Liquid Cooling Leak Points, Component by Component: CDUs, Manifolds, Quick Disconnects, Hoses and Rear Door Heat Exchangers

• A liquid cooled hall has two loops. The facility water system (FWS) is the building side; the technology cooling system (TCS) runs from the coolant distribution unit (CDU) to the cold plates.
• Quick disconnects fail from debris on the faces, damaged O-rings, partial engagement, side load from heavy hoses and blind mate misalignment. A drip at disconnect means wear or damage.
• Any coolant in a drip tray is a finding, even if no alarm fired. Clear PG25 is hard to see on a dark tray; a sticky film is the sign.
• Coolant chemistry turns into leaks. Glycol breaks down into acids, inhibitors are used up, and corrosion attacks thin cold plates and fittings.

A direct to chip liquid cooling loop has two sides separated by a heat exchanger in the coolant distribution unit (CDU). The facility water system (FWS) supplies the CDU from the building. The technology cooling system (TCS) is the closed loop from the CDU to the cold plates.

The cold plates sit directly on the central processing units (CPUs) and GPUs. OCP uses the TCS term for this loop, and it is the loop that carries PG25.

• Facility water system (FWS): chilled or condenser water from the building plant, often at higher pressure than the TCS.
• Coolant distribution unit (CDU): pumps, a heat exchanger, a reservoir, filters and controls between the two loops.
• Rack manifolds: vertical supply and return headers that feed each server.
• Quick disconnects and hoses: the connections between manifold and server.
• Cold plates: thin walled heat exchangers on the chips.
• Rear door heat exchangers: coils on the rack door that cool exhaust air.

Each part has its own leak points. The sections below take them in order, from the building side to the rack.

PG25 is inhibited propylene glycol at about 25% concentration in water. It is the default TCS coolant because it balances heat transfer, freeze protection and biological stability. An earlier OCP draft gives the range as 24.5 to 27.5% (OCP draft).

Propylene glycol is chosen partly because at these concentrations it is biostatic, so it resists microbial growth. Bacteria in a closed loop foul filters and cold plates and can drive corrosion, as covered in Part 3.

The same earlier draft lists PG25's freeze point as 9 to 14°F (−13 to −10°C). That protects equipment during shipping, storage and outages in cold sites.

"Inhibited" matters as much as the glycol. The inhibitor package protects the metals in the loop, and when it is used up the fluid becomes corrosive.

The OCP guideline

The OCP document is titled Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Liquid Cooled Racks. Revision 2.0, Version 1.0.0, dated September 15, 2026, gives recommendations for inhibited PG25 in the TCS (OCP PG25 guidelines). It is the reference to check fluid selection, testing and maintenance against.

A CDU leaks on the facility water side mainly at its plate heat exchanger gaskets, its pump seals and its pipe joints. These are the same failures found in any mechanical room, brought into the white space.

• Plate heat exchanger gasket leaks at the edges of the plate pack.
• Pump seal leaks, covered in Part 1.
• Joint leaks at flanges, unions and threaded fittings, covered in Part 2.

Facility water often runs at higher pressure than the TCS. A leak on this side can push more water, faster, than a TCS weep.

An internal leak across the heat exchanger is a separate risk. Because of the pressure difference, facility water can cross into the TCS and dilute the coolant, which shows up as chemistry drift rather than a puddle.

A CDU leaks on the technology cooling side mainly at points people open for service. Reservoir and fill points, filters and strainers, pump seals and sensor ports are the usual places.

• Reservoir and fill points, where coolant is topped up and caps are removed.
• Filters and strainers opened for service and closed again.
• Pump seals on the TCS pumps.
• Sensor ports for pressure, temperature and flow.

Anything opened for service is a joint that was remade by hand. A filter housing closed in a hurry, or a fill cap left loose, is a common source of slow TCS loss.

The CDU's pressure and level displays show it first. A reservoir level that drops between top ups, or a TCS pressure that slowly falls, means coolant is leaving the loop somewhere.

Rack manifolds leak at their port connections, at vents and drains, and at the manifold to hose fittings. Every port is a joint, and a manifold has many.

• Threaded or brazed port connections along the manifold body.
• Vents and drains at the top and bottom of each manifold.
• Manifold to hose fittings, where each server connects.

Vibration and repeated server swaps loosen fittings. Each swap works the connections a little, and the lower ports carry the weight of hoses that hang below them.

A manifold leak runs down the rack, often behind equipment. Staining or residue on the rack frame below a manifold is the trace to look for.

Quick disconnects fail when their sealing faces or O-rings are damaged, dirty or not fully engaged. They are designed to be connected and disconnected often, which makes them both convenient and vulnerable.

