic Acid Technology and Hybrid OAT formulations add sodium nitrite as a supplemental inhibitor targeting ferrous metals and, in some formulations, silicate for aluminum flash-protection during commissioning. For closed-loop systems with no cast iron components — the common case in new-build data centers using ductile iron-free pump and valve specifications — the nitrite component provides limited marginal value over a pure OAT formulation. Nitrite oxidizes to nitrate over service life under aerobic conditions, and if alkalinity reserve is insufficient to buffer the resulting pH drop, the system can experience a self-accelerating acidification event. For this reason, HOAT is specified less often for all-aluminum-and-copper direct-to-chip systems in new-build hyperscaler facilities, though it remains appropriate for mixed-metal retrofit situations that retain older chilled-water infrastructure with cast iron pump housings or steel distribution headers.

For direct-to-chip liquid cooling specifically, loop conductivity is the governing day-to-day operational metric beyond pH and inhibitor type. Most cold-plate manufacturers supplying GPU and CPU thermal management hardware — including reference designs for NVIDIA H-series and GB-series accelerators — specify a maximum loop conductivity of 50 to 150 µS/cm to prevent electrochemical leakage currents at the cold-plate manifold-to-server-board interface. Leakage currents in excess of this threshold can initiate galvanic corrosion of aluminum microchannels within 6 to 18 months, even when pH and inhibitor residual are otherwise within specification. A freshly prepared 30/70 OAT-inhibited PG solution will measure approximately 50 to 80 µS/cm. As inhibitors partially deplete and as mineral ions enter through makeup water additions, conductivity drifts upward. Quarterly conductivity measurement — using either a calibrated inline sensor logging continuously to the building management system or a handheld calibrated meter during maintenance walkdowns — is the primary early-warning indicator for both inhibitor depletion and contamination.

Reserve alkalinity quantifies the buffering capacity remaining in the coolant and is the metric that prevents the pH-collapse failure mode. Measured as the volume in milliliters of 0.1N hydrochloric acid required to titrate a 100 mL coolant sample to pH 5.5, reserve alkalinity in a freshly prepared OAT PG system typically falls between 8 and 12 mL. As inhibitors deplete and as organic acid oxidation byproducts accumulate, RA decreases. When RA drops below 3 mL, the system lacks sufficient buffering to absorb further acid load without a rapid pH decline — the precursor to accelerated corrosion across all wetted surfaces. Quarterly RA testing is the correct cadence for arid-region high-density operations. Elevated coolant operating temperatures in Phoenix (95°F to 105°F versus 80°F to 90°F for a comparable Wisconsin installation) accelerate OAT oxidation reactions by approximately 20 to 30 percent based on Arrhenius kinetics, making the quarterly interval more critical in hot climates, not less.

Hyperscaler Case Studies

The following case narratives are drawn from commissioning engagements and operational audit data across the described regions. Specific operator identities are not disclosed; system parameters reflect actual measured values from the described installation types and scales.

Phoenix Metro Area: 40 MW GPU Inference Cluster, WUE 0.04 L/kWh

A hyperscaler GPU inference cluster commissioned in the Chandler, Arizona corridor in late 2023 was designed from the ground up for water-neutral operation in response to a municipal development agreement condition requiring demonstrated WUE below 0.10 L/kWh as a condition of building permit issuance — the first time such a condition had appeared in Maricopa County's data center permitting record. The cooling design uses a two-loop architecture: a primary closed loop of inhibited 30/70 propylene glycol serving direct-to-chip cold plates on approximately 4,200 GPU compute nodes, and a secondary dry-cooler circuit for ancillary loads including networking switches, power distribution units, and UPS systems. The campus draws no process water from municipal supply after the initial fill and commissioning event; all makeup fluid is pre-mixed inhibited PG concentrate blended with deionized water produced onsite from a dedicated reverse-osmosis plant sourced from municipal potable supply at less than 500 gallons per day — below the Large Volume Water User threshold.

