Propylene Glycol vs Ethylene Glycol for Hyperscale Data Centers: Toxicity, Liability, and Regulation

The glycol selection decision that once lived quietly in a commissioning specification has migrated, at hyperscale, to the desks of ESG directors, insurance underwriters, and general counsel. With single campuses now operating continuous fluid inventories of 50,000 gallons or more across chilled water loops, direct-to-chip (D2C) secondaries, and facility cooling infrastructure, the choice between propylene glycol (PG) and ethylene glycol (EG) is no longer an engineering footnote — it is a material liability decision. Both fluids work. Both deliver corrosion protection when properly inhibited. The gap that matters for hyperscale operators running millions of square feet across watershed-sensitive geographies is not thermal performance. It is the regulatory, environmental, and human-health exposure profile that follows every gallon of fluid from installation through inevitable spill. This article examines both chemistries across every dimension that matters to facility engineers, procurement teams, and risk managers: toxicity mechanisms, thermal specifications, spill liability, regulatory trajectory, total cost of ownership, and the practical migration pathway for operations already running EG.

The Toxicity Profile Head-to-Head

The first number engineers reach for when comparing glycol toxicity is the oral LD50 — the dose required to kill 50 percent of a rodent test population. For propylene glycol, that figure in rats is approximately 20,000 mg/kg body weight. For ethylene glycol, it is 4,700–4,825 mg/kg. The ratio is roughly 4:1 in PG's favor, placing PG in the same acute toxicity tier as table salt (LD50: 3,000 mg/kg) and well below the threshold of concern for incidental occupational exposure. Ethylene glycol, at its LD50, corresponds to an estimated lethal human dose of roughly 1.4 mL/kg body weight — approximately 100 mL for an average adult — a quantity achievable through oral ingestion of a moderately contaminated beverage.

Acute LD50, however, is a blunt instrument. The more operationally significant difference between the two fluids is the metabolic pathway each follows inside a mammalian body — specifically, what toxic intermediates accumulate during clearance, and what organ systems bear the consequences.

Propylene glycol is metabolized primarily through the same enzymatic machinery that handles normal carbohydrate intermediates. Alcohol dehydrogenase converts PG to D- and L-lactaldehyde; aldehyde dehydrogenase then oxidizes these to D- and L-lactic acid. Both lactic acid enantiomers are physiologically normal intermediates in glycolysis and gluconeogenesis. At exposure levels relevant to occupational contact — dermal absorption during maintenance, inhalation of aerosol from a pressurized line failure, or incidental ingestion — the metabolic load is unremarkable. The kidneys and liver handle clearance efficiently, with no accumulative metabolite of concern. Urinary excretion of unchanged PG accounts for the small remainder at higher doses. No nephrotoxic endpoint, no neurological sequelae, no crystal deposition pathway exists in PG metabolism at any realistic occupational dose. This biochemical benignity is why the U.S. Food and Drug Administration granted propylene glycol Generally Recognized as Safe (GRAS) status under 21 CFR 184.1666 for direct food contact — a designation that requires published evidence of safe ingestion across a range of doses, not merely acceptable LD50 values on paper.

Ethylene glycol metabolism takes a structurally similar starting point and arrives at a profoundly different destination. Alcohol dehydrogenase converts EG to glycolaldehyde, which aldehyde dehydrogenase oxidizes to glycolic acid. From glycolic acid, a major oxidative branch yields glyoxylic acid, and from glyoxylic acid, terminal oxidation yields oxalic acid. Oxalic acid is the metabolite responsible for EG's systemic toxicity. In the presence of physiological calcium concentrations, oxalate anion precipitates immediately as calcium oxalate monohydrate — needle-like crystals that deposit preferentially in the proximal tubular epithelium of the kidney, causing acute tubular necrosis. Secondary deposition occurs in the myocardium, cerebral vasculature, and pulmonary parenchyma. At sublethal but clinically significant doses, renal injury manifests within 24–72 hours of exposure. Antidotes exist — fomepizole (4-methylpyrazole) and, secondarily, ethanol act as competitive alcohol dehydrogenase inhibitors, blocking the conversion of EG to its toxic metabolites — but their availability in an occupational emergency context depends on early clinical recognition of EG exposure, which is not always straightforward. Ethylene glycol carries no GRAS designation and is not approved for any food, pharmaceutical, or leave-on personal care formulation in the United States.

