Liquid cooling loop water treatment and chemistry field guide

Loop chemistry, and why the fluid touching a chip is a precision system

Liquid cooling loop chemistry is the specification and the upkeep of the fluid that runs through the cold plates clamped to the CPUs and GPUs. The technology cooling system, the clean secondary loop the CDU feeds, does not carry plain water. It carries a tightly controlled coolant, usually treated water or a propylene-glycol-and-water mix, with a corrosion-inhibitor package and a biocide, held inside a band on pH, conductivity, inhibitor level, biocide level, and cleanliness. That band is the difference between a loop that runs for years and a loop that fouls a cold plate in a season.

The reason the fluid is a precision system and not a utility is the geometry it has to pass through. A cold plate is a sealed block with internal microchannels pressed against the silicon, and those channels can be a fraction of a millimeter wide. Anything the chemistry lets happen, corrosion dropping metal oxide, biology growing a film, particulate riding in suspension, scale precipitating out, ends up narrowing or plugging the one path that takes heat off the most expensive hardware in the building. Get the chemistry wrong and the loop quietly cooks the chip it was built to protect.

This guide stays on the chemistry. How the loop gets built, flushed, pressure-tested, and balanced is in the liquid cooling loop commissioning guide, and how the CDU isolates and conditions the loop is in the CDU commissioning guide. Here the subject is the fluid itself: what it is, what corrodes it, what grows in it, what fouls it, and the sampling program that proves it is still in spec long after the commissioning agent has gone home.

Why does the chemistry matter so much in a cold-plate loop?

The chemistry matters because the cold-plate microchannel is the least forgiving piece of plumbing in the building, and everything the chemistry governs ends up there. A chilled water coil can tolerate some scale and a little biology and keep working. A microchannel cannot. Drop a flake of corrosion product, grow a layer of biofilm, or let particulate settle, and the channel restricts the flow to the chip that can least afford it. The chip throttles to protect itself, and if it cannot shed the heat, it fails.

Three failure paths run through the same narrow geometry. Corrosion of the mixed metals in the loop drops oxide and metal particles that lodge in the channels and also eats the cold plate from the inside. Biological growth lays down biofilm, which is a poor conductor of heat and a flow restriction at the same time, so it insulates the chip and chokes the channel in one move. Particulate, whether it came in with construction debris or precipitated out of a bad chemistry, simply plugs the gap. The wrong fluid does all three.

So the chemistry values are not housekeeping. They are the controls that keep those three paths shut. The exact targets belong to the coolant and cold-plate manufacturer and to the design, and they vary by the metallurgy and the fluid, so treat any number here as a starting point to confirm against the spec, not a setpoint to dial in blind. The principle does not vary: hold the fluid in spec, or the microchannel will tell you that you did not.

What is the difference between the TCS and the FWS chemistry?

The technology cooling system, the TCS, is the clean secondary loop from the CDU to the cold plates. The facility water system, the FWS, is the dirtier primary loop, the building chilled water the plant makes and the tower or dry cooler rejects. The CDU sits between them with a heat exchanger that passes heat from one to the other without passing the water, and that separation is exactly why the two loops carry different chemistry. The TCS guide covers the wall itself in the CDU commissioning guide; what matters here is that the chemistry on each side is not the same job.

The FWS is treated like an industrial cooling loop. It is larger, it can run open or makeup-fed, it tolerates more dissolved solids, and its water treatment looks like classic cooling-system practice: scale and corrosion inhibitor, biocide, and control of hardness and dissolved solids against the equipment it serves. The cooling-tower and chilled-water side is a different discipline with its own guides. The TCS is the tighter, cleaner, smaller loop, closed, low-volume, and held to far stricter limits on conductivity, particulate, and chemistry, because it ends in a microchannel and the FWS does not.

The line a crew most needs to keep straight is which loop they are working on, because the flush, the fill, the sample, and the treatment all reference one side or the other. Treat the TCS like facility water and you will fill a cold-plate loop with water that was never meant to go near silicon. Treat the FWS like the TCS and you will over-spend chasing a purity the building loop does not need. ASHRAE has expanded the guidance on how the two loops differ in fluid quality, and the equipment manufacturer sets the actual limits for each.

What coolant is used in data center liquid cooling?

