A data centre is a factory whose product is continuous computation.
Servers turn electricity into useful digital work and heat. Network equipment moves the resulting data. Cooling removes the heat. Backup systems keep the process alive when normal power fails. Water may carry heat away through cooling towers or chilled-water systems. Every extra watt of computing therefore creates another infrastructure demand somewhere else in the building.
This is especially important in Singapore because land, electricity and water are all constrained.
IMDA’s Green Data Centre Roadmap therefore frames data-centre growth as a resource-efficiency problem as well as a digital-economy problem. The roadmap aims to provide at least 300 MW of additional data-centre capacity in the near term, with more growth enabled through green energy and higher efficiency. In August 2026, EDB and IMDA provisionally allocated 200 MW of new capacity to four proposals under the second Data Centre Call for Application.
The operating chain is: grid power enters → switchgear distributes → UPS bridges disturbances → generators provide longer backup → power distribution feeds IT racks → servers produce heat → air or liquid cooling removes heat → cooling plant rejects heat outdoors → water and energy use are monitored → capacity and redundancy are kept inside operating limits.
Quick answer: what does a data centre have to keep balanced?
- IT power: enough electricity for servers, storage and network equipment.
- Cooling: enough heat-removal capacity to keep equipment within safe operating limits.
- Water: where evaporative or water-cooled systems are used, enough water with acceptable efficiency and quality.
- Electrical redundancy: alternative source paths when equipment is unavailable or maintenance is occurring.
- Thermal redundancy: spare cooling capacity and alternative equipment paths.
- Space: sufficient room for racks, electrical systems, cooling plant and maintenance access.
- Connectivity: diverse network links so the compute remains reachable.
1. The server is an electrical load first
Every processor, memory module, storage device and network port consumes electricity.
AI accelerators can make rack power density much higher than traditional enterprise computing.
The data-centre building therefore has to be designed around electrical capacity before the customer installs the compute.
A rack with no usable power allocation is empty real estate, however advanced the server hardware might be.
2. Electricity becomes heat almost immediately
Most electricity used by IT equipment eventually becomes heat inside the data hall.
A one-megawatt increase in IT load therefore creates approximately another megawatt-scale heat-removal problem.
Compute capacity and cooling capacity cannot be planned independently.
The data centre is a coupled electrical-thermal system.
3. Utility power enters through controlled electrical infrastructure
Large data centres can connect to Singapore’s grid at high-tension voltage levels appropriate to their contracted demand.
Transformers, switchgear and distribution equipment convert and route that supply toward data halls, cooling plants and essential auxiliary systems.
The electrical distribution tree is designed so maintenance or faults can be isolated without unnecessarily removing every load at once.
4. UPS systems protect the no-break interval
Servers cannot wait several seconds while a standby generator starts.
Uninterruptible Power Supplies use stored energy—commonly batteries—to keep critical IT load powered through short utility disturbances and generator start-up intervals.
The UPS therefore acts as a fast bridge between normal and emergency power.
5. Generators provide longer-duration backup
Standby generators can support the facility for longer periods than ordinary UPS batteries, provided fuel and mechanical systems remain available.
The data centre therefore layers fast finite storage with slower but longer-duration fuel-based backup.
Continuous availability comes from the handoff between systems rather than one perfect source.
6. Redundancy is useful only when failure paths are truly independent
Two UPS units connected to one common switch can still share a critical failure point.
Two network carriers using one common duct may not provide full route diversity.
Resilience design therefore maps dependencies, not just equipment counts.
The question is not “How many backups exist?” but “Which single faults can still remove all of them together?”
7. Air cooling begins with controlling where hot and cold air go
Servers draw cooler air through their fronts and discharge hotter air behind.
If hot exhaust mixes freely with supply air, cooling systems have to work harder and equipment can receive uneven inlet temperatures.
Hot-aisle and cold-aisle arrangements, containment and careful airflow management reduce that mixing so the same cooling plant removes heat more effectively.
8. Cooling can consume up to 40% of total data-centre energy
IMDA’s Tropical Data Centre Standard notes that cooling systems can account for up to 40% of a data centre’s total energy use.
That makes cooling efficiency one of the largest non-IT opportunities available.
A small improvement repeated across megawatts of cooling load can create large annual electricity savings.
9. Singapore’s tropical standard challenges unnecessary overcooling
Data centres historically operated around very low temperatures to create a wide safety margin for IT equipment.
Singapore Standard SS 697:2023 provides a method for safely raising operating temperatures toward 26°C and above where equipment and risk assessment permit.
IMDA states that every 1°C increase can produce roughly 2% to 5% cooling-energy savings under the standard’s framework.
The efficiency comes from reducing the temperature difference the cooling plant has to create.
10. Higher rack density is pushing more liquid cooling
Air carries less heat per unit volume than liquid coolants.
