Coolant Chemistry for Data Centers: Glycol, Inhibitors, and Sampling
A practical coolant program for direct to chip loops: what PG25 protects, which readings matter, how to sample the same way every time, and what to do when a number drifts.

On this page
- At a Glance: Common Coolant Choices
- Two Loops, Two Water Specifications
- What PG25 Protects
- What the Inhibitor Package Does and How It Fails
- Sample the Loop the Same Way Every Time
- Build a Log You Can Compare
- Interpret the Reading and Decide
- Fix the Cause, Not the Number
- Put the Loop on the Maintenance Calendar
A coolant loop is a chemistry experiment that runs for a decade. The fluid looks clean, the temperatures look normal, and then a cold plate starts running hot because the microchannels have collected a layer of corrosion product that nobody saw forming. By the time the chip throttles, the damage is inside a component that costs more than the loop maintenance budget for several years.
The good news is that the chemistry is measurable. Glycol concentration, pH, conductivity, chloride, corrosion products, and microbial counts tell you what the fluid is doing long before the thermal performance changes. This guide covers the two loops and their different specifications, what PG25 actually protects, how to sample without introducing error, and how to turn a reading into a decision.
At a Glance: Common Coolant Choices
| Water with inhibitors | PG25 | PG55 | Dielectric fluid | |
|---|---|---|---|---|
| Freeze protection | None | About -10 C | About -40 C | Varies by product |
| Heat capacity | Highest | About 4 percent lower than water | Lower still | Much lower |
| Biostatic above about 24 percent | No | Yes | Yes | Not applicable |
| Typical use | Large facility loops | Direct to chip TCS | Cold climate or shipping | Immersion |
| Main risk | Freezing, biological growth | Dilution, inhibitor depletion | Viscosity and pump energy | Material compatibility |
Two Loops, Two Water Specifications
The facility water system and the technology cooling system are separated by the CDU heat exchanger, and they are allowed to have different water quality. The facility side has larger passages and a robust exchanger, so it tolerates more chloride, more hardness, and more bacteria. The technology side feeds cold plates with microchannels measured in fractions of a millimeter, so its limits are much tighter.
The ASHRAE guidance makes the gap explicit. The facility loop is commonly specified around pH 7 to 9, chloride below 50 ppm, sulfate below 100 ppm, hardness below 200 ppm as calcium carbonate, and bacteria below 1,000 colony forming units per milliliter. The technology loop is tighter: pH 8.0 to 9.5, chloride below 5 ppm, sulfate below 10 ppm, hardness below 20 ppm, bacteria below 100 colony forming units per milliliter, and suspended solids below 3 ppm. Those technology loop values are written for a water based loop. An inhibited PG25 fluid follows the supplier specification instead, and its pH target is commonly a little wider, around 8 to 10.5 depending on the inhibitor family.
The conductivity row is the one that trips people up. The technology loop guidance includes a 0.2 to 20 microsiemens per centimeter range, which applies to a water based loop with a well maintained deionized fill. An inhibited PG25 fluid typically reads above 2,000 microsiemens per centimeter because the inhibitor package is conductive by design. Applying the water target to a glycol loop would send an operator chasing a number that cannot be reached, and the correct approach is to trend conductivity against the baseline for that specific fluid.
The ownership boundary matters too. The facility loop is usually the building operator’s responsibility, and the technology loop is usually shared between the CDU vendor, the IT vendor, and the operator. A water quality argument that does not name the loop is unresolvable, because each side is measuring against a different table.
What PG25 Protects
PG25 is an inhibited mixture of about 25 percent propylene glycol and 75 percent water, with the inhibitor package counted in the water share. The glycol does two jobs. It lowers the freeze point to roughly -10 C, which protects a loop during shipping, storage, and a winter outage, and above about 24 percent it keeps the fluid biostatic, so bacteria and fungi cannot multiply. The inhibitor package protects the mixed metals in the loop, which usually means copper cold plates and stainless steel piping.
The concentration range is tight for a reason. A fresh PG25 fill is commonly specified at 24.5 to 27.5 percent glycol by volume, and an in service fluid is kept within about 24 to 29 percent. Below 24 percent, the fluid loses its biostatic property and the inhibitor is diluted below its design strength. Above the range, viscosity rises, heat transfer falls, and pump energy increases without a corresponding benefit.
PG25 is not the only option. Deionized water with a corrosion inhibitor package has a higher heat capacity and lower viscosity, which means less flow and less pump energy for the same heat load. It also has no freeze protection and no inherent biostatic property, so it needs a biocide and a tighter microbial monitoring program. PG55 extends freeze protection to about -40 C and is used where a loop may be shipped or stored in a cold climate, at the cost of higher viscosity and lower heat capacity. Dielectric fluids belong to immersion systems and have their own compatibility and maintenance rules.