The most common coolant in a single-phase direct-to-chip TCS loop is a propylene-glycol-and-water mix, commonly around 25 percent propylene glycol by volume, often written PG25, carrying a corrosion-inhibitor package matched to the metals in the loop. Treated or deionized-grade water shows up in some designs where freeze and biology are handled another way. The choice is not the contractor's to make on the jobsite. The equipment and cold-plate manufacturer specifies the fluid, and that specification governs over any rule of thumb.

Propylene glycol earns its place for two reasons beyond the heat-transfer job. It depresses the freeze point, which protects coolant sitting in an outdoor dry cooler or a cold loop in winter, and it suppresses biological growth far better than plain water in the warm, dark, low-flow environment of a cold-plate loop. The price for that is a heat-transfer penalty: a glycol mix carries less heat and pumps harder than water, so the design pays in flow and pump energy for the freeze and biostatic protection. Ethylene glycol moves heat better and turns up on some outdoor or primary loops, but propylene glycol dominates the server-side loop for leak safety and lower toxicity.

Treated water without glycol gives the best heat transfer and the lowest pumping cost, which is why some indoor loops with no freeze exposure use it, but it leans entirely on the inhibitor and the biocide to do what the glycol would have helped with. Whatever the fluid, the chemistry is the point: the coolant lives in contact with microchannels and dissimilar metals for years, so the concentration, the inhibitor, and the biocide are what keep it from corroding, scaling, or growing biology. Confirm the fluid, the concentration, and the additive package against the manufacturer's specification and the OCP or ASHRAE guidance the design referenced, and do not substitute.

Makeup water quality: why you do not use tap water

The water that goes into a TCS loop, whether it dilutes a glycol concentrate or fills a treated-water loop, has to start clean, and clean here means deionized or demineralized, not whatever comes out of the building tap. Tap water carries calcium and magnesium hardness that precipitates as scale, chlorides and sulfates that drive pitting and galvanic attack, and dissolved minerals that raise conductivity. Send that into a mixed-metal microchannel loop and you have seeded the exact failures the chemistry exists to prevent.

Deionized water is the usual starting point because it strips the ionic content that causes the trouble. A common reference for DI-grade dilution water is conductivity well under 1 microsiemens per centimeter, with chlorides and sulfates held to low single-digit parts per million or less, but the specific limits come from the fluid and equipment manufacturer, so confirm them rather than assume. The point of starting with DI water is that the inhibitor package was formulated for a known, low-ion baseline. Dilute a glycol concentrate with hard tap water and you have changed the chemistry the inhibitor was tuned for before the loop even runs.

The blunt version: tap water, softened water, and even reverse-osmosis water that does not meet the DI specification will accelerate corrosion in a mixed-metal loop and drop scale and particulate into the cold plates. The cost of the right water at fill is trivial next to a fouled rack of GPUs. Use the water the manufacturer calls for, document the conductivity and the chemistry of what actually went in, and never top a loop off from a yard hose.

The corrosion problem: mixed metals in one loop

A direct-to-chip loop is a collection of dissimilar metals sharing one fluid, and that is the setup for corrosion. The cold plates are commonly copper. The manifolds, fittings, and structure can mix in aluminum, stainless steel, brass, and brazed joints, each with its own filler metal. Put dissimilar metals in contact through a conductive fluid and you have built a galvanic cell: the less noble metal corrodes to protect the more noble one, dropping metal and oxide into the loop and thinning the part that is corroding.

Corrosion hurts the cold-plate loop two ways at once. It eats the metal, which over years can perforate a thin cold-plate wall or a manifold, and it generates particulate, the oxide and metal flakes that ride the flow straight into the microchannels and plug them. A loop can fail from corrosion long before anything leaks, because the corrosion product chokes the channels first. That is why corrosion control in a TCS loop is as much about keeping the fluid clean of what corrosion produces as it is about saving the metal.

The defenses are three, and they work together. Specify compatible materials so the galvanic couples are mild to begin with, which is a design decision made before the loop is built. Hold the water quality and pH so the fluid is not aggressive. And carry a corrosion-inhibitor package matched to the specific metals in the loop, which is the chemistry's main line of defense. The material list, the inhibitor, and the limits all come from the manufacturer, because the right answer depends entirely on what metals are actually in the system.