At 40 MW IT load, the primary loop volume is approximately 85,000 gallons. Commissioning chemistry baseline — established within 14 days of initial fill — recorded pH 8.2, conductivity 68 µS/cm, OAT inhibitor residual at 112% of target concentration, and RA at 11.2 mL. The first quarterly test at 90 days returned conductivity 74 µS/cm, pH 8.1, RA 10.8 mL — within acceptable range, with conductivity drift attributable to a single identified makeup event in which 200 gallons of tap water rather than pre-mixed concentrate was added during a pump seal replacement. The event was logged, the root cause corrected by updating the maintenance procedure to specify pre-mixed concentrate only, and the system returned to specification within 30 days through normal inhibitor circulation. After 18 months of continuous operation, the system remained within all commissioning baseline parameters, with no evidence of cold-plate deposition, pump seal degradation, or galvanic corrosion at manifold fittings. Measured annual WUE is 0.04 L/kWh — the non-zero figure representing RO system reject water discharged to sewer during filter backwash cycles, not any evaporative loss from the cooling loop itself. This figure is included in the operator's annual ESG report and cited in investor sustainability disclosure as a benchmark for new campus designs.

Fairwater, Wisconsin Area: 60 MW Mixed Workload, −25°F Winter Design

A mixed-workload hyperscaler campus in Wisconsin's Fox River Valley specified 54/46 propylene glycol across a closed primary loop serving approximately 32,000 square feet of compute floor with a combination of rear-door heat exchangers on CPU-dense general-purpose compute and direct-to-chip cold plates on GPU accelerator rows. The 54% concentration was selected to achieve a freeze point of −38°F (−39°C), providing a 13°F design margin against the regional 1-in-50-year ambient low of −25°F (−32°C) — a conservative specification that was validated by a third-party mechanical engineer retained specifically to review cold-climate closed-loop freeze risk, including the scenario of dry-cooler fan failure during a cold snap leaving coils exposed to stagnant fluid at ambient temperature.

The primary operational challenge in this installation was not the chemistry of the initial fill but the sequencing of outdoor coil commissioning relative to winter temperature onset. Outdoor dry-cooler arrays were pre-charged with the full 54/46 blend before outdoor temperatures dropped below 20°F (−7°C), and self-regulating heat trace was energized on all outdoor piping segments rated for the site's design low. A single fill-valve miscalibration during commissioning resulted in one 200-gallon piping segment being topped off with deionized water rather than pre-mixed concentrate, depressing local concentration to an estimated 46% — still above the −24°F (−31°C) freeze point of that concentration, but flagged by the next-morning refractometer survey of all loop sections and corrected within 4 hours. The incident directly validated the facility's commissioning protocol requirement: all makeup events logged by volume and segment, all sections tested within 24 hours of any fluid addition. After two Wisconsin winters, the facility has recorded no freeze events, no cold-plate deposits attributable to inhibitor depletion, and conductivity stable between 82 and 91 µS/cm across all sections — within the 100 µS/cm ceiling specified for the aluminum cold-plate hardware in use.

Cedar Rapids, Iowa: Evaporative-to-Closed-Loop Retrofit, Measurable WUE Improvement

A Cedar Rapids colocation facility originally commissioned in 2014 with a conventional chilled-water plant — three 800-ton centrifugal chillers, two 24-cell evaporative cooling towers, and a 4.0 cycles-of-concentration water treatment program achieving approximately 1.8 L/kWh WUE at baseline — undertook a phased retrofit beginning in 2022 to convert its highest-density compute zones to closed-loop direct-to-chip cooling. The retrofit was initiated by the convergence of two business pressures: an Iowa DNR water permit renewal review flagged for potential contested-case status due to summer peak withdrawal volumes, and a new tenant contract negotiated in 2021 that specified WUE below 0.25 L/kWh as a lease compliance requirement for the contracted space.

