How to Control Coolant Additive Concentrations in Data Centers

A liquid cooling loop has four operating parameters. Three of them — temperature, flow, pressure — show up on every CDU display. The fourth one is chemistry, and it has no gauge on the front panel. Coolant chemistry fails silently: concentrations drift a few percent a month, nothing alarms, and then a cold plate plugs or a copper line pits through, eighteen months after the fluid went out of specification.

The chemistry itself is not exotic. The coolant in a direct-to-chip loop is a recipe with three ingredients — water, glycol, and an additive package — and “controlling concentration” means keeping each ingredient inside its own window while the system actively works to pull all of them out. Water is lost to seeps and maintenance drains. Inhibitors are consumed doing their job. Glycol slowly oxidizes into organic acids. Every top-up dilutes everything at once.

This article covers the target windows for each parameter, why they drift, how to measure them without a full lab, the dosing math for reinhibition and top-up, and the six concentration mistakes that cost operators the most. It ends with a complete annual chemistry review for a 5 MW liquid-cooled hall.

Know what you are controlling

The standard coolant for cold-plate loops is PG25 — roughly 25% propylene glycol by volume, blended with high-purity deionized water and a corrosion-inhibitor package [1][2]. Each of the three ingredients does a different job, and each fails differently when its concentration drifts.

The water does the thermal work. It carries the heat capacity and the thermal conductivity; the glycol mostly gets in its way. Water also determines the corrosion baseline: fill with tap water instead of deionized and you have introduced calcium, magnesium, chlorides and dissolved solids into a loop that ends in microchannels smaller than a human hair [16].

The glycol is a carrier, not a coolant. At 25%, propylene glycol depresses the freeze point to about −10°C (14°F) — enough to ride out a winter transport delay or a chilled-plant shutdown without burst piping [1][5]. Just as important, a glycol concentration above 20% makes the fluid biostatic: microbes that would colonize a plain-water loop grow slowly or not at all [17]. And the glycol is the carrier that keeps the inhibitor package dissolved and evenly distributed. Below about 20% PG, all three of these functions degrade together.

The additive package is what actually protects the metals. A typical package combines azole compounds (tolyltriazole or benzotriazole) that film copper and brass surfaces, organic-acid or molybdate or silicate chemistry that passivates aluminum and steel, and a buffer that holds pH in the mild-alkaline band where the film-formers work [8][9]. This package is consumed over time — filming inhibitors deplete as they do their job, glycol oxidation eats alkalinity, and every metal surface in the loop is drawing on the same reserve [8].

Table 1 puts the working windows in one place. Values are typical industry practice; the fluid manufacturer's datasheet governs wherever it is stricter.

Table 1 — Coolant chemistry windows for a PG25 direct-to-chip loop. Iron and copper are diagnostic indicators, not dose targets: they tell you corrosion is already happening.

Understand why concentrations drift

Closed loops lose very little water — a healthy system holds about 90% of its charge for a year [17]. But “very little” is not zero, and the losses that do occur are exactly the ones that damage chemistry:

• Make-up dilution. Every liter of make-up water added to replace a seep dilutes the glycol, the inhibitor and the biocide simultaneously. This is the most common drift mechanism, and the most underestimated, because each event looks tiny.
• Inhibitor consumption. Film-forming azoles deplete as they coat metal surfaces and heal damage; molybdate and OAT chemistry deplete passivating aluminum. The reserve alkalinity that buffers glycol oxidation acids is consumed doing its job [8].
• Glycol oxidation. Glycol slowly reacts with dissolved oxygen into organic acids. The acids lower pH and, worse, feed the bacteria you were trying to starve [16][17].
• System openings. Every rack insertion, pump swap or manifold repair drains some volume and introduces air, debris and biofilm seeds. IBM's maintenance spec requires biocide re-dosing within two hours of opening the loop to new racks [4].

The dilution effect looks harmless month by month and is anything but over a year. Suppose a 10,000-liter loop seeps 2% of its volume per month — well within normal — and the operator dutifully tops up with deionized water but no chemicals:

Table 2 — Dilution drift from 2%/month DI-only top-ups (computed: each top-up divides concentrations by 1.02). Nothing fails, nothing alarms — and by month twelve the fluid is at the edge of its biostatic window while inhibitor natural consumption has pushed the azole far below target.

