Singapore’s New SS 726:2026: An AI-Ready Liquid-Cooling Maintenance Playbook for Data Centres

A practical Singapore guide to sensors, leak detection, water quality, alarms and AI-assisted maintenance for high-density cooling systems.

Singapore launched SS 726:2026 on 27 August 2026, described by Enterprise Singapore and IMDA as the world’s first standard for liquid cooling in tropical data centres. As artificial intelligence workloads increase rack density and cooling demand, the standard gives data-centre owners, operators and facility teams a timely national reference point.

The practical question for facilities teams is not only whether liquid cooling has been adopted. It is whether the installation can be monitored, maintained and operated with enough visibility to protect availability, equipment condition, energy performance and workplace safety.

This article translates the topic into an engineering and facility-management playbook. It is not a substitute for reviewing the official SS 726:2026 document, project specifications or equipment manufacturers’ requirements. Instead, it highlights the operational data and maintenance controls that should be considered when preparing a liquid-cooled environment.

Why liquid cooling changes the maintenance task

Traditional air-based cooling relies mainly on airflow, temperature, pressure and chilled-water or refrigerant-system performance. Liquid cooling introduces additional interfaces between the cooling distribution system and IT equipment. These may include coolant distribution units, secondary loops, hoses, manifolds, quick-connects, heat exchangers, pumps, valves and leak-detection provisions.

Each interface creates a condition that should be visible to the operations team. A small change in flow, pressure, temperature, conductivity or moisture may provide an earlier warning than a server alarm or a visible equipment failure. In a tropical environment, the wider facility still needs to manage humidity, condensation risk, drainage, access, electrical safety and the reliability of supporting ACMV systems.

Liquid cooling should therefore be treated as an integrated asset-management issue rather than an isolated IT equipment upgrade.

1. Establish a cooling-loop asset register

Begin with a clear asset register and system map. Record the location, function and criticality of each major component, including pumps, CDUs, heat exchangers, control valves, strainers, filters, manifolds, flexible connections and leak-detection devices.

The register should link physical assets to:

• Equipment identification and service area
• Cooling-loop or circuit designation
• Normal operating range
• Isolation points and maintenance access
• Manufacturer instructions and warranty conditions
• Inspection, testing and replacement intervals
• Alarm points and escalation contacts

This information is important during planned work and incident response. A technician should be able to identify which equipment is affected, what can be isolated, which downstream loads may be exposed and which alternative cooling arrangements are available.

2. Define the right sensor points

Sensor quantity alone does not create useful monitoring. The objective is to place reliable measurement points where they can show developing faults and support decisions.

Depending on the system design, facilities teams should consider monitoring supply and return temperatures, differential pressure, flow, pump status, valve position, filter or strainer condition, reservoir level where applicable, and heat-exchanger performance. Sensors should also be considered around high-risk connection points and areas where leaks could reach electrical or IT equipment.

Data should be time-stamped and associated with the correct asset. Trend views are more useful when operators can compare current readings with commissioning values, operating modes, maintenance events and changes in IT load.

Before relying on any sensor, confirm its installation quality, calibration needs, communication status and failure response. A missing or implausible value should create a data-quality alert rather than silently entering an AI model as if it were normal.

3. Build a layered leak and moisture strategy

Leak detection should not depend on a single device. A practical strategy can combine point sensors, sensing cables, drip trays, visual inspection, pressure or flow deviation, and alerts from connected equipment.

Detection zones should be mapped to physical locations so that an alarm identifies where technicians should respond. The response plan should define who receives the alarm, how quickly it must be acknowledged, whether equipment or pumps should be isolated, and how affected areas are inspected before restart.

Moisture detection should also be coordinated with drainage, containment and electrical-risk controls. Teams should understand the difference between an alarm that indicates confirmed water and one that indicates an abnormal operating condition requiring investigation. Testing should be scheduled and documented rather than assumed to be effective because the system is connected.

4. Monitor water or coolant quality

Where the design uses a water-based or treated liquid loop, coolant quality can affect corrosion, fouling, microbial activity, heat transfer and component life. The exact control limits should follow the approved system design, equipment manufacturer guidance and project requirements.

A maintenance programme may include checks such as conductivity, pH, temperature, pressure, flow, filtration condition and signs of contamination. Sampling frequency should reflect the system’s risk, size, operating history and any treatment programme. Results should be recorded against the relevant loop and reviewed for trends.

