Liquid cooling for data centers is the use of a circulating fluid, rather than air, to carry heat away from processors at the chip. As server power densities rise, cloud service providers and equipment manufacturers are adopting it to manage the thermal loads of high-performance processors and densely packed computer hardware.
In glycol-based cooling loops, coolant condition must be monitored to help maintain system reliability, corrosion protection, and stable thermal performance. pH and conductivity are two key parameters used to track coolant health in these systems.
Data centers play an increasingly important role in the global digital infrastructure landscape. Cloud Service Providers (CSPs) such as Google, Microsoft, Amazon, Meta, Alibaba, and Tencent are investing in hyperscale facilities to support growing AI and computing demands.
Modern AI chips now produce over 1,000 W of heat each, pushing server rack densities to 100 kW or more. Graphics Processing Units (GPUs) are being packed more densely than ever, creating a thermal challenge that often exceeds the practical limits of conventional air cooling. To overcome this, the industry is rapidly shifting to liquid cooling. Within this transition, single-phase Direct Liquid Cooling (DLC) is becoming standard for high-density server deployment.
Understanding Power Usage Effectiveness
Before comparing air cooling and liquid cooling, Power Usage Effectiveness (PUE) is an important term to understand. By definition, PUE is the industry-standard metric for measuring data center energy efficiency. It is calculated by dividing the total facility power by the power consumed by IT equipment (servers, storage, and networking).
PUE = Total Facility Energy / IT Equipment Energy
For example, if a data center consumes a total of 150 kW of power and the IT equipment consumes 100 kW, the PUE would be:
PUE = 150 kW/100 kW = 1.5.
An ideal PUE is 1.0, meaning every watt entering the building is used directly for computing. Anything above 1.0 represents “overhead” spent on infrastructure like cooling, lighting, and power distribution. Therefore, a PUE of 1.5 means that for every 1.0 kW used by servers, an additional 0.5 kW is used on support systems. Because data centers consume so much energy, CSPs must obtain approval to build them. Authorities have set regulations requiring low PUEs. In a traditional air-cooled data center, cooling typically consumes 30% to 40% of the total facility power. To have a lower PUE, CSPs are switching to liquid cooling.
Air cooling uses fans to pass ambient air over components. Because air has a low volumetric heat capacity, this approach is less efficient and often results in PUE values in the 1.4–1.6 range. In contrast, liquid cooling uses fluids with nearly 3,500 times the heat-carrying capacity of air. Capturing heat directly at the chip via cold plates eliminates the need for power-hungry fans and room-scale refrigeration. This improvement in efficiency significantly reduces cooling overhead, allowing for high server density, and driving PUE down to a range of 1.2 or lower, reducing both energy costs and the carbon footprint. While liquid cooling improves PUE, it also introduces new operational and material risks that require robust coolant monitoring.
The transition to liquid cooling introduces significant physical and operational risks, primarily centered on coolant leakage and its potential to cause substantial hardware failure.
Mechanical failure often occurs at quick-disconnect points due to improper seating or worn O-rings. Additionally, constant vibration from high-power pumps can cause fatigue cracking in manifolds and thin-walled cold plates.
Beyond physical wear, coolant degradation is a risk that must be addressed. Glycol-based coolant requires careful monitoring to prevent corrosion and damage to server hardware.
Real-time fluid monitoring should be treated as an integral part of the cooling loop architecture rather than an add-on.
Open Compute Project (OCP) guidelines identify pH, conductivity, 25% propylene glycol (PG25) concentration, flow, temperature, and particulate monitoring as a practical foundation.
As such, pH and conductivity sensors are commonly installed within the Cooling Distribution Unit (CDU) as the primary means of monitoring coolant health. For improved reliability, redundant sensors are recommended at critical control points.
In September 2025, NVIDIA released CDU requirements that included recommendations for monitoring coolant quality. The requirements identify pH, conductivity, and turbidity as three key parameters for coolant quality monitoring.
Glycol-based coolants are typically formulated with anti-corrosion salt-based inhibitors as key ingredients. Without them, the coolants will severely damage the cooling system through a process called electrochemical corrosion. Inhibitors in coolants protect the cooling systems through two main mechanisms:
Creates a film on metal surfaces, acting as a physical shield that prevents oxygen and water from causing oxidation.
Alkaline buffers neutralize organic acids that naturally form as glycol degrades over time, keeping the coolant stable and non-corrosive.
The InPro™ 4260i pH sensor is designed for use in coolant applications. With sensors that use conventional ceramic diaphragms, the small pores can become clogged rapidly from fine particles released by components. This causes unreliable measurements because contamination of the diaphragm results in undesirable diffusion potentials that distort readings.
The InPro 4260i features Xerolyt™, a solid polymer electrolyte that eliminates the need for a diaphragm. Instead, the reference system connects with the measuring medium via an open junction. The extremely fine particles that deposit in the junctions can then be rinsed away, either by the flow of liquid in the process or periodic rinsing.
UniCond™ conductivity sensors, with a built-in measuring circuit and temperature sensor, are well suited for glycol-based coolant systems that require accurate conductivity and temperature measurement. Further, METTLER TOLEDO combines accepted, standardized mechanical configurations with temperature-compensation algorithms for conductivity measurement, offering up to 10 preset algorithms, including one for ethylene glycol.
Both 2-electrode and 4-electrode UniCond conductivity sensors are suitable for glycol-based coolant applications, depending on the specific design preferences and requirements of the cooling loop.
Regarding process connections, threadless conductivity sensors are available to address concerns related to cross-threading, leakage, dripping, and the risk of system contamination from thread sealants or PTFE tape.
In such cases, a conductivity sensor with a Tri-Clamp is an alternative, providing a leak resistant, and easy-to-install solution that addresses these concerns while maintaining measurement performance.
Both pH and conductivity sensors mentioned are digital and equipped with METTLER TOLEDO’s Intelligent Sensor Management (ISM™) technology. ISM’s Plug and Measure feature ensures fast, error-free startup, and greater measurement reliability.
In addition, ISM offers predictive sensor diagnostics for efficient routine maintenance, which can be scheduled in advance, eliminating unexpected measurement point downtime.
As data centers rely on liquid cooling to support higher heat loads and greater rack densities, reliable coolant monitoring becomes a core part of system design. Because these facilities are built to operate with limited on-site intervention, sensor robustness, measurement stability, and maintenance efficiency are critical requirements.
For glycol-based coolants, that means tracking pH against the formulation's alkaline operating range, typically pH 8.0 to 10.5, and trending conductivity against a defined baseline rather than a fixed limit. The InPro 4260i pH sensor and UniCond conductivity sensor cover both parameters, helping operators track coolant condition, support corrosion control, and maintain consistent thermal performance.
Bryant Wai is an industry expert with over a decade of experience in analytical instrumentation and process applications. Additionally, he is central to METTLER TOLEDO's data center cooling segment. Bryant has extensive experience across the entire value chain - working closely with system fabricators, integrators, and end-users. His deep technical proficiency ensures customers receive high-performance measurement solutions.