The mixed-metal galvanic couple: aluminum and copper

The galvanic couple that causes the most trouble in these loops is aluminum next to copper. Copper is far more noble than aluminum, so when the two share a conductive fluid, the aluminum becomes the sacrificial anode and corrodes preferentially, sometimes fast. A loop that mixes a copper cold plate with an aluminum manifold or an aluminum component, without an inhibitor package built for that couple, can pit the aluminum and drop corrosion product into the flow in a way that shows up as a fouled cold plate well upstream of any visible damage.

This is why the inhibitor package is not generic. A formula tuned for an all-copper-and-stainless loop is not the same as one that has to passivate aluminum too. Aluminum protection usually leans on different chemistry than copper protection, so a loop with aluminum in it needs an inhibitor that covers both, at the concentration the manufacturer specifies. Drop in the wrong package, or let the right one deplete, and the aluminum starts giving itself up to the copper.

The field lesson is to never introduce a metal the inhibitor was not formulated for. The classic mistake is a quick repair with a zinc-plated fitting, a galvanized component, or an aluminum part that was handy, into a loop whose chemistry never accounted for it. Now there is a galvanic cell the inhibitor cannot cover, corroding on its own schedule. Match every metal that touches the fluid to the compatibility list and the inhibitor package, and confirm both against the manufacturer's specification before anything goes in the loop.

The corrosion inhibitor package and how it depletes

The corrosion inhibitor is the additive package that passivates the metal surfaces so the fluid cannot attack them, and it is tuned to the specific metallurgy of the loop. Inhibitor chemistries vary: organic-acid technology, hybrid organic-acid formulations, and packages using nitrite, silicate, molybdate, or phosphate all appear, depending on the metals being protected. For the copper that dominates cold plates, azole compounds such as tolyltriazole and benzotriazole are common, because they lay a protective film on copper and brass that resists pitting and dealloying. Aluminum, where present, typically needs its own passivating chemistry in the same package.

The thing to understand about an inhibitor package is that it gets consumed. It is not a permanent property of the fluid. As it does its job, plating out on metal and neutralizing aggressive ions, the active concentration drops, and a loop running on a depleted inhibitor is a loop with no corrosion protection even though it is still full of the right-looking coolant. Depletion is gradual and invisible without testing, which is exactly why inhibitor level is one of the parameters a sampling program tracks.

The practical job is to test the inhibitor level against the manufacturer's acceptance band and top it up or replace the fluid before it falls out of range, not after the corrosion has already started. The depletion rate depends on the loop volume, the metals, the temperature, and the makeup losses, so it is not a fixed calendar interval, it is a trend you watch. Treat the inhibitor like brake pads, not like a permanent part: it wears, and you replace it on a schedule the testing sets, with the product and the band the manufacturer specifies.

What is biofouling in a cooling loop?

Biofouling is biological growth inside the loop, bacteria and other microorganisms that colonize the wetted surfaces and lay down biofilm, a slime layer on the channel walls. A cold-plate loop looks like a hostile place for life, but a warm, dark, low-flow, water-bearing loop is exactly where biology will grow if the chemistry lets it, and the consequences land squarely on the cold plate. Biofilm is a poor conductor of heat, so it insulates the chip even where it is thin, and it is a flow restriction, so it narrows the microchannel and chokes the coolant the chip needs.

Biofouling also feeds corrosion. Some microbial colonies drive microbiologically influenced corrosion, attacking the metal under the biofilm where the inhibitor cannot reach, so a bio problem becomes a metal problem on top of a flow problem. The growth is slow and quiet until it is not. By the time a cold plate is throttling from biofilm, the loop has been growing it for a while, and clearing an established biofilm out of a microchannel is far harder than keeping it from forming.

Two things hold biology back, and they work together. The glycol in a PG mix suppresses growth far better than plain water, which is one reason a glycol loop is more forgiving than a treated-water loop on the bio front. And the biocide in the package kills what tries to establish. A treated-water loop with no glycol leans entirely on the biocide and the cleanliness, so it is less forgiving and needs tighter attention to the bio side. Either way, biological count is a parameter the sampling program watches, against the manufacturer's limit.

