A liquid-cooled AI hall is commissioned with a coolant that passes every lab test. The fluid report goes into the handover file, and the next sample is scheduled for three months later. In the weeks between, a technician tops up the loop with plain water after replacing a quick disconnect, a small amount of facility water seeps through a plate heat exchanger, and fine debris left over from installation keeps circulating through the cold plates. None of it shows up on a temperature or flow screen until thermal performance starts to slip.
This is the gap coolant chemistry monitoring is meant to close. The question for most operators and CDU integrators is no longer whether to monitor coolant chemistry, but how: which events each approach will actually catch, and what form the monitoring hardware should take inside a CDU or facility control system.
The short answer is this. Periodic lab sampling confirms a fluid’s condition at one moment, but misses fast events between samples. A single inline sensor catches some events and misses others, because each failure mode shows up in a different parameter. Continuous multi-parameter monitoring of glycol concentration, pH, conductivity, turbidity and temperature, wired into the CDU controller or BMS, is the only approach that sees the full range of events as they happen.
Why coolant does not stay the way it was commissioned
A technology cooling loop is closed, but it is not static. Maintenance introduces water and air. Quick disconnects weep and get topped up. Inhibitors are consumed as they protect metal surfaces. Glycol slowly oxidises under heat. Particles from fabrication and installation break loose over the first months of operation. Any leak in the CDU heat exchanger brings facility water chemistry into a loop that was never designed for it.
Each of these events changes the fluid in a specific way, and each leaves its mark on a different measurement. That is why the choice of monitoring approach matters as much as the decision to monitor at all. For background on how the fluid choice itself (PG25, DI water or immersion) shapes these risks, see our earlier article on choosing a CDU coolant and why monitoring decides whether it works.
Which events each monitoring approach catches
The table below maps common coolant events against three approaches: quarterly lab sampling, a single inline conductivity sensor, and continuous multi-parameter monitoring.
| Event | What changes in the fluid | Quarterly lab sampling | Conductivity sensor only | Multi-parameter inline |
|---|---|---|---|---|
| Water top-up dilutes glycol | Glycol concentration falls; inhibitor strength falls with it | Caught weeks or months later | Often missed (small change) | Caught immediately by glycol measurement |
| Heat exchanger leak from facility side | Conductivity steps up; hardness and chlorides enter | Caught late, after exposure | Caught | Caught, and confirmed by pH and turbidity |
| Glycol oxidation and inhibitor depletion | pH falls gradually as organic acids form | Caught as a trend over several samples | Weak signal, easy to miss | Caught early by pH trend |
| Particulate release or fouling | Turbidity rises; cold plate flow can be restricted | Caught only if a particle count is requested | Missed | Caught by turbidity |
| Microbial growth | Turbidity rises; pH may shift | Caught if culture testing is included | Missed | Early indication via turbidity and pH |
The pattern is clear from the table. No single parameter catches everything, and sampling intervals measured in months cannot respond to events that develop in days. A loop can show acceptable conductivity while its glycol is diluted, its pH is falling, or particles are building up in the cold plates.
Why turbidity deserves more attention than it gets
Turbidity is the measurement most often left out of coolant monitoring, largely because conventional cooling water instrumentation was designed for systems with wide flow passages. AI cooling hardware is different. Cold plates and microchannel heat exchangers have passages so narrow that small amounts of particulate can reduce flow at the chip.
Rising turbidity is often the first sign of corrosion products, installation debris, biological growth or precipitated inhibitor. It also tells an operator when filtration is not keeping up. Placing the turbidity measurement where it sees the fluid before the CDU’s filter shows what the loop is generating, while a reading after the filter confirms what is reaching the IT equipment.
Embedded module or complete analyser
Pyxis Lab’s KRYOPTIX IK-Series packages the five key coolant measurements (glycol concentration, pH, conductivity, turbidity and temperature) in two formats. Which one fits depends on who will be looking at the data and where.
IK-1500 Series: the embedded chemistry module. This version has no local display or operator interface. It is designed to sit inside or alongside a CDU and transmit its measurements to a control system that already exists, such as the CDU controller, a PLC, a DCS, a BMS or OEM cooling equipment. It suits CDU manufacturers building chemistry monitoring into their product, system integrators delivering factory-integrated cooling skids, and hyperscale sites where all data is already centralised and a local screen would go unused.
IK-1600 Series: the complete analyser. This version measures the same parameters and adds the UC-80-PLUS industrial touchscreen terminal, which provides live display, historical trending, data logging, a calibration interface, sensor diagnostics, alarm management and local configuration. It still connects to PLC, DCS, CDU controller, BMS, EPMS and DCIM platforms through Modbus RTU and Modbus TCP. It suits colocation operators and enterprise facilities retrofitting monitoring onto existing CDUs, sites where maintenance staff need to see and calibrate at the equipment, and any installation where a local data record is valuable during commissioning and troubleshooting.
A practical way to decide is to ask where the technician will be standing when an alarm goes off. If the answer is a central control room with a screen already showing CDU status, the embedded module is the leaner choice. If the answer is beside the CDU with a calibration buffer in hand, the complete analyser saves time on every visit.
Getting value from the data, not just collecting it
Continuous measurement only pays off if it drives action. A few design choices make the difference.
Alarm on rate of change, not only on limits. A heat exchanger leak shows as a sudden step in conductivity that may still sit inside the normal band. A rate-of-change alarm catches it on day one, while a fixed upper limit may not trip until contamination is well established.
Baseline during commissioning. Record every parameter after the final fill and flush. The values that matter later are deviations from this baseline, because every fluid formulation and every loop settles at slightly different values.
Correlate chemistry with operating data. Temperature, flow and pressure data already sit in the CDU controller. Trending chemistry alongside them shows whether a pH change follows a period of high return temperature, or whether a turbidity rise follows a maintenance event, which turns an alarm into a diagnosis.
Install for maintenance. Mount the sensors in a side-stream flow assembly with isolation valves, so they can be removed for cleaning and calibration without draining the loop or interrupting cooling.
Where this leaves operators in Malaysia
Malaysia’s liquid-cooled capacity is growing quickly, much of it in Johor, and much of it built around direct-to-chip loops running PG25 coolant. Many of these loops are being commissioned now, which is the cheapest point to build chemistry monitoring into the CDU design rather than retrofit it after the first fouling event. For CDU integrators and OEMs supplying the Malaysian market, an embedded module adds a differentiating capability to the product. For operators, a complete analyser provides the evidence that the cooling loop is being maintained to specification.
Autoflo Technology is the Malaysian distributor for Pyxis Lab, including the KRYOPTIX IK-1500 and IK-1600 coolant chemistry monitoring platforms. For help specifying coolant monitoring for a CDU design or an existing liquid-cooled facility, contact us at info@autoflotechnology.com.