For facility engineers, these biochemical distinctions have practical consequences in three domains: the occupational health protocol posted in the mechanical room differs materially between the two fluids; the medical management chain activated after a maintenance spill differs (EG exposures require monitoring for metabolic acidosis and renal function over 72 hours; PG exposures rarely require more than observation and supportive care); and the liability narrative if a contractor or visitor sustains a significant exposure on site differs fundamentally in severity and legal exposure.

Thermal Performance Reality Check

Any honest evaluation of these two fluids must start with EG's genuine thermal advantage. At 30°C and a 50/50 volume concentration, ethylene glycol solution delivers approximately 7–10% higher thermal conductivity than an equivalent propylene glycol solution. This is a real, measured, reproducible property rooted in EG's lower molecular weight (62 g/mol versus PG's 76 g/mol) and its tighter hydrogen-bonding network with water molecules. The advantage is not a rounding error and not a marketing artifact — it is the primary technical justification EG suppliers have cited for decades. It deserves a direct technical response rather than dismissal.

That response begins with a careful accounting of where thermal conductivity sits in the total system thermal resistance budget at GB200-class rack densities. In a direct-to-chip cooling loop serving a rack at 100–120 kW, the thermal resistance chain runs from silicon junction to coolant across die, thermal interface material, cold plate base metal, and fluid-side film. The convective resistance at the fluid-metal interface accounts for roughly 15–20% of total system thermal resistance in a well-designed cold plate operating at typical flow velocities. A 7–10% improvement in coolant thermal conductivity therefore translates to approximately 1–2% improvement in total junction-to-coolant thermal resistance — well within the design margin of any competent cold plate specification. At the facility chilled water level, serving cooling distribution units or dry coolers at 35–55°C differential, the arithmetic is even less favorable for EG's conductivity argument, because the fluid-side film resistance represents an even smaller fraction of the total thermal circuit.

The viscosity picture cuts against EG in one dimension that matters to system designers. At 25°C and 50/50 concentration, PG solution viscosity runs approximately 5.5–6.0 mPa·s compared to EG's 2.5–3.0 mPa·s — roughly double. This means PG loops require more pumping energy to achieve the same mass flow rate at a given pressure drop, and at low temperatures PG's viscosity rise is steeper than EG's. Neither of these constraints is operationally binding for D2C secondary loops operating at 18–24°C supply temperature. Engineers select pump sizes and pipe diameters based on design flow requirements; the incremental pump energy difference between equivalent PG and EG loops at typical operating temperatures is measurable in kilowatts against a facility consuming megawatts. The viscosity penalty only becomes a design-relevant constraint below approximately −5°C — addressed in the warning callout below.

Specific heat slightly favors PG. At 50/50 concentration, PG solution carries approximately 3.70–3.75 kJ/kg·K versus EG's 3.55–3.60 kJ/kg·K. This means a PG loop carries marginally more thermal energy per unit mass flow — a modest offset to the conductivity disadvantage when evaluating total heat transport capacity rather than film coefficient alone. For a facility loop sized on total thermal capacity rather than cold plate heat flux, this difference partially compensates for PG's conductivity shortfall in the total system heat removal calculation.

Freeze protection at common data center concentrations is effectively equivalent for the geographies where hyperscale clusters operate. At 30% volume concentration, both fluids protect to approximately −13 to −16°C. At 50% concentration, EG provides protection to around −36°C compared to PG's −33°C — a 3°C difference that is operationally irrelevant for any chilled water loop operating above −15°C ambient, which covers every major hyperscale cluster geography in the contiguous United States including PNW, upper Midwest, and Mountain West winter operating conditions.

The engineering summary for mainstream hyperscale applications: EG's 7–10% thermal conductivity advantage at 50/50 concentration is a real fluid property that is design-margin-trivial for D2C secondary loops and facility chilled water applications at current rack densities. The thermal argument for EG is legitimate and meaningful in specific high-flux specialty compute applications — addressed in Section 5 — but it does not automatically recommend EG for a 200 MW campus-scale deployment where pump sizing, pipe diameters, and facility loop design margins dominate total thermal system performance.