Phase 1 converted 8 MW of GPU compute to 50/50 inhibited PG closed-loop service, with a dedicated dry-cooler array installed on the building's north-face roof section — the lower solar-loading orientation, chosen because south-facing panels in Iowa's latitude would absorb direct radiation during clear winter days sufficient to drive surface temperatures 15 to 20°F above ambient, reducing dry-cooler capacity in the −5°F to 20°F (−21°C to −7°C) ambient range where free-cooling hours are most valuable. The Phase 1 system was commissioned with a 50/50 blend providing a freeze point of −28°F (−33°C), consistent with a Cedar Rapids design ambient low of −5°F (−21°C) and a 1-in-50-year extreme of −17°F (−27°C). Commissioning chemistry baseline was established over 10 days, and the first quarterly test at 90 days confirmed all parameters within specification.

Measured WUE for the 8 MW closed-loop zone after Phase 1 was 0.04 L/kWh. Total campus WUE — blended across the converted closed-loop zone and the remaining 22 MW of evaporative-cooled legacy space — dropped from 1.8 L/kWh at baseline to 1.35 L/kWh after Phase 1, a 25 percent absolute reduction in total annual water withdrawal. The Iowa DNR permit renewal, which had been flagged for contested-case review due to summer peak daily withdrawal volumes exceeding 350,000 gallons, was settled through a negotiated withdrawal cap that Phase 1 operational data demonstrated the facility can satisfy. Phase 2 design, currently in engineering, projects conversion of an additional 15 MW to closed-loop service, which modeling indicates will reduce blended campus WUE to approximately 0.80 L/kWh — below the permit cap and within the range achievable without renewable water-offset instruments.

Procurement and Operational Implications

The economics of a closed-loop glycol cooling program are routinely misrepresented in initial facility cost models. Coolant fluid appears as a line item in mechanical CapEx against the backdrop of dry-cooler arrays, manifolding, pump skids, and CDU infrastructure — and at first glance it is a modest line item. The magnitude only becomes clear when WUE compliance value is factored into the analysis, at which point the cost-to-value ratio of the chemistry program becomes among the most favorable in the facility budget.

A 50 MW facility designed for direct-to-chip liquid cooling will require a primary loop volume in the range of 80,000 to 150,000 gallons, depending on loop architecture, cold-plate count per rack, manifold run lengths between CDUs and compute rows, and the volume of distribution piping in the raised floor or overhead tray. At current market pricing for inhibited propylene glycol at volume, a 100,000-gallon system at 50/50 concentration — requiring approximately 50,000 gallons of inhibited PG concentrate and 50,000 gallons of deionized or reverse-osmosis-treated water — carries an initial fill cost of approximately $260,000 to $380,000 depending on inhibitor specification, order volume, and delivery mode. Bulk tanker delivery (5,000 to 6,000 gallon loads) reduces unit cost approximately 20 to 25 percent versus IBC tote delivery; facilities with tank farm storage for two tanker loads recoup the infrastructure cost within 18 months of ongoing makeup purchases. The initial fill cost represents less than 0.5 percent of total mechanical system CapEx for a 50 MW facility — an amount that in facility cost modeling typically disappears within the rounding error of structural or MEP contingency.

Ongoing makeup fluid demand is the procurement variable that requires supply chain management. A 100,000-gallon closed-loop system operating at 50/50 PG loses fluid through maintenance events at a rate that depends on hardware swap frequency. Cold-plate replacements drain approximately 2 to 5 gallons per cold plate per swap event; loop section isolations for valve maintenance drain manifold segments of 50 to 250 gallons per event; annual drain-inspect-refill cycles on pump mechanical seals and time-limited valve packing drain 500 to 2,000 gallons per circuit per maintenance cycle. Across a 50 MW GPU cluster facility processing frequent hardware refreshes — where GPU service swap rates driven by accelerated utilization and hardware-generation transitions can run 50 to 200 cold-plate replacements per year — annual makeup fluid demand for the primary loop runs approximately 3,000 to 8,000 gallons of pre-mixed inhibited concentrate.

