Why Does Cooling Use So Much Power in a Data Centre?

Why does cooling use so much power in a data centre? Almost every watt a server draws is converted straight into heat, and that heat has to be pumped out of the building continuously or the equipment throttles itself, fails early or shuts down altogether. Industry estimates commonly put cooling at roughly a quarter to 40 per cent of total facility power in older, air cooled halls, which is the single biggest reason a data centre needs an electrical supply so much larger than its IT load alone. Newer liquid cooled designs cut that overhead sharply, but cooling remains the largest controllable share of a site's Power Usage Effectiveness.

⚡ Quick Answers: Data Centre Cooling and Power

• What actually uses the power in cooling? Chillers, pumps, computer room air handlers and cooling towers running continuously to move heat out of the building and the racks inside it.
• How much of a data centre's power goes to cooling? Industry estimates put it at roughly a quarter to 40 per cent of total power in older air cooled facilities, falling sharply in modern liquid cooled designs.
• Does Sydney's climate make cooling harder? Yes. Humid summers limit how many hours a year free air cooling alone can carry the load compared with cooler, drier climates, so mechanical cooling still has to be sized for the worst days.
• Is liquid cooling replacing air cooling? It is growing fast for high density AI racks. One 2026 industry survey found 36 per cent of operators had already deployed it, with more than one in four racks now running on liquid cooling.
• Does cooling load affect my electrical supply size? Yes. The switchboard, mains and network application all have to cover the cooling load, not just the IT equipment itself.
• Can an electrician help reduce cooling related costs? We cannot redesign your air handling or chiller plant, but we can measure real demand and make sure your electrical supply, cables and switchboard are sized correctly for the cooling system you actually run.

1. Why every watt becomes a cooling problem

A server does not use electricity so much as convert it. Almost all of the power going into a processor, a power supply or a disk array is turned into heat, with only a small fraction doing useful computing work in any thermodynamic sense. That heat cannot simply be left in the room, because every piece of IT equipment has a maximum safe operating temperature, and running hot shortens component life, increases failure rates and can trigger automatic throttling that slows the workload down.

This is why a data hall is never just a room full of racks. It is a room full of racks paired with an equal and opposite cooling system working around the clock to remove exactly as much heat as the IT load generates. Our guide to why data centres use so much power covers the bigger picture of servers, redundancy and AI density; this guide goes deeper on the cooling half of that equation specifically.

2. Inside a CRAC or CRAH unit

Most traditional data halls cool the room with computer room air conditioning (CRAC) or computer room air handling (CRAH) units, which pull hot air from the racks, pass it across a cooling coil and push cold air back out through a raised floor or overhead ducting. CRAC units have their own built in refrigeration circuit and compressor, while CRAH units rely on chilled water supplied by a separate chiller plant, which is usually more efficient at larger scale because the compression work happens once, centrally, rather than many times over in each unit.

Either way, the electrical load is real and continuous. Fans run constantly to move large volumes of air, compressors or chillers cycle to remove heat, and pumps circulate chilled water where that design is used. None of this is optional or intermittent in the way a home air conditioner might cycle on a mild day. A data hall generates heat every hour it operates, so the cooling plant runs every hour too.

3. Hot aisle and cold aisle containment

A large share of the improvement in modern facilities comes from simply stopping hot and cold air from mixing. Contained hot aisle or cold aisle designs use physical barriers, usually plastic strip curtains or solid panels, so that the cold air delivered to the front of a rack cannot leak into the hot exhaust coming out the back. Without containment, cooling units have to work harder and run colder overall to compensate for air that has already been warmed by mixing, which wastes a meaningful amount of energy for no benefit.

This is a relatively low cost fix compared with replacing a chiller plant, and it is one of the first things an efficiency review usually looks at in an older server room or comms room. A well contained aisle layout can noticeably improve a facility's PUE without touching the cooling equipment itself. Our guide to what PUE is and why it matters explains how that ratio is measured and what counts as good in 2026.

4. Free air cooling in a Sydney climate

Free air cooling, sometimes called economisation, brings outside air straight into the building instead of running mechanical refrigeration, and it works well when the outside air is cool and dry enough on its own. Climates with long, dry, moderate winters can run free air cooling for a large share of the year, cutting chiller energy use dramatically during those months.

Sydney's climate is more of a mixed case. Winters and shoulder seasons are mild enough that free air and evaporative assistance can genuinely do a lot of the work, but Sydney's humid summers limit how far evaporative cooling can push outside air down, since evaporative cooling depends on dry air to work well. That means a Sydney facility still has to provision full mechanical cooling capacity for the hottest, most humid days of the year, even if free air carries much of the load the rest of the time. The electrical supply has to be sized for that worst case, not the average, which is a point operators sometimes miss when estimating a new connection.

5. Liquid cooling and the push from AI racks

Traditional racks drawing 5 to 15 kilowatts can usually be cooled with air alone. AI racks packed with GPU servers, now commonly quoted from 40 kilowatts up to 100 kilowatts or more, generate a heat density that air alone struggles to remove efficiently, which is pushing the industry toward direct liquid cooling, where coolant runs through cold plates attached to the chips themselves, or immersion cooling, where whole servers sit in a dielectric fluid.

Adoption is moving quickly. One 2026 industry survey, reported through the AFCOM State of the Data Center Report, found that 36 per cent of operators had already deployed liquid cooling and a further 28 per cent planned to adopt it within 12 to 24 months, with more than one in four racks already running on it. A separate survey of IT professionals reported through The Register found 38.3 per cent expected to deploy some form of liquid cooling within two years, up sharply from 20.1 per cent in early 2024, while Uptime Institute's own 2023 Global Data Center Survey had found only 11 per cent of facilities of any kind using liquid cooling at that point. The trend line is clear even if the exact figures vary by survey: liquid cooling is moving from a niche option to a mainstream requirement for high density AI work faster than most facilities were built to expect.