The fix is not vigilance, it is arithmetic: make-up fluid must be PG25 premix, not water. Every top-up then restores all three ingredients in the correct ratio at once. Section on top-up discipline below shows the math.

Price both failure directions before you dose

Concentration errors are not symmetric, and operators who assume “more protection is safer” get this wrong. Both directions are expensive; they are just expensive differently.

Too little PG, too little inhibitor: the corrosion math is brutal and measured. In ASTM D1384 glassware tests, formulated PG25 held copper corrosion to about 0.03 mils per year; the same glycol without inhibitors corroded copper at 0.16 mpy — more than five times faster. Plain water sits in between at 0.08 mpy [5]. And once glycol concentration falls below ~20%, the fluid loses its biostatic character, and biofilm growth in warm low-flow zones — exactly where your manifolds dead-leg — becomes a matter of time, and a loop that grew a biofilm needs the full fouling remediation sequence, not just more biocide [6][17].

Too much PG: the freeze curve flattens while the viscosity penalty compounds. Going from 25% to 30% PG buys only 3°C more freeze protection while adding roughly 20–30% more viscosity at room temperature — a 30% blend runs 2.5–3.0 cP against water's 0.89 cP, and PG25 already runs 2.88 cSt at 20°C [5][9]. Every added point of glycol lowers specific heat and thermal conductivity, which means either colder supply temperatures or more flow — both cost pump energy. The curve below shows how little the last points of glycol buy:

Freeze protection per added point of glycol. An indoor secondary loop at 15–45°C supply has no credible freeze risk — the 25% point exists to cover transport and commissioning, not year-round weather [1][9].

Too much of the wrong inhibitor is the worst outcome of all. Silicate-based formulas are safe in a healthy alkaline loop; add one to a loop whose pH has already sagged below 7.5 and the silicate precipitates as amorphous silica gel — a gel deposit inside heat exchangers and cold-plate microchannels that cannot be reversed in service [8]. Similarly, stacking an OAT concentrate on top of a legacy nitrite program can form insoluble deposits. This is why every dosing event must be preceded by a measurement, not a hunch.

Measure what you dose

Every parameter in Table 1 has a field method that needs no laboratory. What separates a working program from a paper program is matching the right instrument to the right parameter and the right cadence:

Table 3 — The measurement stack. Field instruments catch drift between lab samples; the quarterly lab panel catches what field kits cannot see and gives you the numbers the dosing math needs.

Set the sampling cadence

Sampling frequency is where most programs fail — not because operators disagree about chemistry, but because nobody owns the calendar. The cadence below merges the IBM rack-loop spec, industry closed-loop practice and fluid-supplier programs [3][4][10]:

Table 4 — Sampling cadence. New loops run tighter than established ones: monthly field tests for the first year, relaxing to quarterly only after twelve clean months [3].

Do the dosing math

Two formulas cover every dosing decision you will make. Both are conservation statements — the chemical you add does not multiply, only the volume around it changes [15].

Formula 1 — the initial fill (C₁V₁ = C₂V₂): how much concentrate is needed to reach a target concentration in a known volume. A 2,000-liter loop to be filled at 25% PG from 95% inhibited concentrate:

The classic error here is treating V₂ as the water to add — pour 526 L of concentrate followed by 2,000 L of water and you finish at 26% in 2,526 liters, if the loop had somewhere to put the extra 526 L at all [15].

Formula 2 — the slug dose: how much product to add to raise a concentration by a target ppm in an existing loop:

Product density is the step people skip. Nitrite and molybdate inhibitors run 9.5–10.5 lb/gal (1.14–1.26 kg/L); glycol-based products near 8.6 lb/gal (1.03 kg/L). Assuming water density puts your dose 25% off [15].

Worked azole example. IBM's reference program holds azole at 40 ppm active, dosed as 120 ppm of a concentrate whose active fraction is therefore about one-third (40/120) [4]. In a 10,000-liter loop:

Azole top-up arithmetic at 33% active concentration. Always confirm the active fraction from the product data sheet — it varies by vendor [4].

Reserve alkalinity dosing follows the same logic with one extra input: the lab titration tells you the RA is at, say, 74% of target on a 60,000-gallon loop; the deficit (18% of the target charge) times the loop volume gives the concentrate quantity, added through the expansion tank in a two-to-four-hour job that costs $1,500–$4,000 in chemistry — versus a full fluid change-out an order of magnitude larger [8].