• Debris or fibers on the mating faces, which hold the seal open.
• Damaged or swollen O-rings from incompatible fluid or age.
• Partial engagement, where the coupling is not fully latched.
• Side load from a heavy hose pulling on the coupling.
• Misalignment in blind mate connectors beyond the connector's tolerance when a server is racked.

Flat face dry break designs are meant to avoid spillage when connecting or disconnecting. Drips at disconnect therefore point to wear or damage, not normal behavior.

OCP connector specifications

OCP also publishes specifications for the Universal Quick Disconnect (UQD), in hand mate and blind mate versions. It also covers the Open Rack version 3 (ORv3) Blind Mate Quick Connector (BMQC) (OCP Cold Plate sub-project). Connectors built to a common specification make inspection and replacement more consistent across vendors.

After a server is swapped, every quick disconnect it touched should be checked for full engagement, wetness and hose routing. Server swaps are the moment connections are most likely to be disturbed.

1. 1. Confirm each coupling is fully latched, by feel and by any visual indicator.
2. 2. Look and feel for wetness or residue at each coupling a few minutes after flow resumes.
3. 3. Check that hoses are not kinked, crushed in the door or rubbing on rack edges.
4. 4. Check that hose weight is supported and not hanging on the coupling.
5. 5. Look in the drip tray below the server for fresh coolant.
6. 6. Note the swap and the check in the rack's record.

This takes a few minutes per server. It catches the partial engagement and side load faults that cause most connector leaks.

Hoses fail mostly from how they are routed and supported, not from the hose material itself. Kinks, crushing, abrasion and poor support account for most problems.

• Kinks and crushing restrict flow and stress the hose wall.
• Abrasion where hoses rub on sharp rack edges.
• Unsupported hoses hang their weight on the fittings.
• Tight bends at the fitting fatigue the hose end first.

A kinked hose also starves a cold plate of flow, so it can cause an overheating problem before it ever leaks. Thermal imaging can show a server running hotter than its neighbors on the same loop.

Routing is set at installation and disturbed at every change. A hose that was fine at commissioning can be trapped in a door or bent hard after the third server swap.

Rear door heat exchangers leak at their coils and at the fittings on the hinge side. They also produce condensation when supply water runs too cold.

• Coil leaks at brazed joints and return bends.
• Fitting leaks at the door hinge side, where hoses flex every time the door opens.
• Condensation if supply water runs below the room's dew point, covered in the Moisture and Air sub-series.

The hinge side hoses take a bend every time someone opens the door. That makes them a fatigue point, and the fittings there the most likely to weep.

Water at the bottom of a rear door can be a leak or condensation. Checking the supply water temperature against room conditions usually tells which.

Coolant in a drip tray or under a rack means the loop has leaked, even if no alarm fired. Any coolant in a tray is a finding.

Leak sensors only report where they are placed and only when liquid reaches them, as covered in liquid cooling leak detection. A small weep can sit in a tray for days without touching a sensor.

Clear PG25 can be hard to see on a dark tray. A sticky film is a sign, because glycol stays tacky long after water would have dried.

Trays and containment are the evidence store of a liquid cooled hall. Routine visual inspection of every tray, not just a check of the leak alarms, is how small leaks are caught. Reading residue over time is covered in Part 4.

Coolant chemistry turns into a leak problem because degraded coolant corrodes the metal it is meant to protect. The leak comes later, but the cause is chemical.

Glycol breaks down over time and with heat into organic acids. That lowers pH and consumes inhibitors.

Corrosion then attacks thin walled cold plates and fittings. Corrosion debris fouls quick disconnect faces and filters, which leads to the connector failures described above.

Discoloration in a sight glass is a visible sign that chemistry has drifted. Coolant that has darkened or clouded is worth sampling before anything leaks.

How often should coolant be tested?

Test coolant on the schedule set by the OCP guideline and the fluid supplier. Test again after any event that could change it: a large top up, a facility water crossover or a repair. Testing pH and inhibitor levels is what turns chemistry from a surprise into a trend.

Where coolant lines run matters because a leak's consequence depends on what is below it. Coolant lines routed above or beside power distribution units (PDUs), busway and electrical panels turn a small leak into an electrical event.

Liquid cooled racks bring water and high power density into the same space. Routing that seemed acceptable on a drawing can put a manifold drain directly above a power strip.

Recording which lines run above power equipment is part of commissioning. It should be rechecked whenever racks are added or rearranged.

A liquid cooling leak can cost hardware, compute time and safety, often all at once.

In a data center

Coolant on IT hardware can damage servers outright. Loss of TCS pressure and flow leads to thermal throttling or shutdown of GPUs, and lines near power add electrical risk.