Spill Scenarios and the Liability Calculus

No maintenance protocol eliminates glycol releases from large-scale data center operations at scale. Flexible hose fatigue failures, quick-disconnect seal wear, pump mechanical seal failures, seismic events at code-minimum fixture specifications, and contractor procedural errors collectively guarantee that a campus operating 50,000 or more gallons of loop inventory will eventually experience a significant release event. The question is not whether a spill will occur — it is what happens in the 72 hours after the response team arrives. That question has very different answers depending on which glycol is in the pipe.

Consider a 55-gallon drum equivalent release — roughly 460 pounds of fluid — from a secondary D2C loop into a mechanical room floor drain that routes to a municipal storm sewer. With propylene glycol, the immediate response pathway is administratively straightforward: contain, recover, clean, and document internally. PG does not appear on the CERCLA/Superfund table of reportable quantities for emergency release notification. It is not listed under EPCRA Section 302–304 as an Extremely Hazardous Substance triggering immediate notification to the State Emergency Response Commission. Most state environmental agencies classify PG releases as nuisance spills — requiring internal incident documentation and standard spill kit deployment but not emergency notification to state authorities. Total direct response cost for a 55-gallon PG release contained within the facility: a few hundred dollars of absorbent material, three hours of maintenance labor, and a one-page internal incident report. The event does not leave the building.

With ethylene glycol, the same drain event triggers a substantially different administrative chain. EG is classified as a hazardous substance under Clean Water Act Section 311, and many state environmental agencies require notification to the state environmental department within 24 hours of a release that reaches or threatens surface water or stormwater infrastructure. In Virginia — home to the world's densest concentration of hyperscale data center capacity in Loudoun, Prince William, and Fairfax Counties — the Virginia Department of Environmental Quality (DEQ) maintains active watershed oversight of the Potomac, Rappahannock, and James River drainages under Chesapeake Bay Program obligations. A confirmed EG release into a storm sewer serving the Potomac watershed requires formal written notification to the DEQ, activation of the facility's spill response plan, potential third-party environmental assessment to determine whether any release reached the receiving water, and — depending on the receiving water's designated use classification — a formal closure letter from the agency before the incident can be administratively closed. Incident closure timelines typically run 8–12 weeks from notification. In Columbus, Ohio, where Amazon, Google, and Meta collectively operate over 5 million square feet of data center space in the Rickenbacker and Dublin corridors, the Scioto River serves as the municipal water supply intake for Columbus downstream. An EG release detected in Scioto basin stormwater and triggering the Ohio EPA release notification protocol carries the additional dimension of public communications management with a municipal water utility whose customer relations department will ask pointed questions. Cedar Rapids, Iowa — where Meta's campus sits adjacent to the Iowa River corridor — presents the same scenario under Iowa DNR jurisdiction. Rural Wisconsin sites near Fairwater, where shallow aquifer systems serve surrounding residential wells, face a distinct but equally consequential exposure: EG entering a high-water-table groundwater system creates persistent biological oxygen demand that can affect well water quality for months, generating regulatory attention and neighbor relations issues that do not arise from equivalent PG releases.

"Every time we walk a facility team through the regulatory exposure map for a Northern Virginia campus — watershed buffers, groundwater reporting thresholds, the DEQ notification chain — the conversation shifts. The 7 percent thermal conductivity advantage starts looking a lot smaller when you put it next to a 72-hour incident report, a state agency notification, and a remediation vendor on speed dial. The fluid in the pipe is not just a thermal engineering decision. It is a risk management decision that your EHS team, your insurer, and your legal department will be living with for the operational life of that facility."

The insurance and reputational calculus follows directly from the environmental exposure profile. Environmental liability underwriters have systematically repriced EG-containing data center facilities upward since approximately 2020, as EG spill claim frequency at hyperscale has increased with facility density and fluid inventory volumes. Insurers offering pollution liability coverage to large campus operators increasingly differentiate PG and EG in their actuarial models — not as a policy mandate but as a risk-adjusted pricing input. Operators who have documented a full migration to PG have received pollution liability premium reductions of 8–15% at renewal in several verifiable cases, because underwriters recognize that PG claims almost never escalate beyond the facility fence line and do not generate state agency notification events that extend the claim duration and documentation burden.

The public communications dimension of a large EG release near a sensitive water resource is increasingly material for operators with published ESG commitments and public sustainability reporting. A reportable EG discharge that reaches local news through routine DEQ public records — with accurate headlines about a chemical release near a municipal intake or watershed buffer — is a reputational problem that no retrospective communications management fully resolves. A PG release at equivalent volume generates a cleaning crew dispatch note in the maintenance log. That asymmetry is not a marketing claim. It is a regulatory and media exposure differential with direct dollar value that any risk management team can model.