As AI and high-performance computing increase rack density, direct-to-chip liquid cooling and other liquid-based designs can move heat more efficiently from processors into facility cooling systems.
The transition does not eliminate pumps, heat exchangers or heat rejection. It moves the thermal interface closer to the electronics producing the heat.
11. PUE measures how much facility energy surrounds the IT load
Power Usage Effectiveness compares total data-centre energy use with energy used by IT equipment.
A PUE closer to 1 means less additional facility energy is being spent on cooling, power conversion, lighting and other overheads for each unit of IT energy.
PUE does not tell us whether the computing workload itself is efficient. It measures facility overhead around that workload.
12. IT equipment itself is now an efficiency target
IMDA states that IT equipment accounts for around 60% of total energy use in a typical data centre.
Singapore Standard SS 715:2025 therefore focuses on more efficient servers and IT operation, including workload consolidation and virtualisation.
IMDA says the standard aims for at least 30% energy savings at the IT equipment level through efficient hardware selection and management.
The greenest cooling plant cannot compensate indefinitely for poorly utilised compute.
13. Water can be a cooling resource
Many cooling systems use cooling towers to reject heat through evaporation.
That can reduce electricity consumption compared with some purely air-cooled arrangements but consumes water.
Singapore therefore has to optimise energy and water together rather than reducing one resource metric while ignoring the other.
14. Water Usage Effectiveness creates a second resource metric
WUE compares annual water consumption with IT energy delivered.
The Green Data Centre Roadmap states that the 2021 median WUE among large Singapore data-centre water users was about 2.2 cubic metres per MWh.
Singapore is working with PUB and the industry toward 2.0 m³/MWh or lower over the coming decade for new and existing facilities where feasible.
Water efficiency therefore becomes measurable rather than an environmental footnote.
15. Alternate water can reduce potable-water demand
Singapore’s Green Mark criteria recognise alternate cooling-tower makeup sources such as NEWater, rainwater or air-handling condensate where suitable.
The objective is to avoid using high-grade potable water when another treated or recovered water stream can perform the cooling job safely.
Fit-for-purpose water use therefore becomes part of digital infrastructure design.
16. Continuous availability requires maintainability
A data centre designed never to stop still needs equipment replacement.
UPS modules age. Batteries degrade. chillers need overhaul. switchgear needs inspection. filters clog. pumps require maintenance.
Redundant architecture allows one component to be isolated for planned work while remaining equipment keeps the service operating.
Maintainability is therefore part of availability, not a separate facilities concern.
17. Singapore is adding capacity selectively rather than returning to unconstrained growth
The Green Data Centre Roadmap aims to support at least 300 MW of additional capacity in the near term while raising requirements for energy efficiency and green-energy use.
In August 2026, EDB and IMDA selected four proposals under DC-CFA2 for provisional allocation of 200 MW.
The capacity-allocation process therefore treats megawatts as a scarce national resource tied to economic value and sustainability performance, not simply a property-development entitlement.
18. A worked example: new AI cluster enters a data hall
Imagine a customer installs a dense GPU cluster.
Facility planners verify that the electrical distribution can supply the new rack density. UPS and generator capacity are checked against the higher critical load. Liquid or high-capacity air cooling removes the added heat. Water and cooling-plant impacts are included in the operating plan. The building-management system monitors electrical loading, temperatures and equipment conditions so the additional compute remains inside both power and thermal limits.
The servers arrived in boxes. The facility had to create an invisible megawatt-scale ecosystem around them before they became useful.
19. Common misconceptions
Misconception: Data centres mainly consume electricity for cooling.
No. IT equipment itself is typically the largest energy load; cooling can account for up to about 40% depending on facility design.
Misconception: Colder data halls are always safer.
No. unnecessarily low temperatures consume more cooling energy; Singapore’s tropical standard provides a controlled route toward higher operating temperatures where equipment permits.
Misconception: A low PUE proves the servers are efficient.
No. PUE measures facility overhead relative to IT energy, not whether the workloads or servers themselves are well utilised.
Misconception: Water-free cooling is automatically environmentally superior.
No. designs must compare energy, water, climate and system efficiency together.
Misconception: Redundancy means equipment can be ignored until failure.
No. continuous availability depends on planned maintenance, testing and replacement while alternate capacity carries the load.
20. The deeper idea: a data centre is a machine for converting national resources into digital capability
The visible product is cloud computing, AI, banking, streaming and digital services.
The physical inputs are less glamorous: megawatts of electricity, cooling plant, water, land, fibre, switchgear, batteries and maintenance access.
Singapore’s data-centre strategy becomes sustainable only when those inputs are managed as carefully as the compute output.
The question is no longer merely how many servers fit in a building.
It is how much useful computation the city can produce from each scarce unit of power, water and space while keeping the service continuously available.