The heat capacity difference is small but real. Water is about 4.18 kilojoules per kilogram per kelvin, and PG25 is around 3.9 to 4.0, with a density slightly above water. The practical result is that a PG25 loop needs roughly 4 percent more flow than a water loop to hold the same temperature rise, and it has a higher pressure drop at low temperatures because the viscosity is higher.
What the Inhibitor Package Does and How It Fails
The inhibitor package is a mixture of corrosion inhibitors, pH buffers, and stabilizers. It works by forming a thin protective layer on the metal surfaces, and it is consumed over time. Every liter of fluid in the loop is a reservoir for that chemistry, and every addition of water, every air bubble, and every hot spot consumes some of it.
The failure mode is not usually a sudden drop. The pH drifts down as the glycol oxidizes into acidic compounds, the reserve alkalinity falls as the buffer is consumed, and the corrosion rate rises as the protective layer thins. Copper and iron are the usual early markers in the lab panel, and a loop that tests fine for two years can cross a threshold in one quarter if it starts ingesting air or gets topped off with hard water.
Three rules keep the chemistry inside its design envelope. First, do not mix fluids from different suppliers unless both suppliers confirm compatibility, because two inhibitor packages can precipitate each other. Second, do not add an unapproved inhibitor or biocide, because overdosing can foul the cold plates as effectively as underdosing can corrode them. Third, fix the cause of any dilution or contamination before adding fresh fluid, because topping off a leaking loop hides the leak and concentrates the damage.
Sample the Loop the Same Way Every Time
A sample is only useful if it is comparable to the last one. Take it from the same port, at the same operating condition, with the same procedure, and label it with the date, the loop, the operating temperature, the glycol reading, and who took it. A sample pulled from a stagnant dead leg tells you about the dead leg, not the loop.
Industry practice is to sample a hot, flowing closed loop, which usually means a sample valve on the return side while the CDU is running. Flush a small volume first to clear the port, fill the sample container to the top to limit air contact, and record the loop temperature at the moment of sampling. For a baseline, take the sample after the system has circulated long enough for the fluid to be fully mixed, not immediately after a fill.
The schedule that fits most operations is a baseline at fill, field checks every quarter, and a full laboratory panel every year. Increase the frequency during the first months after startup, after a large makeup addition, and after a leak repair. Industry guidance often lands on a 50 to 100 day interval for the field checks, which is close enough to quarterly that either is defensible as long as it is consistent.
Field checks cover appearance, pH, and glycol concentration. A digital refractometer with temperature compensation is the standard tool for glycol, and it should be checked against the supplier’s calibration fluid. Laboratory panels add anions such as chloride and sulfate, reserve alkalinity, elemental metals such as copper and iron, inhibitor concentration, and a microbial count. For an OAT based fluid, the lab needs the specific test for that inhibitor family, so tell the lab which product is in the loop.
Build a Log You Can Compare
A spreadsheet works, but a small text log is easier to keep next to the loop and easier to diff. The important part is a fixed column order that never changes: date, loop, pH, glycol percent, conductivity, chloride, iron, copper, appearance, and action. If the column order changes, the history becomes unreadable.
The script below appends one reading to a file and prints the change against the first line for that loop. It is deliberately simple, and it runs anywhere with awk.
cat > coolant-log.tsv <<'TSV'
date loop pH glycol_pct conductivity_us_cm chloride_ppm iron_ppm copper_ppm appearance
TSV
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
2026-11-11 R07-L1 8.6 25.4 2380 3.1 0.04 0.02 clear >> coolant-log.tsv
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
2026-12-09 R07-L1 8.4 24.8 2410 3.4 0.05 0.02 clear >> coolant-log.tsv
awk -F'\t' -v loop="R07-L1" '
NR == 1 { next }
$2 != loop { next }
first == "" { first = $3; firstGlycol = $4; firstCond = $5 }
last = $3; lastGlycol = $4; lastCond = $5
END {
printf "loop %s pH %s -> %s glycol %s -> %s conductivity %s -> %s\n",
loop, first, last, firstGlycol, lastGlycol, firstCond, lastCond
}
' coolant-log.tsv
loop R07-L1 pH 8.6 -> 8.4 glycol 25.4 -> 24.8 conductivity 2380 -> 2410
A note on units: record conductivity in microsiemens per centimeter and metals in parts per million or milligrams per liter, and write the unit in the header. Mixing units between labs is one of the most common reasons a trend looks alarming when nothing has changed.