The biocide and keeping the dosing controlled

The biocide is the additive that kills the microorganisms before they establish a biofilm, and like the inhibitor, it is consumed as it works and has to be maintained. Some loops carry biocide in the fluid package, some dose it through the CDU, and some pair it with ultraviolet treatment on a side stream to knock down biology without adding more chemistry. Whatever the method, the goal is the same: keep the biological activity below the level where biofilm forms in the microchannels.

The warm-water trend in liquid cooling raises the stakes on the bio side. Running the loop warm, in the 30s of degrees C, is good for energy efficiency because it lets the facility reject heat for more of the year, but warm water is also better for growing biology than cold water. So the warmer the loop runs, the more the biocide and the glycol have to carry, and the less margin there is for a depleted biocide before growth starts.

Dosing is a balance, not a maximum. Too little biocide and biology grows. Too much, or the wrong biocide, can attack the elastomers and the metals or interact with the inhibitor, so the dosing is held to the band the manufacturer specifies, not pushed high for insurance. Test the biocide level on the same schedule as the inhibitor, top up or re-dose before it falls out of range, and use the product the fluid manufacturer calls for, because a biocide that is wrong for the loop chemistry causes its own problems.

Holding the pH in band

The pH of the coolant is one of the controlling variables for corrosion, and it sits in a band the inhibitor package was built around. Common practice keeps a treated water-glycol loop on the neutral-to-mildly-alkaline side, often cited somewhere in the range of about 7 to 9, but the exact target belongs to the fluid manufacturer and depends on the metals and the inhibitor chemistry. The point is that the inhibitor works inside a pH window, and outside it the protection degrades.

Out of band cuts both ways. Let the pH drift acidic, below neutral, and corrosion of both ferrous and nonferrous metals accelerates, with the rate climbing fast as the pH falls. Push it too alkaline and you can drive scaling and attack on aluminum. pH drifts on its own as the fluid ages, as the inhibitor depletes, as gases dissolve in, and as the loop reacts with its metals, so a loop that filled in band will not stay there without attention. A buffer in the package resists the drift, but it is not infinite.

Test the pH on the sampling schedule and read it as an early warning, not just a pass-fail. A pH that has started to move is often the first sign the inhibitor is depleting or something has contaminated the loop, before the corrosion shows up in the metal counts. Hold it in the manufacturer's band, investigate a drift rather than just correcting the number, and confirm the target against the fluid spec, because the right pH is the one the inhibitor was formulated for.

Why does low conductivity matter near electronics?

Conductivity is a measure of the dissolved ionic content of the coolant, and a TCS loop is generally held to low conductivity for two related reasons. The first is corrosion: ionic content is what makes a fluid electrically conductive, and a conductive fluid is what carries the galvanic current between dissimilar metals. Low conductivity means fewer ions to drive that corrosion, so conductivity doubles as a clean indicator that the water quality and the chemistry are still where they should be. A climbing conductivity often means contamination or inhibitor breakdown before anything else shows it.

The second reason is the leak. In a sealed direct-to-chip loop the coolant does not normally touch the electronics, so conductivity is not a dielectric requirement the way it is in immersion cooling. But a leak puts the coolant onto a powered board, and a conductive fluid landing on energized electronics is a short waiting to happen. Lower conductivity reduces, though it does not eliminate, what that leak does, and many conductive-fluid leak-detection schemes rely on sensing the coolant's conductivity, so the number ties into how a leak gets caught as well as what it does.

The targets vary and the guidance is not uniform. Some direct-to-chip references call for treated water conductivity held low, with figures in the single-digit to low-tens of microsiemens per centimeter cited for the loop, while a DI makeup is far lower still, and a glycol-and-inhibitor fluid reads higher than pure water because the additives themselves carry ions. ASHRAE TC 9.9 does not fix a single conductivity ceiling, so the limit comes from the fluid and equipment manufacturer. Track conductivity as a trend against that limit, and treat a rise as a question to answer, not a number to ignore.

Particulate and the cold-plate channel size

Particulate is the suspended solid matter in the coolant, and in a microchannel loop it is a direct threat, because the particle does not have to be large to plug a channel that is a fraction of a millimeter wide. Particulate comes from two places: construction debris left in a new loop, and the ongoing production of corrosion product, precipitated scale, and biological matter from a loop whose chemistry is off. Filtration handles both, but only if the rating matches the geometry it is protecting.