Regulatory Tailwinds

Regulations governing cooling chemistry at large data centers are in the early stages of a one-way ratchet. Understanding the landscape as it exists today — and where the trajectory points — is minimum due diligence for any procurement team making a 10-to-20-year infrastructure chemistry decision. The directional conclusion is not ambiguous: no state or federal environmental agency has moved to relax glycol regulations in recent years. Several have tightened them. PG is on the favorable side of every regulatory trend line affecting large-scale industrial coolant use. EG is not.

At the federal level, ethylene glycol is subject to reporting under the Toxic Substances Control Act (TSCA) Chemical Data Reporting (CDR) rule. Entities manufacturing, processing, or importing EG above CDR thresholds must submit production volume data, worker exposure assessments, and environmental release estimates to the EPA on a four-year reporting cycle. While this is a reporting obligation rather than a prohibition, it creates a documentation burden and establishes EPA visibility into EG use patterns at large industrial facilities — visibility that has no equivalent for propylene glycol, which does not appear on the CDR priority chemical list. The EPA's ongoing TSCA Section 8(a) rulemaking agenda has signaled increasing focus on high-volume industrial chemicals with identified aquatic oxygen depletion and toxicity profiles. EG's biodegradation-driven biological oxygen demand (BOD) impact on receiving waters — where aerobic decomposition of a large release can create hypoxic conditions affecting aquatic populations — places it squarely within the category of chemicals receiving heightened regulatory scrutiny in the current rulemaking cycle.

Virginia has moved further than most states toward data-center-specific environmental guidance. The Virginia DEQ's updated stormwater management guidance for large impervious-surface industrial facilities explicitly addresses glycol-based coolant releases as a recognized spill risk category, recommending secondary containment design standards and written spill response plan documentation that goes beyond general industrial facility requirements. The guidance does not mandate propylene glycol, but it imposes notification and remediation documentation requirements for EG releases that PG releases do not trigger — creating a de facto administrative advantage for PG operations in every Virginia DEQ permit review and renewal cycle. Virginia's data center development permitting process, which has become more rigorous as Loudoun and Prince William counties address cumulative environmental impact, increasingly requires applicants to demonstrate secondary containment adequacy for loop fluid inventories above defined volume thresholds. Facilities using PG consistently report faster regulatory review cycles than equivalently configured EG facilities in Virginia jurisdictions, because the state agency environmental risk assessment is abbreviated when the primary loop fluid does not carry a hazardous substance designation under state water quality rules.

Ohio and Iowa, both significant data center growth markets, operate Clean Water Act Section 401 water quality certification processes for large facility construction permits that evaluate chemical storage and release risk as part of the project approval pathway. Facilities specifying PG have documented shorter permitting timelines — often 60–90 days faster — compared to EG facilities in both states, because the state agency aquatic risk review is abbreviated when the loop fluid lacks the aquatic toxicity trigger. For fast-moving hyperscale construction programs where permit delays have direct capital carrying cost consequences, this timeline difference is a quantifiable benefit of the PG specification.

The longer-term regulatory vector is shaped by two converging pressures beyond current rule sets. First, the EPA's intensifying focus on industrial facility stormwater management, driven partly by PFAS enforcement infrastructure development, is creating regulatory machinery for chemical-specific stormwater oversight that will extend over time to other facility chemicals including glycols. Once stormwater chemical reporting infrastructure is normalized at the federal level, EG's BOD and aquatic toxicity profile puts it in the category most likely to receive stricter release thresholds. Second, the proliferation of state-level environmental justice legislation specifically targeting industrial facilities in communities near waterways and in historically underserved areas will increase scrutiny of any chemical with a demonstrated aquatic impact profile. Several major hyperscale development corridors — including parts of Northern Virginia, the Columbus exurbs, and rural Midwest greenfield sites — intersect with community water resource concerns that will only intensify as data center density increases. PG is the future-proof chemistry choice on every regulatory trend line visible from 2026 forward.

When EG Still Wins

A technically credible analysis requires stating clearly where ethylene glycol remains the correct engineering choice. There are real applications where EG's thermal advantage is not design-margin-trivial, real geographies where the liability asymmetry discussed in Section 3 does not apply in the same magnitude, and real retrofit economics where migration cost outweighs regulatory optionality benefit. Operators in these situations should not feel compelled to switch on the strength of arguments that do not apply to their specific operational context.