The same log becomes the evidence for a warranty conversation. If a cold plate vendor asks how the loop was maintained, a dated history with sample results, filter changes, and makeup volumes is a much stronger answer than a statement that the loop was checked regularly.
Interpret the Reading and Decide
A single out of range value is a reason to resample, not a reason to drain the loop. A trend across three or four samples is a reason to act. The table below is the decision layer that turns numbers into work.
| Reading | Likely cause | First check | Action |
|---|---|---|---|
| Glycol below 24 percent | Dilution from makeup or a leak | Makeup volume, leak sensors | Find the source, add premixed PG25 |
| pH drifting down | Inhibitor depletion or glycol degradation | Reserve alkalinity, lab panel | Contact the fluid supplier, plan a correction |
| Conductivity rising | Makeup water, corrosion, contamination | Chloride and metals | Stop the contamination, do not dilute blindly |
| Chloride rising | Tap water or a failing exchanger | Makeup source, FWS pressure | Fix the source, replace affected fluid |
| Iron or copper rising | Active corrosion | pH, inhibitor, flow, aeration | Escalate to the supplier and inspect the loop |
| Cloudy or discolored fluid | Particulate or biological growth | Filter differential pressure, bacteria count | Change filters, test for microbes, flush if needed |
| Makeup volume rising | A slow leak | Loop walk, quick disconnects, CDU drain pan | Repair the leak before adding more fluid |
The hardest case is a loop that passes every individual limit but is drifting toward several of them at once. That is the point to call the fluid supplier with the history, not to wait for a thermal alarm. A supplier who made the fluid can tell you whether the drift is normal consumption, contamination, or the start of a chemistry problem, and they can recommend an approved correction instead of a generic additive.
Fix the Cause, Not the Number
Most coolant problems start outside the fluid. A loop that keeps needing makeup has a leak. A loop with rising chloride has a water source problem. A loop with rising iron has air ingress, low flow, or a corroding component that is shedding into the fluid. Adding fresh glycol treats the symptom and leaves the cause in place.
The service practices that prevent most of this are unglamorous. Use deionized or reverse osmosis water for any makeup, or use premixed PG25, and never tap water. Flush the system properly before the first fill, and keep the flush water within the same limits as the fill water. Change filters on differential pressure rather than on a calendar, because a filter that loads early is telling you something. Keep the loop closed, because every opening is an opportunity for oxygen and particles to enter.
Filtration is part of chemistry, not a separate topic. Inline filters on the CDU are commonly 25 to 50 micrometers, and side stream filters are often finer. Finer filtration protects the cold plates, but it loads faster and needs more attention. Record the clean filter differential pressure at commissioning, then trend it. A filter that climbs quickly after a repair is doing its job by catching the debris from that repair.
Put the Loop on the Maintenance Calendar
Chemistry only works as a routine. The useful cadence is daily monitoring of alarms, temperature, flow, and pressure; weekly walk of the loop for drips and filter differential pressure; quarterly sampling and quick disconnect inspection; and an annual laboratory panel, sensor calibration, and failover test. The exact intervals come from the CDU and fluid suppliers, and the first months after startup usually need more attention than the steady state.
The same loop should be documented in the inventory, not only in the maintenance log. A feed record with a supply temperature, a return temperature, and a maximum flow gives the chemistry program a design baseline to check against, and the rack summary shows whether the heat load still matches the loop. The data center management pages are a reasonable place to keep those numbers together, so the next person who opens a sample log also knows what the loop was designed to do.
The first step is small: pull one sample from one loop, log it with the date and the operating temperature, and compare it to the fill record. If there is no fill record, that missing baseline is the first finding, and creating one today is still better than starting next quarter.
Frequently Asked Questions
Is PG25 the same as antifreeze?
Why does glycol concentration matter if the loop never freezes?
What should the pH of a PG25 loop be?
How often should coolant be tested?
Should conductivity be low in a glycol loop?
Can I top off a glycol loop with tap water?
Stop reaching for a spreadsheet
Obelinf keeps every subnet, device, circuit, and rack in one live source of truth, with audit logs and a topology view. Free for personal use.
Related Articles

Data Center Liquid Cooling Operations
A day to day operating guide for liquid cooled data center halls, covering CDU and coolant loop monitoring, service boundaries, leak response, and preventive maintenance.
Read more
What Is a CDU? Facility Water, In Row, and In Rack Explained
A practical guide to coolant distribution units: the two loops they separate, the three form factors, how approach temperature sets the plant, and how to size flow from rack heat load.
Read more
Quick Disconnects and Coolant Connectors: UQD, Drip Free, and Flow Basics
How coolant quick disconnects work, what the OCP UQD sizes actually carry, where blind mate fits, and how to keep a connector from becoming the narrowest point in the loop.
Read more