The number that governs is the relationship between the filter micron rating and the cold-plate channel size. A filter rated coarser than the smallest channel passes particles that can lodge in the plate, so the filtration has to be fine enough to catch what the channel cannot tolerate. Cold-plate and quick-disconnect specs commonly call for filtration to 50 microns or finer for the main flow, with many direct-to-chip designs going finer still on a side stream, down toward the low single-digit micron range, to polish the loop. The cold-plate manufacturer's filtration target is the one that governs, because the manufacturer knows the channel size.

Cleanliness is a chemistry job as much as a filtration job, because a loop with a controlled chemistry produces far less particulate to filter in the first place. Good inhibitor and biocide control means less corrosion product and less biological matter entering suspension, so the filter is catching construction debris and the occasional upset rather than fighting a loop that is constantly making its own particulate. Keep the chemistry clean and the filter has an easy job. Let it drift and the filter loads up fast and the cold plates see what gets past it.

Full-flow and side-stream filtration

Filtration in a TCS loop usually runs in two arrangements at once. A full-flow filter sits in the main coolant path and catches particles before they reach the manifolds and the cold plates, commonly rated around 50 microns or 25 microns in the CDU. A side-stream filter takes a slipstream off the main flow, often on the order of 10 percent of the total, and runs it through a much finer element, sometimes down toward a few microns or finer, polishing the whole loop volume over time without choking the main flow with a fine filter the pump could not push through.

The two filters do different jobs and both belong in the design. The full-flow filter protects the cold plates in real time from anything large enough to plug a channel right now. The side-stream filter slowly cleans the bulk fluid, pulling out the fine particulate that the full-flow rating lets pass, so the loop gets cleaner over time rather than just holding steady. Confirm both are installed, at the rating the cold-plate spec calls for, and that the filter differential-pressure alarm is set and reports, because a loaded filter starves the loop the same as a closed valve.

Filters are a maintenance item, not a fit-and-forget part. They load up, the differential pressure across them climbs, and a fouled filter restricts the flow the chips depend on. The change-out interval is not a fixed calendar date, it is driven by how much the loop is making the filter catch, which is itself a signal: a filter loading fast is telling you the chemistry is producing particulate. Track the filter differential pressure, change the elements on the trend the manufacturer's limits set, and read a rapidly loading filter as a chemistry problem to investigate, not just a part to swap.

Cold-plate fouling: the clog, the hotspot, the throttle

Cold-plate fouling is the end state every other chemistry parameter exists to prevent: deposit or growth inside the microchannels that restricts the coolant flow or insulates the chip. It is where corrosion product, biofilm, scale, and particulate all end up, and it is the failure that actually takes the hardware down. A fouled channel carries less coolant to the chip, the heat does not leave fast enough, the silicon runs hot, and the chip throttles to protect itself. Push it past throttling and the chip fails.

The symptom is a hotspot that does not match the loop. The CDU is making its supply temperature, the facility water is on spec, the flow at the manifold looks right, and yet one node or one chip runs hotter than its neighbors and throttles under load. That signature, good loop numbers and a bad chip, points at the cold plate itself, because the restriction is downstream of everything the loop instruments can see. By the time it shows, the fouling is established, and a cold plate cannot be flushed back to clean the way a loop can.

This is the whole argument for chemistry control stated as a consequence. Every parameter, the inhibitor, the biocide, the pH, the conductivity, the filtration, the water quality, exists to keep the microchannel clear, because the microchannel is the one part of the system that cannot tolerate the deposit and cannot be cleaned once it has it. Hold the chemistry in spec and the fouling never starts. Let it drift and the cold plate is where the bill comes due, on the most expensive hardware in the room.

Flushing the construction debris from a new loop

A new loop comes full of construction debris, and it has to be flushed clean before any coolant or any cold plate goes in. New piping, manifolds, and hoses carry weld slag, pipe scale, cutting oil, thread sealant, plastic shavings, and the fine grit that gets into anything built on a jobsite. Send that into a microchannel and you choke the channel on day one, before the chemistry has had any chance to matter. The flush is the first cleanliness control in the loop's life, and it is the one most likely to be rushed.