A single result outside an expected range should not automatically trigger a major intervention. However, repeated drift, unexplained changes after maintenance or disagreement between instruments should prompt engineering review. Water-quality records should be connected to work orders so teams can relate chemical or filtration changes to later equipment behaviour.

5. Link alarms to clear escalation actions

Alarm flooding can be as dangerous as a lack of alarms. Facilities teams should classify liquid-cooling alarms by consequence and required response. Examples may include sensor communication loss, abnormal temperature difference, low flow, high differential pressure, pump failure, leak detection, water-quality deviation and loss of control-system connectivity.

Each alarm should have an owner, an acknowledgement target, an investigation procedure and an escalation path. Critical alarms may require immediate coordination between the data-centre operations team, facilities personnel, IT owners and specialist contractors. Lower-priority alerts may be reviewed during a planned maintenance window, provided the risk is understood and documented.

Alarm logic should be tested during commissioning, planned maintenance and system changes. Operators should also be trained on manual fallback actions if the BMS, monitoring platform or network connection is unavailable.

6. Make maintenance records AI-ready

AI-assisted fault prediction depends on consistent operational data. A facility may have many sensors but still lack usable intelligence if asset names, alarm descriptions, timestamps and work-order records are inconsistent.

At minimum, maintenance records should capture the asset, symptom, alarm or trend involved, inspection findings, corrective action, parts used, operating condition and return-to-service result. Planned and reactive tasks should be distinguished. Repeated failures should be coded using a common failure-mode library, such as leakage, blockage, sensor drift, pump degradation, valve failure, communication loss or water-quality deviation.

Data governance also matters. Access rights, retention, cybersecurity, backup arrangements and human approval should be defined before automation is introduced. AI outputs should support engineering judgement, not replace it. A prediction is a prompt to investigate; it is not proof that a component has failed.

7. Use AI as a maintenance assistant

Once the data foundation is reliable, AI tools can help identify unusual combinations and gradual changes that may be difficult to see in individual readings. Possible applications include detecting abnormal pump performance, comparing similar cooling circuits, identifying repeated nuisance alarms, prioritising work orders and predicting when inspection may be warranted.

Useful workflows may connect BMS or DCIM data with computerised maintenance-management-system records, inspection forms and asset history. For example, a rising differential pressure trend combined with reduced flow and repeated filter alarms could create a recommended inspection task. A model should also show the data and reasoning behind its recommendation, allowing a competent person to validate the next step.

Start with a controlled pilot on a defined group of assets. Establish a baseline, agree what constitutes an actionable alert, measure false alarms and review results with the operations team. Expand only when the workflow improves response quality without creating unnecessary intervention.

8. Design for maintainability and safe access

Maintenance planning should begin before installation. BCA’s work on Design for Maintainability highlights the value of considering access, operational requirements and lifecycle effort upstream. For liquid cooling, this includes safe access to connections, valves, filters, sensors and isolation points; adequate space for inspection; clear labelling; drainage and containment; and practical routes for replacement components.

Digital permits, environmental sensors and remote monitoring may support safer work planning, consistent with Singapore’s wider interest in WSH technology. These tools should complement, not replace, risk assessment, isolation procedures, competent supervision and site-specific controls.

A practical starting checklist

1. Review the official SS 726:2026 document and applicable project requirements.
2. Map every cooling loop, major component, sensor and isolation point.
3. Identify leak, moisture, flow, pressure, temperature and coolant-quality monitoring gaps.
4. Standardise alarm names, priorities, owners and escalation procedures.
5. Link asset data, maintenance records and failure-mode codes.
6. Test sensors, alarms, communications and manual fallback procedures.
7. Pilot AI-assisted trend and fault detection on a controlled asset group.
8. Review findings with engineering, facilities, IT and safety stakeholders.

Conclusion

SS 726:2026 provides Singapore’s data-centre sector with an important reference point as liquid cooling supports higher-density AI workloads. The operational advantage will depend on more than the cooling technology itself. It will depend on visible system data, disciplined maintenance, reliable alarms, maintainable design and well-governed automation.

ISS can support organisations reviewing engineering, facility-management and AI automation requirements for complex cooling environments. Contact ISS to discuss how your facility can improve cooling visibility, maintenance workflows and operational decision-making.