And one rule that is not arithmetic: after dosing, let the loop run at least three full turnovers before sampling. A sample pulled ten minutes after a slug dose will read low and tempt you into a second, unnecessary charge [15]. (If your sampling points and turnover times need a refresher, they are in our CDU maintenance guide.)

Top up with premix, never with water

The single most valuable discipline in coolant chemistry costs nothing: every liter of make-up goes in as ready-mixed PG25, not as water. Premixed fluid restores glycol, inhibitor and biocide in the correct ratio simultaneously; water top-ups dilute all three at once, exactly as Table 2 showed [13].

The premix rule turns make-up from a dilution event into a non-event. Log every make-up — the log is also how you notice that a loop suddenly “drinking” 5% a month has a leak you have not found [3][13].

And the water half of the premix is non-negotiable too: dilution water must be deionized grade at 5–10 µS/cm or better [4][16]. The most expensive documented chemistry mistake in this industry is a top-up from a yard hose. At one hyperscale facility, a single hard-water top-up introduced enough calcium and magnesium to scale cold plates across 40 racks — a week-long derating estimated at $2 million in lost compute. Microchannel blockage does not throw an error code; it silently throttles processors [16]. Tap water, softened water and standard RO water all fail the spec — softening exchanges calcium for sodium but leaves the conductivity and chlorides; RO alone can still carry ionic carryover [9][13].

Dose it safely

Closed loops are batch-dosed, not continuously fed — the chemical stays in the loop, so the tool is a bypass (pot) feeder for occasional charges, or direct addition through the expansion tank for compatible concentrates. Metering pumps earn their keep on open systems with continuous make-up, not on sealed loops [8][14].

A pot-feeder charge has a short, strict procedure [14]:

1. Isolate and relieve pressure before opening the pot — every time. The feeder sits across a pump differential and is a pressure vessel.
2. One chemical per charge. Never pre-mix inhibitor and biocide in the pot; add them in separate, sequential charges. Product compatibility inside a canister is not the same question as compatibility in a 10,000-liter loop.
3. Leave headspace — overfilling prevents the loop flow from entraining the concentrate.
4. Open the isolation valves and circulate for at least three full system turnovers before testing [15].
5. Verify the result with a test kit — the dose is not done until the residual confirms it. IBM's program, for example, requires azole confirmation with a test kit after every charge [4].

Run the biocide program on data, not on a calendar

Glycol helps you here — PG25 is biostatic, and a nitrite-free OAT formulation adds another layer of biological suppression [2]. But biostatic is not biocidal, and every system opening, dead leg and warm low-flow zone is an inoculation opportunity. Nitrifying bacteria metabolize nitrite inhibitors into ammonia (which attacks copper); sulfate-reducers generate hydrogen sulfide in anaerobic pockets behind biofilm; pseudomonas slimes glue the whole mess into microchannels [12][17].

The reference program is trigger-based [4][12]:

• Test quarterly (lab TVC) and monthly (dipslides) [4][12].
• Trigger level: 1,000 CFU/mL. Above it, dose a non-oxidizing biocide: glutaraldehyde at 100–200 ppm (strong against anaerobes — SRB, nitrate-reducers) or isothiazolinone at 100 ppm (strong against aerobic bacteria, fungi and algae) [4].
• When in doubt, use isothiazolinone — fungi in cold plates are the failure mode water testing misses entirely; the field symptom is falling cold-plate flow, not a positive water sample [4].
• Alternate chemistry between glutaraldehyde and isothiazolinone across treatments to prevent resistance [17].
• Re-dose within 2 hours of any system opening or rack addition [4].

Note what is absent from that list: a fixed monthly biocide shot. A healthy PG25 loop with clean dipslides needs no biocide at all; dosing on a calendar instead of a trigger just selects for resistant strains and adds organics that eventually feed the bacteria you were suppressing [11][12].