The flush is staged from coarse to fine. You circulate a procedure fluid at a velocity high enough to scour the pipe walls, capture the heavy debris on a coarse filter, then step down to progressively finer filtration until the loop holds a measured cleanliness target. The flush fluid is the one the spec calls for, often a conditioned water meeting a documented conductivity and microbiological limit, not whatever is in the yard hose, because filling a clean loop with dirty flush water defeats the point. The staged flush procedure and the cleanliness targets are covered in the liquid cooling loop commissioning guide.

The chemistry point is that the flush sets the baseline the chemistry program builds on. Connect cold plates only after the cleanliness result passes, because the flush is the one chance to clear the construction debris before the channels are in the path. Skip it or rush it and you have started the loop dirty, and no amount of inhibitor or biocide cleans out construction debris that is already lodged in a cold plate. Flush first, prove it clean, then fill.

Fill, de-air, and the acceptance sample

Filling a TCS loop is a chemistry step, not just adding fluid. You fill with the coolant the spec calls for, the right glycol concentration and the right inhibitor and biocide package, not plain water and not a top-off of something convenient, and you fill in a way that does not trap air. Air in a liquid loop blocks flow at a high point or in a cold plate, cavitates a pump, and degrades heat transfer, so the fill works the air up to the separator and bleed points and vents it, then tops off and repeats until the loop holds steady flow and pressure with no air signature.

The initial chemistry is the baseline everything after is measured against. Once the loop is filled, de-aired, and at operating condition, you draw an acceptance sample and confirm the as-filled chemistry: the glycol concentration, the inhibitor level, the biocide level, the pH, the conductivity, the particulate or cleanliness, and the biological count, against the manufacturer's acceptance criteria. That sample is the zero point. A reading next year only means something measured against what the loop was accepted at, so a loop turned over with no acceptance sample is a loop nobody can trend.

The fill, de-air, and acceptance sequence is covered as a commissioning step in the liquid cooling loop commissioning guide. The chemistry job inside it is narrow and concrete: the loop got the specified fluid at the specified concentration with the inhibitor and biocide in band, and the as-accepted numbers are recorded as the baseline the operations program will sample against. Fill it right, prove it in spec, and write down what in-spec looked like on day one.

Why does liquid cooling water need a sampling program?

A TCS loop needs a regular sampling program because every chemistry parameter that protects the cold plates drifts on its own, and drift is invisible without testing. The inhibitor depletes, the biocide is consumed, the pH wanders, the conductivity creeps, particulate accumulates, and biology grows, all slowly, all silently, and all heading toward a fouled cold plate if nobody is watching. The sampling program is what turns those slow, invisible trends into numbers someone can act on before the chip throttles.

A working program samples on a schedule and tests the parameters that matter: pH, conductivity, glycol concentration, inhibitor level, biocide level, particulate or total suspended solids, and a biological count, with some loops adding dissolved oxygen or oxidation-reduction potential and a check on chlorides and sulfates. Some of those read inline through CDU instrumentation, and some go to a lab, and the lab results are what catch the things the inline sensors cannot. The schedule is tighter early in a loop's life and on warm loops, then settles as the trend establishes, but it never stops.

The value of the program is the trend, not the single reading. One sample in spec tells you the loop is fine today. A series of samples tells you the inhibitor is depleting on a slope that will cross the limit in two months, so you act now instead of after the corrosion starts. That is why the samples have to be logged somewhere durable and trended, not read and forgotten. A field platform like FieldOS holds the sampling schedule, the per-loop history, and the limits, so a deviation surfaces against the baseline instead of sitting in a notebook nobody opens until a chip is already throttling.

The ongoing water-treatment program

The chemistry does not end at commissioning. A TCS loop is a slow chemistry experiment running every hour for years, and keeping it in spec is an ongoing water-treatment program, not a one-time fill. The program is the schedule that ties together the sampling, the inhibitor and biocide top-ups, the filter change-outs, and the periodic cleaning, all driven by the trends the sampling produces rather than a fixed calendar. It is the difference between a loop that ages gracefully and one that quietly degrades until a cold plate fouls.

The program looks like a cooling-tower water-treatment program in its bones, the same cycle of sample, trend, dose, and clean, but it runs cleaner and tighter. A cooling tower is an open, dirty, makeup-heavy system fighting scale, dissolved solids, and heavy biology with a large chemical feed. A TCS loop is a closed, clean, low-volume system holding a precision fluid to far stricter limits, so the doses are small, the cleanliness target is high, and the consequence of falling out of spec lands on silicon rather than on a chiller barrel. The cooling-tower water-treatment side is its own guide; the principle carries over but the limits do not.