The clearest legitimate case for EG is in ultra-high heat-flux direct-to-chip applications where cold plate thermal resistance is the primary constraint on achievable TDP — not the 100–120 kW per rack typical of current GB200 NVL deployments, but the bespoke liquid-cooled compute modules being developed for next-generation AI training clusters with per-GPU TDP targets above 1,200 W and per-rack densities above 400 kW. Cold plate designers operating at the boundary of current copper microchannel fabrication capability can make a defensible argument that the 7–10% fluid conductivity advantage directly enables a higher sustainable heat flux at the same chip junction temperature. At this extreme end of the heat flux envelope, the fluid conductivity difference is not a marginal system-level correction factor — it is a material input to the cold plate thermal circuit that affects whether the thermal specification is achievable with a manufacturable cold plate geometry. Engineering teams developing hardware in this regime have correctly concluded that EG's thermal advantage justifies the added regulatory handling and liability overhead, because without it the chip thermal specification cannot be met.

Facility geography and site hydrology create a second legitimate EG exception. A data center sited in a semi-arid environment with stormwater routing to an evaporation pond rather than a navigable waterway, with no municipal water supply wells within the regulatory setback, and with state-level environmental guidance that treats glycol releases in the nuisance category rather than the hazardous discharge category, faces materially different regulatory exposure than a Loudoun County campus 400 meters from a Chesapeake Bay watershed drainage. Hyperscale facilities in Nevada's high desert, West Texas, eastern New Mexico, and the Permian Basin region often fall into this lower-regulatory-exposure category. For these sites, the liability case against EG is substantially weaker, and when combined with high-density workload requirements, the thermal case for EG can be decisive.

Retrofit economics occasionally favor EG continuation, independent of thermal and regulatory arguments. A facility commissioned on EG, operating with EG-formulated inhibitor packages, fitted with EG-compatible elastomers throughout its cooling infrastructure, and staffed with an O&M team trained on EG handling protocols has real sunk cost in its existing chemistry. If that facility is not proximate to sensitive water resources, is not facing tightening state guidance, is not experiencing insurance pressure, and has a remaining operational life of fewer than ten years, the financial case for a $300,000–500,000 flush-and-refill migration is not self-evident. The correct engineering decision in this scenario may be to maintain EG with enhanced secondary containment at high-risk connection points and a pre-positioned incident response capability, rather than migrating for regulatory optionality benefit that may not materialize within the facility's remaining service life.

EG wins in ultra-high-flux specialty compute, low-regulatory-risk geographies, and certain retrofit economics scenarios. It does not win for greenfield hyperscale campuses in watershed-sensitive geographies, for operators with published ESG commitments that will be evaluated by investors and enterprise customers, or for facilities where the liability asymmetry between the two fluids is a material board-level concern. The honest analysis supports both conclusions — and the geography of your site, the thermal profile of your workload, and the remaining life of your existing infrastructure determine which column you are in.

Cost Analysis: Per-Gallon Price vs. Total Cost of Ownership

The per-gallon cost comparison is straightforward and genuinely unfavorable to PG, and it should be put on the table directly. Industrial-grade ethylene glycol — the specification used for the majority of facility chilled water loops — trades in bulk truck quantities at approximately $0.60–0.90 per pound depending on market conditions and contract structure. USP-grade propylene glycol, the specification recommended for D2C secondary loops and any application where incidental contact with potable systems is possible, runs approximately $1.20–1.50 per pound in equivalent bulk quantities — a premium of 20–35% over industrial EG. Industrial-grade PG (technical grade) runs $0.90–1.15 per pound, narrowing the premium to 10–25% over industrial EG for facility chilled water applications where USP grade is not required. For a greenfield hyperscale campus requiring 60,000 gallons of initial glycol fill across all loops, the material cost difference is $120,000–$270,000 depending on grade specifications. That is a real number. Procurement teams are not wrong to put it on the spreadsheet and ask what they get for it.

The total cost of ownership analysis begins when you move past that first purchase order and account for the costs that do not appear on the material purchase invoice.