The six mistakes that cost the most

1. Topping up from a hose. Tap, softened or unconditioned water introduces hardness, chlorides and silica into microchannels. The 40-rack, $2M scale-up event above is the industry's cautionary tale; the anvilfield field guide lists it as mistake number one for good reason [13][16].
2. DI-only top-ups. Even pure water dilutes glycol, inhibitor and biocide simultaneously (Table 2). Make-up is premix or it is a slow leak in your chemistry [13].
3. Blind dosing. Adding concentrate without measuring pH, RA and metals first risks silicate gelation and inhibitor stacking — both irreversible in service [8].
4. Mixing inhibitor packages. Different vendors' PG25 fluids use different chemistries (OAT, silicate, nitrite, hybrid). Topping a Dober loop with a Clariant concentrate is an uncontrolled chemistry experiment on a loop full of copper and aluminum [1][7][8].
5. Filtering coarser than the cold plate. Full-flow filtration at 50 µm or finer, with a side-stream loop polishing toward single-digit microns. Particles that pass a coarse filter go straight into microchannels — and corrosion products from mistake #2 or #3 become the particles [13].
6. Applying facility-water chemistry to the TCS. The technology cooling loop is a closed, clean, low-volume system ending in a microchannel; the facility loop is an open, dirty, make-up-heavy system. Borrow the method (sample, trend, dose) from cooling-tower practice, never the numbers or the chemicals [13][18].

Worked example — annual chemistry review, 5 MW liquid-cooled hall

A 5 MW direct-to-chip hall: 125 racks at 40 kW, row-level CDUs, roughly 80 liters of TCS volume per rack including manifolds and CDU internals — call it 10,000 liters of PG25, commissioned 18 months ago. The quarterly lab panel comes back:

Panel results for the worked example. Two parameters out of band, one to watch, no biocide trigger.

The remediation math, in one block:

Complete remediation for the worked example: 211 L of concentrate restores the glycol window, 0.4 L of azole concentrate restores copper protection, total chemistry cost a few hundred dollars — against a loop charge that would cost tens of thousands to replace and weeks of downtime to flush.

Notice what the review did not do: no blind inhibitor slug — the azole deficit came from a measured titration; no biocide — the CFU count was below trigger; no fluid change-out — every parameter was recoverable in place. A managed loop with this discipline runs 5–7 years on its original charge in moderate-stress service, 4–5 years at high thermal density, with the eventual change-out as a planned capital event rather than an emergency [8].

Keep the log that makes next year easy

The chemistry itself is simple; the discipline is bookkeeping. The minimum viable program:

• Baseline at fill: archive an as-filled sample and the lab panel that goes with it. Every future deviation is measured against this, not against a datasheet [13].
• Monthly, ten minutes: refractometer PG%, pH, conductivity, dipslide incubation, make-up log, filter dP [3][7][12].
• Quarterly, one courier box: lab TVC, azole, reserve alkalinity, ICP-MS metals [4][8].
• Every event: opening, repair, make-up >1% of volume — immediate retest, biocide within 2 h of any opening [3][4].
• Every dose: what, how much, computed from what measurement, verified by what residual [4][15].

Facilities that outsource this — coolant-as-a-service programs with sensors and quarterly lab panels bundled — currently run $0.10–$0.30 per gallon per month in North America [8]. Facilities that keep it in-house keep the log themselves. What nobody gets away with is keeping no program at all: the loop does not alarm on its way out of spec, and the first symptom of depleted chemistry is thermal throttling on chips worth more than the building.

When the chemistry program is defined, the coolant spec becomes a procurement line item, and it should be quoted as one. Send us your loop volume, metallurgy, minimum site temperature and CDU architecture, and we will come back with fluid options, the monitoring kit and the dosing accessories to run it: request a quote.

Frequently asked questions

Can I just top up with distilled water from the grocery store?

Once, in an emergency, and then re-test immediately. Store-distilled water is usually clean enough (a few µS/cm), but a water-only top-up still dilutes glycol, inhibitor and biocide at once — so follow it with a chemistry check and reinhibition if needed. As a routine practice, use PG25 premix for every make-up; the premix discipline makes dilution impossible by construction [13].

How often should I actually test?

Monthly field tests (PG%, pH, conductivity, dipslide) for the first year; quarterly field tests after a clean year; the quarterly lab panel (TVC, azole, reserve alkalinity, metals) always; and an immediate retest after any opening or significant make-up [3][10]. Loops over 250 gallons should have continuous inline conductivity and pH monitoring with BMS alarms [4].

Is it safe to mix two vendors' PG25 fluids?

Do not plan to. Both are “PG25”, but the inhibitor packages differ — OAT, silicate, hybrid, nitrite-containing legacy chemistry — and some combinations form insoluble deposits or cancel each other's film-forming action. If you must change fluid vendors, plan a proper drain-flush-fill, not a top-up [8].