What makes the program work is ownership and records. Someone owns the schedule, the samples get drawn and logged on time, the trends get reviewed, and the top-ups and changes happen before the limits are crossed, not after. A loop handed over without a named program, a sampling schedule, and a place to trend the results is a loop that will run fine until it does not, and nobody will see it coming. The maintenance program is a deliverable, the same as the commissioning record.

Makeup and top-up: the right fluid, not water

A closed loop should not lose much fluid, but it loses some, to small weeps, to service disconnects, to evaporation at vents, and over the years the level drops enough to need makeup. The rule that gets broken here is simple and expensive: top up with the specified coolant, not with plain water. A glycol loop topped off with water has its glycol concentration diluted, its freeze and biostatic protection weakened, and its inhibitor and biocide diluted below the protective band, all at once, from a quick top-off that looked harmless.

Makeup is a chemistry event, not a level adjustment. Adding fluid changes the concentration of everything in the loop, so the makeup has to be the right fluid at the right concentration with the additive package, and the loop should be re-sampled after a significant makeup to confirm it is still in band. A large makeup also raises the question of why the loop lost that much fluid, which is usually a leak worth finding before it grows.

The discipline is to treat the makeup fluid the same as the fill fluid: the manufacturer's coolant, the manufacturer's concentration, the manufacturer's additives, and a sample afterward. Keep the right fluid on hand so nobody is tempted to reach for the hose, log every makeup with the volume and the fluid used, and confirm the chemistry after. The loop that gets topped off with whatever is convenient is the loop that drifts out of spec one top-off at a time.

A leak is a chemistry loss and an electronics risk at once

A leak in a TCS loop is two problems in one event. It is a chemistry loss, because the loop is losing the precision fluid it depends on and the level and concentration both move, and it is an electronics risk, because the coolant can land on a powered board. Even a small weep at a quick-disconnect is worth chasing, because it dilutes the chemistry as it gets made up and it drips onto whatever is below it. The leak-detection and containment strategy that catches it is covered in the liquid cooling loop commissioning guide and the CDU guide; the chemistry point is that the fluid lost is not free water, it is a treated coolant that has to be replaced in kind and re-verified.

The two failure modes argue for the same response: find the leak, fix it, make up with the right fluid, and re-sample. A leak that gets topped off with water instead of coolant trades an electronics risk for a chemistry problem, which is no trade at all.

Immersion dielectric fluids are a different chemistry

Immersion cooling is a different fluid and a different chemistry from the water-and-glycol cold-plate loop this guide covers. Immersion drops the whole server into a bath of dielectric fluid, single-phase or two-phase, and that fluid is non-conductive by design because it is in direct contact with the energized electronics. Its chemistry concerns are about material compatibility, fluid degradation, and dielectric strength, not about corrosion inhibitors, biocides, and the water quality that govern a TCS water loop.

The distinction matters because the two get confused. A direct-to-chip loop runs treated water or a glycol mix through sealed cold plates and the fluid never normally touches the electronics, so it is treated for corrosion and biology like a closed hydronic loop. Immersion runs a dielectric fluid in contact with the boards, so its rules are different. This guide is the water and glycol cold-plate loop. The immersion fluid is its own scope, cross-linked by topic rather than covered here.

What to document

The chemistry record is what lets the next engineer tell a new problem from how the loop has always run. A loop sampled but never logged is a loop with no baseline, and the first fouled cold plate becomes a guessing game. Capture the as-accepted chemistry at fill and every sample after, with the date, the loop, the values, the limits, and who drew it, so the trend is visible and a deviation surfaces against the baseline instead of getting lost.

The table below is a starting frame for the parameters a TCS loop typically tracks. The targets are deliberately written as ranges and directions, because the actual numbers belong to the fluid and equipment manufacturer and vary by the metallurgy and the fluid. Confirm every value against the spec, and record what the loop was actually accepted at, not the generic figure.