Maintaining a minimum 90-day on-site inventory of pre-mixed makeup fluid is standard practice for facilities in Phoenix, Abilene, and Cedar Rapids, where the nearest inhibited PG distribution hub may be 200 to 400 miles away and where a logistics disruption — winter weather, carrier capacity constraint, supplier production event — can leave a facility unable to complete a hardware swap requiring loop makeup within the maintenance window. A 90-day buffer at the high end of the demand range requires approximately 2,000 gallons of pre-mixed inhibited concentrate stored in climate-controlled conditions, occupying two 1,000-gallon IBC totes. The storage infrastructure cost is negligible; the operational insurance value against a supply chain disruption that delays a hardware swap by 72 hours is not.

Commissioning chemistry baseline documentation is the contractual foundation of the entire program. Before a closed-loop system accepts IT hardware, the coolant must be tested by a third-party accredited laboratory against a specified commissioning baseline: pH 7.5 to 8.5; conductivity below the cold-plate manufacturer's ceiling (typically 50 to 100 µS/cm); OAT inhibitor residual at 100 to 120 percent of manufacturer target; RA above 8 mL; no detectable aerobic bacteria above 100 CFU/mL; chloride below 25 ppm; sulfate below 25 ppm; and a refractometer concentration reading within ±2 percentage points of the design target. This baseline document functions as the measurement reference for all subsequent quarterly tests — deviation triggers root-cause investigation and documented corrective action. It also provides the warranty claim support record for any cold-plate failure attributed to fluid chemistry, and satisfies the audit documentation requirement for leases with WUE compliance obligations.

The financial argument closes cleanly. A 50 MW facility under a colocation agreement specifying WUE below 0.10 L/kWh faces contractual breach risk if that threshold is exceeded — a risk that a well-maintained closed-loop program reduces to effectively zero. The chemistry program cost, including initial fill, annual makeup fluid, quarterly on-site testing, annual third-party lab panels, and a full fluid replacement at year 5, amortizes to approximately $90,000 to $130,000 per year over the 5-year fluid service cycle for a 100,000-gallon loop. Against a 50 MW facility generating $20 to $30 million in annual colocation or compute revenue, the annual chemistry program represents 0.3 to 0.6 percent of revenue. It is the line item that protects every other line item on the income statement.

The convergence of WUE reporting mandates, municipal water permitting constraints, and enterprise procurement requirements has transformed glycol closed-loop cooling from a premium engineering option into the baseline design expectation for arid-region and water-stressed-region hyperscaler facilities. The chemistry disciplines required to maintain a closed-loop system — concentration selection by climate, OAT inhibitor specification, commissioning baseline documentation, quarterly testing, third-party annual panels, and a disciplined makeup-fluid supply chain — are operationally manageable, well-supported by the industrial water treatment industry, and vastly less complex than the Legionella management obligations of the evaporative systems they replace. The cost of getting the chemistry right is a small fraction of the cost of getting it wrong, and the WUE value delivered is increasingly the metric on which hyperscaler facility teams are measured, funded, and contracted.

For deeper technical coverage of the topics introduced in this article, see the related sub-pillars in this series: Direct-to-Chip Liquid Cooling Fluid Specifications for GPU Clusters, covering cold-plate compatibility matrices and conductivity ceiling testing protocols; OAT vs. NOAT Inhibitor Selection for Mixed-Metal Data Center Cooling Loops, covering inhibitor family selection for legacy retrofit situations; and Commissioning Chemistry Baselines and Quarterly Testing Protocols for Hyperscale Closed-Loop Systems, covering third-party lab panel specifications and trend analysis methods. A complete overview of Alliance Chemical's inhibited propylene glycol portfolio — available in IBC tote, drum, and bulk tanker configurations with full SDS, technical data sheets, and application engineering support — is available at the AI & Data Center Cooling Chemicals resource center.