My dipslides are clean every month. Can I skip the quarterly lab panel?

No. Dipslides see planktonic bacteria; they say nothing about azole residual, reserve alkalinity, glycol oxidation products or the dissolved metals that indicate active corrosion. The quarterly panel is what tells you the inhibitors are holding before the corrosion counters show it [8][10].

Do immersion-cooled systems need any of this?

No dosing — dielectric fluids (synthetic hydrocarbons, fluorocarbons, esters) are single-component chemistries with no water to treat. Their equivalent discipline is periodic dielectric strength and moisture testing, plus particulate control. This article's program applies to water-based technology cooling loops: cold plates, CDUs, rear-door heat exchangers [9][16].

1. Dober, “PG 25 Coolant for Data Centers” — PG25 composition (~25% PG v/v + high-purity water + inhibitor package), −10°C freeze protection, tolyltriazole + silicate buffering chemistry, ASHRAE TC 9.9 and OCP framework, 1.5 LPM/kW cold-plate reference flow, and why higher glycol fractions trade heat transfer for freeze margin. dober.com
2. Open Compute Project, “SOLUTHERM OAT PG 25 Heat Transfer Fluid” product entry — OCP Accepted PG25 reference, the OCP contribution “Guidelines for Using Propylene Glycol Based Heat Transfer Fluids in Single Phase Cold Plate Based Liquid Cooled Racks”, modified ASTM D8040 corrosion test, nitrite-free OAT biostatic formulation, and elastomer compatibility (HNBR, NBR, VMQ, FKM, EPDM). opencompute.org
3. KVRM Engineering, “Liquid Cooling Infrastructure: CDU Sizing, Manifold Design and Facility Modifications” — TCS water-quality targets: pH 7.5–9.0, conductivity 500–2,000 µS/cm, total hardness <150 mg/L CaCO₃, Fe <0.3 mg/L, Cu <0.1 mg/L, TVC <1,000 CFU/mL, glycol 25–40%, monthly testing in year one then quarterly, and immediate retest after any breach or top-up. kvrm.in
4. IBM, “Water cooling system specification and requirements” (RDHx) — the reference dosing program: 40 ppm azole target dosed as 120 ppm Nalco 3DT-199 with test-kit confirmation, quarterly bacteria testing, >1,000 CFU/mL trigger for 100 ppm glutaraldehyde (H-550) / 200 ppm glutaraldehyde (73500) / 100 ppm isothiazolinone (77352), glutaraldehyde vs isothiazolinone spectra, biocide within 2 hours of rack addition, continuous 3D TRASAR monitoring above 250 gallons, annual azole recalibration, 50 µm inline filtration and DI-cartridge water polishing above 1 MΩ·cm. ibm.com
5. Caplinq, “Heat Transfer Fluids for Direct-to-Chip Cooling in Data Centers” — DOWFROST LC25 (25 vol% inhibited PG) and LC55 (55 vol%) specifications, PG concentration vs freeze point table (20%: −7.3°C through 55%: −40°C), ASTM D1384 corrosion data (LC25 ~0.03 mpy copper vs ~0.08 mpy water vs ~0.16 mpy uninhibited PG), PURAGUARD USP/EP base-glycol purity rationale. blog.caplinq.com
6. Schmidt + Haensch, “Propylene Glycol Concentration Monitoring in Data Center Cooling Systems” — why 25% PG must be controlled in both directions (biostasis below, viscosity/heat-capacity above), inline process refractometry with temperature compensation, and biofilm risk in cold-plate microchannels. schmidt-haensch.com
7. Clariant, “Antifrogen PG 25” product data — 25% v/v MPG with corrosion inhibitors, pH ~8.8, density 1.024 g/cm³ at 20°C, refractive index 1.365, freeze point −11°C (ASTM D1177), reserve alkalinity 1.4 mL 0.1M HCl (ASTM D1121), hybrid-OAT inhibitor class. clariant.com
8. Alliance Chemical, “Inhibitor Depletion Lifecycle and the Case for Coolant-as-a-Service” — depletion mechanisms, sensor layer (pH, conductivity, ORP into BMS), quarterly reserve alkalinity per ASTM D1121 and ICP-MS metals, the 60,000-gallon RA 74%→92% reinhibition example ($1,500–$4,000), never-blind-dose protocol, silicate dropout below pH 7.5, inhibitor stacking, 5–7 year managed fluid life (4–5 at high thermal density), CaaS pricing $0.10–$0.30/gal/month. alliancechemical.com