Common mistakes

• Filling or topping a TCS loop with tap, softened, or unconditioned water instead of the specified DI-grade water or coolant.
• Running no corrosion inhibitor, or letting the inhibitor deplete unnoticed, in a loop full of mixed metals.
• Introducing an incompatible metal, a zinc-plated or aluminum fitting, that the inhibitor package was never formulated for.
• Letting biology grow from a missing or depleted biocide, especially on a warm or treated-water loop, until biofilm narrows the microchannels.
• Filtering coarser than the cold-plate channel size, so particulate that plugs a channel rides straight through the filter.
• Connecting cold plates to a new loop that was never flushed to the cleanliness target, choking a channel with construction debris.
• Running no regular sampling program, so the inhibitor, biocide, pH, and conductivity drift out of spec invisibly.
• Topping off a glycol loop with plain water and diluting the glycol, inhibitor, and biocide all at once.
• Treating the TCS like facility water, or the facility water like the TCS, and applying the wrong chemistry to the wrong loop.

Field checklist

Standards and references

The thermal and fluid-quality framework comes from ASHRAE Technical Committee 9.9, whose thermal guidelines and liquid-cooling guidance address the facility water classes and the differing fluid-quality requirements of the TCS and FWS loops. The Open Compute Project publishes the cold-plate and liquid-cooling requirements many AI deployments design to, including guidelines for propylene-glycol-based and water-based heat transfer fluids in single-phase cold-plate racks, with guidance on coolant chemistry, cleanliness, and filtration. Name the OCP document by topic and confirm the current revision, because the documents revise on their own cycle, and ASHRAE does not fix every limit, for example it sets no single conductivity ceiling.

The pH, the conductivity, the inhibitor and biocide chemistry and concentrations, the particulate and filtration targets, and the glycol type and concentration are governed first by the coolant, cold-plate, and CDU manufacturer's fluid specification, which overrides any general guideline where it is stricter. General water-treatment practice informs the program, the sampling, the trending, the dosing, and the cleaning, but the TCS limits are tighter than a typical cooling loop and the manufacturer sets them. Where the loop chemistry borrows from cooling-system water treatment, borrow the method, not the numbers.

Three things hold across every loop, whatever the spec says. The cold-plate microchannels are unforgiving, so keep the fluid in spec or they foul. The loop is a mix of metals in a wet, warm environment, so the inhibitor and the biocide have to be matched, maintained, and topped up before they deplete. And the loop makes and accumulates particulate and biology, so filter it, flush it clean before fill, and sample it on a program. The edition numbers, the OCP revisions, and the specific limits move, so confirm them against the published documents and the equipment manufacturer before citing them on a submittal.

Units, terms, and acronyms

Loop chemistry borrows vocabulary from water treatment, from the IT side, and from the chip and fluid vendors, and the same idea reads differently across a fluid datasheet, a CDU submittal, and an ASHRAE guideline. The terms below travel across the chemistry scope.

TCS

Technology cooling system, the clean conditioned secondary loop from the CDU to the cold plates, held to the strict chemistry

FWS

Facility water system, the dirtier primary building water loop the CDU rejects heat into, treated like an industrial cooling loop

Propylene glycol (PG)

The common glycol in the secondary loop, often near 25 percent (PG25), for freeze protection and biostatic effect at a heat-transfer penalty

Corrosion inhibitor

The additive package that passivates the loop metals, such as azoles for copper, matched to the metallurgy and consumed over time

Biocide / biofouling

Biofouling is biological growth and biofilm in the loop; the biocide is the additive that kills it before it forms

Conductivity

The ionic content of the fluid in microsiemens per centimeter (uS/cm); held low to limit galvanic corrosion and as a contamination indicator

Particulate / filtration

Suspended solids in the coolant, removed by full-flow and side-stream filters rated to the cold-plate channel size

Cold-plate microchannel

The fraction-of-a-millimeter passages inside the cold plate where the coolant takes heat off the chip; the least forgiving part of the loop

DI water

Deionized water, the low-ion makeup and dilution water the fluid is formulated for, far cleaner than tap or softened water

TSS

Total suspended solids, the particulate load in the fluid, trended as a sign of corrosion or biological activity

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is a CDU in a data center?
• What is direct-to-chip liquid cooling?
• What is direct-to-chip liquid cooling?
• Why do data centers need liquid-cooling leak detection?
• What is data center waste heat reuse?