In office and mixed use buildings

A small liquid cooled room carries the same risks at smaller scale, often with less monitoring. It may sit outside the facilities team's normal rounds entirely.

Not all liquid near a rack is a loop leak. Several sources look the same at first glance.

• Condensation on cold supply lines and rear doors when the room dew point is high.
• Residue from a past service event rather than an active leak, covered in Part 4.
• Spilled coolant from a top up rather than a system leak.

Condensation forms over cold surfaces and follows room humidity. A loop leak starts at a connection and continues whether or not the room is humid.

This is why no single sensor is enough. Residue, pressure trends, temperature and the service log each rule out a different explanation.

A walk past can prove that coolant has leaked, that a hose is damaged or unsupported, and that coolant chemistry has visibly changed. It cannot prove inhibitor levels or the condition inside the cold plates.

• Directly observable: coolant in drip trays, wet or sticky fittings, kinked, crushed or unsupported hoses, discolored coolant and a low reservoir level.
• Strong indicator: a TCS pressure trend falling between top ups, frequent CDU makeup, and temperature differences between racks on the same loop.
• Suspected: coolant chemistry and inhibitor levels, and internal cold plate corrosion. Those need a fluid sample sent for lab analysis.

That is the limit of any route based check. It finds the leak and the damaged hose; a fluid sample confirms the chemistry.

A walkthrough record that captures trays, couplings and CDU readings each round turns these checks into a trend, the core of predictive maintenance.

• Every drip tray and the floor beneath racks and CDUs.
• Each quick disconnect for wetness, residue or partial engagement.
• Hoses for kinks, crushing, rubbing and missing support.
• Rear door hinge side fittings.
• CDU reservoir level and pressure displays.
• Coolant color in any sight glass.
• Any coolant line routed directly above electrical equipment.

Coming Soon

1. 1. When was your TCS coolant last sampled, and what were the pH and inhibitor results?
2. 2. How often does each CDU need topping up?
3. 3. Who inspects quick disconnects after servers are swapped?
4. 4. Which coolant lines run above power equipment?
5. 5. Is anyone assigned to check drip trays, or only the leak alarms?

Where do liquid cooling systems leak most often?

Leaks concentrate at connections: quick disconnects, manifold ports, hose fittings, CDU service points and rear door hinge side fittings. Anything opened for service or disturbed during a server swap is most at risk.

What is PG25 coolant?

PG25 is inhibited propylene glycol at about 25% concentration in water, used in the technology cooling system of liquid cooled racks. OCP publishes guidelines for its use, most recently Revision 2.0 dated September 15, 2026.

Why is propylene glycol used in data center liquid cooling?

At PG25 concentrations propylene glycol is biostatic, so it resists microbial growth, and it adds freeze protection. With an inhibitor package, it also protects the metals in the loop.

What is the difference between the FWS and the TCS?

The facility water system (FWS) is the building side that supplies the CDU. The technology cooling system (TCS) is the secondary, closed loop from the CDU to the cold plates on the chips.

How do quick disconnects fail?

Common causes are debris on the mating faces, damaged or swollen O-rings, partial engagement, side load from heavy hoses and blind mate misalignment. Dry break designs should not drip at disconnect, so a drip means wear or damage.

What does coolant in a drip tray mean if no alarm went off?

It means the loop has leaked. Leak sensors only report where liquid reaches them, so a small weep can sit in a tray for days unalarmed. Treat any coolant in a tray as a finding and trace it to its source.

Why does coolant chemistry matter for leaks?

Glycol breaks down into organic acids that lower pH and use up inhibitors. The coolant then corrodes thin cold plates and fittings, and corrosion debris fouls quick disconnects and filters.

Do rear door heat exchangers leak?

Yes. Coils can leak at brazed joints and return bends, and fittings on the hinge side weep because hoses flex every time the door opens. Rear doors can also drip condensation if supply water runs below the room dew point.

How often should liquid cooling coolant be tested?

Follow the schedule in the OCP guideline and the fluid supplier's recommendations. Also test after a large top up, a repair or any sign of discoloration, and track pH and inhibitor levels over time.

• Liquid Cooling Leak Detection: What Sensors Miss and Inspection Catches
• Data Center Commissioning Checklist: What to Inspect Before Handover, Trade by Trade
• Pump Seal Leaks: How to Tell an Acceptable Weep From a Failing Mechanical Seal or Packing
• Why Pipe Joints Leak: Flanges, Threads, Unions, Welds, Valve Bonnets and Expansion Joints

• OCP: PG25 guidelines, Revision 2.0 (September 15, 2026)
• OCP Cold Plate sub-project: UQD and BMQC specifications
• Earlier OCP PG25 draft: concentration and freeze point