9. Alliance Chemical, “Direct-to-Chip vs Immersion Cooling: Fluid Types, Chemistry, and Selection Rules” — why PG over EG for server-side loops, DI-water requirement (~20 µS/cm threshold), pH 7.0–8.5 window and nonlinear corrosion outside it, 30% PG viscosity (2.5–3.0 cP vs 0.89 cP water), glycol 20–35% D2C spec, and why the glycol is an inhibitor carrier more than a freeze fluid. alliancechemical.com
10. ChemREADY, “Closed Loop Water Treatment Corrosion: Catch It Before It's Visible” — lagging indicators of depleted chemistry, pH <7.0 immediate-action flag, first-audit findings (pH 6.0–6.5 with reserve <20%), inhibitor reserve <30% replenishment rule, dissolved iron >2 ppm active-corrosion line, and testing cadence 2–4×/year by system age and metallurgy. getchemready.com
11. QualiChem, “Best Practices for Nitrite Closed Loop Programs” — why oxidizing biocides (hypochlorite, stabilized bromine/chlorine, chlorine dioxide) are not recommended for routine biological control in closed systems, glutaraldehyde biodegrading into a nutrient source and reacting with ammonia, isothiazolinone handling limits, nitrite-reducing bacteria converting inhibitor into ammonia that attacks yellow metals, and side-stream filtration as best practice. watertreatment.qualichem.com
12. Accepta, “Closed System Biocides” — non-oxidizing biocide spectra table (isothiazolinone 1.5%, DBNPA 20%, glutaraldehyde 24%, Bronopol, THPS) with pH/temperature stability and effectiveness vs Pseudomonas/SRB/NRB, quarterly dosing cadence with monthly dipslide verification, shock-dose tactic, and the BSRIA BG29 biofilm warning. accepta.com
13. Anvilfield, “Liquid cooling loop water treatment and chemistry field guide” — PG25 as default TCS coolant, never top up with tap/softened/RO water, full-flow 50 µm with side-stream single-digit polishing, pH 7–9 band, the mistakes list (DI-only dilution, no inhibitor, incompatible metals, coarse filtration, unflushed new loops, TCS/FWS confusion), and the field checklist from fill baseline to event retests. anvilfield.com
14. ERE Inc., “Chemical Pot Feeders: How Bypass Feeders Work and How to Size One” — pot feeder sizing (system volume × dose rate), one-chemical-per-charge rule, pressure relief before opening, headspace, pot feeder vs metering pump selection, and closed-loop batch dosing rationale. ereinc.com
15. Formula Expert, “Closed Loop Slug Dose Volume” and “Glycol Dilution” — V​p = C·V​s·ρ​w/ρ​p with the density trap (nitrite/molybdate products 9.5–10.5 lb/gal), the three-turnover rule before post-dose sampling, C​1V​1=C​2V​2 with V​2-as-final-volume worked example, and the “system already full” volume constraint. formula.expert
16. Machinery Lubrication India, “The Two Coolant Families Every Data Centre Operator Is Betting On” — the Chemical Mandate (DI/RO water, conductivity ≤5–10 µS/cm), the hyperscale hard-water top-up case (40 racks of cold plates clogged, week-long derating, ~$2M lost compute), glycol oxidation into organic acids, and the $40–45B cooling market context with liquid cooling at $15–20B. machinerylubricationindia.com
17. Chardon Labs, “Closed Loop Water Treatment Standards” — closed loops retaining ~90% of charge annually, glycol above 20% suppressing bacteria, nitrate-reducing bacteria converting nitrite inhibitor to ammonia, SRB hydrogen sulfide behind biofilm, glutaraldehyde + isothiazolinone alternation against resistance, and dead-leg risk. chardonlabs.com
18. Saint-Gobain ICS, “Inside the Direct-to-Chip Cooling Loop” — ASHRAE TC 9.9 W-class supply temperatures and the CDU as the FWS/TCS chemical boundary, OCP UQD and PG25 material-compatibility standards (2025), and the coolant spec items procurement should demand (inhibitor identity, biocide identity, pH band, conductivity limit, biological testing interval). ics.saint-gobain.com