Environmental and pollution liability insurance for a large data center campus typically runs $0.12–0.25 per square foot annually, varying by geography, construction quality, fluid inventory volume, and chemical profile documentation. Operators who have executed a full transition from EG to PG and provided documentation of the change to their pollution liability insurers have received premium reductions of 8–15% at renewal in multiple verifiable cases — because underwriters recognize that PG claims almost never escalate to state agency notification events, third-party environmental assessment, or multi-week incident documentation timelines. The claims cost trajectory for a PG spill event is predictable and bounded; the EG equivalent is neither. For a 500,000-square-foot campus paying $125,000 annually in pollution liability premiums, a 12% reduction represents $15,000 per year and $300,000 over a 20-year facility life. That figure alone approaches the entire material cost premium for the initial PG fill at a 60,000-gallon campus inventory.

Incident response budget is the second TCO factor that rarely appears in direct material cost comparisons. Large campus operators with EG loops typically budget $50,000–150,000 per year in incident response reserve — encompassing environmental consultant retainers, sampling and analysis protocols, remediation vendor pre-qualification, state agency correspondence management, and legal review of notification submissions. PG operators carrying equivalent fluid inventory maintain significantly lower reserve requirements because the expected regulatory escalation path after a PG release is shorter, cheaper, and does not involve state agency notification timelines. Across a 20-year operational period, the difference in incident response provisioning commonly exceeds $500,000 for a campus-scale facility in a regulatory-sensitive geography.

Regulatory documentation overhead is the third factor. EG operations at hyperscale require TSCA CDR filings when applicable thresholds are met, state hazardous material inventory reporting in jurisdictions with community right-to-know requirements, and more detailed environmental permit documentation in water-quality-sensitive geographies. The labor and legal costs associated with this overhead — typically 40–80 hours of engineer and environmental counsel time annually for a large campus — are modest per event but compounding across a multi-facility portfolio.

The TCO conclusion for regulatory-exposed campuses is that PG's material cost premium — the number that leads every initial procurement comparison — is typically recovered within 3–5 years through insurance and incident response savings alone, and the 20-year TCO favors PG by a factor that makes the chemistry specification financially straightforward once the full model is on the table. The per-gallon price comparison is not the right frame for a decision with 20-year liability consequences.

Migration Playbook for Retrofit

The most common question from facility engineers who have concluded PG is the right long-term choice is not whether to migrate — it is how to execute the transition without generating the chemistry problems that damage infrastructure and void equipment warranties. A poorly executed glycol migration creates worse outcomes than maintaining the status quo: contaminated inhibitor packages, blended fluid chemistry, and incompatible elastomers can produce corrosion rates that exceed either pure chemistry by an order of magnitude, particularly in aluminum cold plates and brazed copper heat exchangers.

The governing rule for any glycol-to-glycol migration is absolute and non-negotiable: never blend propylene glycol and ethylene glycol chemistry in a live loop. The two fluids are individually water-miscible and can be physically mixed, but their respective inhibitor packages — whether organic acid technology (OAT), nitrite/molybdate hybrid, or azole-based formulations — are engineered specifically for the pH buffering chemistry and metal passivation behavior of their base fluid. An EG loop running a nitrite-based inhibitor package that receives a PG addition without complete inhibitor replacement will have an inhibitor system that is simultaneously over-concentrated in components incompatible with PG's buffering chemistry and deficient in the components PG systems require for aluminum protection. The resulting corrosion in the transition zone is not correctable by adding more inhibitor after the fact.

The migration protocol for a facility loop follows four sequential phases that together require 8–24 hours of loop downtime depending on loop volume and heat exchanger complexity.

Phase 1: Pre-drain sampling and chemistry audit. Before any fluid is moved, sample the existing EG loop at a minimum of three points: supply header, return header, and the most remote terminal unit in the loop. Measure pH, glycol concentration by refractometry, inhibitor reserve (nitrite or carboxylate level depending on current inhibitor package type), dissolved metals by ICP (copper, iron, aluminum as corrosion state indicators), and aerobic microbial plate count. This baseline establishes whether the existing system has pre-existing corrosion or microbiological conditions that must be addressed before new chemistry is introduced. A loop running degraded inhibitor or active microbial contamination should be treated before migration, not migrated into a contaminated state. Retain the baseline report — it becomes part of the equipment warranty compliance file.