For an operator, this matters beyond the cooling plant itself. Liquid cooling systems bring their own pumps, coolant distribution units and leak detection, all of which draw power and need their own electrical circuits, and retrofitting a room built for air cooling often means new switchboard capacity as well as new plumbing.

6. Redundancy means cooling capacity is built bigger than average load

A data centre cannot afford its cooling to fail even for a short period, since IT equipment can overheat within minutes if airflow stops. That is why cooling plant is built with the same N+1 or 2N redundancy thinking used for electrical supply: an extra chiller or CRAH unit beyond what the average load needs, or two complete independent cooling systems for the most critical sites. Our guide to why data centres use so much power covers how that same redundancy logic applies to electrical supply and backup generation.

The practical effect is that the electrical supply has to cover not just the cooling running on a normal day, but the cooling capacity sitting in reserve. A facility that uses 60 per cent of its cooling capacity most of the time still needs the switchboard, mains and backup power to support the full redundant design, not the average draw.

7. What cooling load means for your cables, mains and switchboard

Cooling is not a side issue when an electrician sizes a supply for a server room, comms room or data hall. Chillers, CRAC and CRAH units, pumps and coolant distribution units are all significant electrical loads in their own right, often drawing as much current as the IT equipment they are cooling, and they need to be accounted for in the maximum demand calculation that feeds into a network application. Cable selection follows AS/NZS 3008 for current carrying capacity, and the connection itself is governed by the Service and Installation Rules of NSW.

Getting this wrong in either direction is expensive. Undersizing the supply means cooling plant cannot run at full capacity on the hottest day of the year, exactly when the IT load is also highest. Oversizing means paying for capacity that sits unused. Our guide to how much power a data centre uses in Australia works through the load calculations in detail, including how PUE and cooling overhead feed into the final number.

8. Measuring your real cooling draw before you upgrade

A rough estimate of cooling load is worse than no estimate at all if it leads to an undersized switchboard or an oversized, overpriced network application. Proper measurement means metering the cooling plant separately from the IT load over a period that captures a genuine Sydney summer, not just a mild week, since cooling demand swings hard with outside temperature and humidity. Our power logging service does exactly this, recording real demand at both the cooling circuits and the IT load so any upgrade is sized on measured peaks rather than nameplate guesses.

Once you have reliable numbers, a switchboard upgrade and larger consumer mains can be planned with confidence, and the same data supports the network application lodged with Ausgrid, Endeavour Energy or Essential Energy. You can check your network area in seconds with our postcode checker before you start. Our guide to data centre power supply upgrades in Sydney walks through that process end to end, and where a site is overhead supplied and needs three phase capacity for new chiller plant, our sister site HD Power Pole explains how that works in its guide to three phase power through a private power pole.

Why it's Level 2 ASP work

Increasing the supply to cover a bigger cooling and IT load changes the connection between the network and your installation, which is Level 2 ASP territory rather than standard electrical work. We prepare the network application, install larger consumer mains and coordinate approval with the distributor. Our Level 2 electrician page explains what that accreditation covers. For the very largest campuses, where cooling and IT load together exceed what a low voltage connection can carry, the distributor's own high voltage design and large customer process takes over, and our Level 3 design and Level 1 network installation pages explain that side of the network purely as background, since that scale of connection sits with the distributor's own engineers.

How High Demand Electrical helps

We are a licensed local team, not a call centre, and for data centres, server rooms and comms rooms we focus on the supply side of the cooling equation: measuring real demand across cooling and IT load together, sizing the cables and mains that feed the site, preparing the network application and upgrading the switchboard behind it. We work across Sydney's West, South West and Sutherland Shire, and our data centre power upgrades page sets out the full scope of that work. We will always say plainly if your facility's load points toward a high voltage connection, since that sits outside what a low voltage Level 2 ASP can deliver.

Data Centre Cooling FAQs (Sydney)

Is cooling really the biggest power user after the servers themselves?

Usually yes. Industry estimates commonly put cooling at roughly a quarter to 40 per cent of total power in older air cooled facilities, second only to the IT load itself. Modern liquid cooled designs cut that share sharply.

Can free air cooling work in Sydney all year round?

Not on its own for the hottest, most humid days. Free air and evaporative assistance can carry a lot of the load for much of the year, but mechanical cooling capacity still has to be sized for Sydney's humid summer peaks.

Does liquid cooling reduce my electrical load overall?

It typically improves PUE by cutting fan and compressor energy, but it adds its own pumps and coolant distribution units, so total facility power still needs to be measured rather than assumed. See our guide to PUE and data centre power.

Why does cooling load affect my switchboard, not just my air conditioning contractor?

Chillers, CRAC and CRAH units and pumps all draw significant current that has to be covered by your electrical supply, cables and switchboard, which is why cooling is part of the maximum demand calculation for any upgrade.

Who can measure cooling and IT load together and plan a supply upgrade?

Call High Demand Electrical on 0402 559 777. We log real demand across cooling and IT equipment, then size the mains and network application from measured numbers. See power supply upgrades.

Get a Power Supply Upgrade Quote: Call High Demand Electrical on 0402 559 777 to measure your facility's real cooling and IT load and plan the right supply for your data centre, server room or comms room. Every job needs a site assessment before scope or pricing is confirmed.