Eight Questions That Stall a Data Center Coolant Decision — Answered Straight

Coolant decisions rarely stall on price. They stall on questions that never got a written answer. Those are the ones that surface in a review meeting, get a vague reply, and reappear next quarter.

Here are the eight we hear most. Each one is answered the way we would want it answered: the conclusion first, then what to check, then who owns the check. Take the page into your review and cross the questions off one at a time.

Glacier Coolant fields these questions across three decades of circulating loops — industrial service, mostly, which is harder on a loop than a data hall ever is. The answers below are the ones that survived review meetings.

1. Is “PG 25″ enough on the specification line?

PG 25 is propylene glycol at roughly 24.5–29.5 vol%, with pH 8.0–10.5 in the Open Compute Project’s reference. It is a sensible starting point for an indoor technology loop, where the wetted list carries no aluminium and the conductivity requirement goes no further than corrosion control.

The word to watch is “reference”. Where the loop runs outdoors, touches aluminium, or carries a conductivity limit, the reference stops short and the spec needs a named grade. Two of the inputs that decide the grade hide inside the grade number itself: the inhibitor package and the conductivity trim. A spec that says PG 25 outsources those two decisions to whoever fills the truck.

What to check: temperature band, wetted materials, conductivity requirement, loop volume, make-up water, test regime.

Who to ask: the supplier should decode the grade number you are buying. How temperature bands map to grade families shows what the mapping looks like.

2. Can glycol run on aluminium cold plates?

The base fluid is rarely the problem. Aluminium holds a protective film inside a pH window. It loses that film when the chemistry drifts alkaline — the opposite of what steel-conditioned maintenance habits expect. What decides the outcome is the inhibitor package and the pH discipline, not the words “propylene glycol”.

In our range, grades built for aluminium service carry an L suffix. The letter is not decoration; it names an inhibitor package formulated and validated for aluminium metallurgy. A grade without it is a different product, whatever the base fluid says.

What to check: the wetted-materials list first, then the pH window of the grade on offer, then the make-up water quality that will drive drift.

Who to ask: us for the grade and its pH window; the plate maker for brazing and metallurgy questions. What aluminium in the wetted list asks of a fluid covers the distinction in detail.

3. Will the microchannels clog?

They clog from three sources, in roughly this order of frequency. Debris left in the loop at install. Corrosion products made inside it. Biofilm shed from warm, slow pockets. The passages are sub-millimetre, so anything that passes a strainer can still sit in a plate.

The Open Compute Project’s commissioning guidance calls for filtration down to about 5 μm before service. It also calls for flush velocities of 3–5 ft/s. The cold plate maker’s numbers take precedence over both. After commissioning, the defence is filtration discipline and a sampling schedule, not hope.

What to check: the installed filter grade against the plate maker’s number, and the last commissioning flush record.

Who to ask: the cold plate maker for channel and filtration numbers; us for particle and biofilm control through the fluid programme.

4. How often do we sample or replace it?

By trend, not by calendar. A sample answers three questions. Is the inhibitor still carrying? Is contamination entering? Is the chemistry drifting? None of those reads a calendar.

The rhythm that works:

Equipment makers publish the six- and twelve-month guidance for high-density and general service; the intervals are theirs, not ours. The trend report is where a replacement decision actually lives. A loop whose numbers hold needs no replacement. A loop whose numbers slide needs one before the calendar says so.

What to check: whether a baseline exists at all. Most loops we are asked about have samples but no baseline, which makes every reading unanswerable.

Who to ask: us for the sampling plan and the trend report.

5. What happens if it leaks?

In a direct-to-chip loop the coolant does not normally touch electronics, which changes the emergency. The immediate risks are loss of flow and damage to whatever the fluid reaches — not short circuit. Leak-detection hardware reads conductivity precisely because a leak changes the chemistry it sees.

The order that works: locate, isolate the affected node, then decide. Draining a whole loop because one quick-disconnect wept throws away good chemistry. It also invites a refill without a baseline. After the repair, sample before returning the loop to service, because the event changed its history.

What to check: where detection sensors sit, and whether the response plan distinguishes a node-level event from a loop-level one.

Who to ask: the CDU and detection hardware makers for sensor placement; us for the chemistry decision after the event.

6. How is it shipped and stored?

Propylene glycol based coolants generally ship as non-hazardous goods, which keeps freight simple and routine. The operative word is “generally”: classification is grade-specific, and the MSDS for the exact grade travels with the delivery. Ask for it before the shipment, not after.

Storage asks for sealed containers, out of direct sun, away from oxidising materials — standard practice for any glycol product. Drums opened for top-ups are where contamination enters a loop, so decanting discipline is part of fluid discipline.

What to check: the MSDS for the named grade, and who on site owns drum hygiene.

Who to ask: us — the documents ship with the fluid, and we will send them ahead on request.

7. Can it mix with what is already in the loop?

Different inhibitor packages are built for different metals and can react with each other into sludge. Mixing is how a working loop becomes a flushing job.

Where a loop is being converted, the sequence is: drain, flush, verify, fill. The verification is a measurement, not an appearance — water that looks clean can still carry the previous fluid’s markers. Flush criteria belong in the conversion plan, and the fill gets a fresh baseline the day it goes in.

What to check: what is currently in the loop, if anyone knows. A surprising number of loops run a mix nobody documented.

Who to ask: the contractor owns flush execution; we supply the criteria and the post-fill baseline. Where coolant actually works in a data center maps which loops need which answer.

8. How long does it last?

The honest unit of coolant life is a trend report, not a year count. Inhibitor reserve, pH stability and contamination levels decide when a fluid needs attention, and those are measured values with a history.

A loop with a baseline and a sampling rhythm gets an answer every time it is sampled. The numbers are holding, or the numbers are sliding and here is what to do. That is a better contract than a year figure nobody can stand behind, and it is the one we offer. The report comes with the fluid, and the decision dates come from your loop’s own data.

What to check: the reserve alkalinity and inhibitor marker trend, against the commissioning baseline.

Who to ask: us, with the sample history. The full product range by temperature band shows which grades carry which monitoring programme.

Who does what

The boundaries are the point. A supplier who claims every column of that table is a supplier who owns none of it when a review asks.

What arrives with the fluid

Documents first. Grade datasheet, MSDS, fill and commissioning sequence — sent ahead of the truck, not in it.

A baseline, not just a barrel. The first sample after fill becomes the reference every later sample is judged against, and the report says so in writing.

A rhythm matched to your loads. High-density loops and general service loops do not share a calendar, and the sampling plan should not pretend they do.

A name behind the answer. When one of the eight questions needs a called shot, the answer comes with the documents that support it.

Final thoughts

None of the eight questions is exotic. They stall projects because they arrive in meetings without written answers, and a verbal “it should be fine” does not survive a review cycle.

Written answers change the meeting. Question one gets a spec line with a named grade. Question two gets a wetted-materials check and a suffix on the grade number. Question four gets a baseline and a rhythm. Question eight gets a trend report instead of a promise.

Glacier Coolant has answered these questions across three decades of circulating loops, most of them running colder and dirtier than a data hall ever will. Bring us the ninth question, the one specific to your loop. It gets the same treatment: a conclusion, the criteria behind it, and the documents to take back into the review.