Phase 2: Drain and primary water flush. Drain the loop to the maximum practical extent using existing lowest-point drains. A complete gravity drain leaves 8–15% of loop volume as residual film on pipe walls, dead legs, and heat exchanger internal passages — this residual cannot be avoided and must be managed by the subsequent flush steps. Follow the gravity drain with a deionized or softened water flush: fill the loop to full volume with flush water, circulate at normal operating flow velocity for 3–4 hours, then drain. The water flush removes the bulk of EG and its inhibitor chemistry. Avoid municipal tap water with chloride content above 25 ppm for the flush — chloride deposits on metal surfaces from the flush water will seed crevice corrosion in the new PG system. A second water fill-and-drain cycle is recommended for loops with complex dead leg geometry or brazed plate heat exchangers where the first flush cannot achieve full turnover velocity in all passages.

Phase 3: Secondary flush with dilute PG solution. For loops with brazed plate heat exchangers, microchannel cold plates, or manifold-fed distribution headers with low-velocity zones, a secondary flush with 10–15% PG solution pre-wets metal surfaces with PG-compatible chemistry and displaces any residual EG from dead zones where the water flush cannot achieve adequate velocity for complete removal. Circulate the dilute PG solution for 4–6 hours at operating flow, then drain. Measure the drain sample glycol chemistry — if EG content is detectable above 3%, repeat the dilute PG flush before proceeding to the final charge.

Phase 4: Charge with inhibited PG solution and verify. Fill the loop with PG at the target concentration — typically 30–50% by volume for freeze protection requirements in the applicable climate zone — pre-blended with the inhibitor package specified for PG chemistry. Do not transfer EG inhibitor chemistry into a PG charge; the inhibitor package must be matched to the new base fluid from a supplier that specifies PG compatibility. Circulate at operating conditions for 4 hours, then sample and verify: pH should be 7.5–8.5 for most OAT PG packages, glycol concentration should be within 2% of target, and inhibitor reserve should be within the supplier's target range. Metals (copper <0.1 ppm, iron <0.3 ppm, aluminum <0.2 ppm) in the 4-hour circulation sample confirm that no acute corrosion response has been triggered by the transition.

Facilities operating multiple independent loops — a D2C secondary loop and a facility chilled water loop, for example — can sequence migrations across separate maintenance windows to maintain partial cooling capacity during the transition. The recommended sequence is D2C secondary first (smaller volume, simpler geometry, faster execution) followed by facility chilled water (larger volume, more complex distribution). Sequencing reduces maximum cooling capacity reduction to the zone served by the loop being migrated, rather than the entire facility.

Post-migration monitoring should include quarterly chemistry checks for the first year: pH, glycol concentration, dissolved metals, and inhibitor reserve. EG contamination from residual deposits releases slowly over the first 60–90 days and will appear as a slight glycol concentration anomaly in the first post-migration chemistry check. Contamination below 3% EG in a PG loop does not require corrective action and will dilute further with normal top-up. Contamination above 5% warrants an additional dilute PG flush cycle before returning to standard quarterly monitoring intervals. Most well-executed four-phase migrations show EG contamination below 2% by day 30 and below detection threshold by day 90.

Propylene glycol for direct-to-chip and facility cooling applications — including USP grade for D2C secondaries, technical grade for facility chilled water, and inhibitor package specifications matched to aluminum, copper, and stainless steel infrastructure — is available in bulk tote and truck quantities through Alliance Chemical's cooling chemistry practice. Pre-migration chemistry sampling analysis and post-migration verification sampling are standard services for bulk PG customers, along with site-specific migration protocol development that accounts for your loop volume, heat exchanger inventory, climate zone freeze protection requirements, and operational scheduling constraints. Contact the cooling chemistry team to build a migration plan before the maintenance window is booked.

The thermal performance data in Section 2, the spill liability analysis in Section 3, the regulatory trajectory in Section 4, and the total cost modeling in Section 6 all point toward the same operational conclusion: for hyperscale facilities in watershed-sensitive geographies, with ESG reporting obligations and long operating life horizons, propylene glycol is the defensible chemistry choice — and the migration to get there is an 8–24 hour engineering exercise, not a capital program. The complete technical resource covering direct-to-chip inhibitor selection, aluminum cold plate corrosion protocols, deionized water specifications for single-phase immersion, and water treatment chemistry for facility cooling towers is available at the AI & Data Center Cooling Chemicals resource center. Each sub-pillar in that resource provides the same specification-level depth as this article, written for the engineers who will be responsible for the chemistry decision for the next two decades.