As AI computing continues to expand, data centres are deploying increasingly dense GPU infrastructure that generates significant amounts of heat. Cold plate liquid cooling provides an effective way to remove this heat, but the cooling loop itself requires careful monitoring. Poor coolant quality can contribute to corrosion, deposits and clogging, potentially affecting cooling performance and equipment reliability.
For a Silicon Valley AI cloud provider operating 3,200 GPU cabinets, Rika Sensor implemented a liquid cooling water quality monitoring solution designed to provide continuous visibility into key coolant parameters. The project supported a 99.99% SLA requirement and integrated monitoring data with the data centre’s existing DCIM infrastructure.
The solution combined pH, conductivity, ORP and turbidity measurement to help the operator identify changes in coolant quality before they could develop into more serious operational problems.
Project Background
The data centre uses a cold plate liquid cooling system to manage the heat generated by high-density GPU cabinets. Unlike conventional air cooling, cold plate systems circulate liquid directly through cooling components positioned close to heat-generating hardware.
This approach can provide efficient heat removal, but the cooling liquid must remain within suitable quality conditions. Changes in pH, conductivity, oxidation-reduction potential (ORP) or turbidity may indicate changes in the condition of the coolant or cooling circuit.
With thousands of GPU cabinets operating within a high-availability environment, the customer required a monitoring approach capable of supporting continuous operation rather than relying solely on periodic manual testing.
Challenge
The cooling system presented several monitoring challenges.
Narrow Micro-Channels
The cold plates contained narrow micro-channels. These small flow paths can be particularly vulnerable to contamination, deposits and other water-quality-related problems. A deterioration that is detected too late may affect coolant circulation and heat transfer.
Weekly Manual Sampling
The existing approach relied on weekly manual sampling. While laboratory or handheld testing can provide useful measurements, sampling at fixed intervals does not provide continuous visibility. A change occurring between two sampling periods could remain undetected for several days.
Distributed Facility
The third challenge was the physical scale of the facility. The data centre contained three geographically dispersed data centre rooms, making centralised monitoring and management important for the operations team.
Monitoring Requirement
The customer required a liquid cooling water quality sensor solution capable of continuously monitoring several important parameters. The monitoring system included:
● pH monitoring
● Conductivity monitoring
● ORP monitoring
● Turbidity monitoring
These measurements created a broader picture of coolant condition than relying on a single parameter. The system was also designed as a data centre coolant water quality sensor solution that could communicate with the customer’s existing monitoring infrastructure.
Solution – Deployment & Commissioning
Rika Sensor installed pH, conductivity, ORP and turbidity sensors at strategic locations throughout the cooling system. The sensors used 316L construction and G3/4 threads, making them suitable for integration into the customer’s liquid cooling piping arrangement.
Monitoring points were positioned at CDU inlets and outlets as well as main pipe branches. This allowed the operator to compare coolant conditions at different stages of the cooling circuit rather than relying on measurements from a single location.
For system communication, the sensors transmitted data through RS485 using Modbus RTU to the existing DCIM platform. This approach allowed water-quality information to become part of the customer’s central data centre monitoring environment.
The project was implemented in two phases. Phase I included 42 monitoring points, while Phase II expanded the deployment to 128 points as monitoring requirements increased.
During commissioning, the engineering team identified an issue involving air pockets in sections of the pipework. These air pockets caused fluctuations in sensor readings. Instead of treating the unstable readings as a sensor fault, the installation arrangement was reviewed. The affected sensors were relocated and additional vent valves were introduced to help remove trapped air and stabilise measurement conditions.
Rated Products
The monitoring system can incorporate dedicated sensors for different liquid cooling parameters, including Rika Sensor’s liquid cooling EC/conductivity sensor, liquid cooling pH sensor, liquid cooling ORP sensor and liquid cooling turbidity sensor.
These products provide dedicated measurement capabilities for liquid coolant quality monitoring, allowing operators to build a monitoring system according to the requirements of their cooling infrastructure.
Results
Following commissioning and go-live, the customer gained continuous visibility into key coolant quality parameters across the cooling network. The expanded monitoring coverage also made it easier to manage water-quality information across the three geographically separated data centre rooms.
One practical improvement was the reduction in manual sampling frequency. Instead of collecting samples weekly, the customer reduced manual sampling to monthly while maintaining continuous online monitoring between sampling intervals.
The project demonstrates how a dedicated coolant water quality sensor system can complement existing DCIM infrastructure in large-scale AI data centres. By continuously monitoring pH, conductivity, ORP and turbidity, operators can identify water-quality changes earlier, support preventive maintenance and improve visibility across complex liquid cooling networks.
For high-density AI infrastructure, where cooling reliability directly supports computing availability, continuous CDU water quality sensor monitoring provides an important layer of operational control and data-driven maintenance.
