Nextdc data centre in Adelaide

What Do Data Centres Look Like?

From the street, a data centre might look like a warehouse, a multi-storey office building or a large industrial complex. Inside, the appearance changes completely: long rows of tall metal cabinets, neatly arranged cables, blinking indicator lights and machinery that keeps thousands of computers operating.

Those cabinets are called server racks. Each can hold multiple computers, storage devices and network switches, stacked horizontally rather than arranged like conventional desktop computers.

An AI data centre can look almost identical from outside. The differences become clearer when you examine the equipment inside: powerful computing systems, high-speed connections between processors and, in some designs, liquid-cooling pipes that carry heat directly away from the chips.

The easiest way to understand what a data centre looks like is to follow it from the outside in, using real Australian facilities and equipment as examples.

What does a data centre look like from the street?

A useful Australian example is NEXTDC’s A1 data centre in Adelaide, designed by architects Hames Sharley.

Rather than resembling a plain industrial shed, the building has an angular façade inspired by the sawtooth roofs of older industrial buildings in the surrounding precinct. Different coloured surfaces become visible depending on where you stand.

From the street, someone unfamiliar with the building might reasonably mistake it for an office development or another type of commercial facility.

Nextdc Adelaide

NEXTDC’s A1 data centre in Adelaide demonstrates how modern computing infrastructure can resemble a contemporary commercial building. Its distinctive façade draws on the industrial architecture of the surrounding area. Source: Hames Sharley.

Compare this with a larger data centre campus.

In Google’s data centre photo gallery, some facilities appear as expansive industrial buildings surrounded by roads, service areas and mechanical equipment.

From above, the size of the supporting infrastructure becomes clearer. Cooling equipment, electrical installations and service areas can occupy substantial space around the main buildings.

This highlights an important point: the exterior of a data centre rarely tells you exactly what computing equipment is operating inside.

A large building isn’t necessarily an AI data centre. A modern façade doesn’t indicate how many servers are running. And two facilities with similar exteriors may use very different electrical and cooling systems.

How can you recognise a data centre from outside?

There isn’t one feature that identifies every data centre, but several visual clues can help.

Exterior featureWhat it might indicate
Large, relatively plain buildingsSpace for computing equipment and supporting infrastructure
Few conventional office windowsAreas designed primarily for machinery and equipment
Outdoor cooling fans or mechanical equipmentSystems that help remove heat from the building
Large generators or electrical installationsBackup power and electricity distribution equipment
Fenced service yards and controlled entrancesSecurity and restricted access to critical infrastructure
Several similar buildings on one siteA campus containing multiple data halls or supporting facilities

None of these characteristics proves that a building is a data centre. Warehouses, hospitals, factories and telecommunications facilities can have similar features.

The building’s documented purpose is a more reliable way to identify it.

For an Australian construction perspective, Mayfair Building Group’s guide to base-building works explains how essential infrastructure, including electrical systems, pumps and heating and cooling equipment, supports the operation of commercial buildings.

Although data centres have specialised requirements, the same broad principle applies: much of a building’s important infrastructure isn’t immediately visible from outside.

What do you see inside a server hall?

Imagine standing at the end of a supermarket aisle, except that the shelves on both sides have been replaced by tall metal cabinets.

That repeated arrangement gives a server hall its distinctive appearance.

Some racks have perforated metal doors. Others appear as open frames containing horizontal equipment trays. The machines often have small status lights, ventilation openings, handles and network connections.

inside an Australian data centre

Inside a typical data centre, computers and networking equipment are arranged in rows of tall racks. The layout helps organise equipment, cabling, maintenance access and airflow.

A rack and a server aren’t the same thing.

  • A rack is the frame or cabinet holding the equipment.
  • A server is a computer installed within the rack.
  • A network switch connects computers so they can exchange information.
  • A storage device holds information used by computing systems.

One cabinet may contain many servers, so counting racks in a photograph won’t tell you how many computers are operating.

How to identify equipment in a server rack

What you can seeWhat you’re looking atWhat it does
Narrow horizontal metal unitsServers, storage devices or network switchesProcess, store or transmit information
Small indicator lightsEquipment status indicatorsShow operating conditions or connection activity
Bundles of network cablesData connectionsConnect servers and networking equipment
Thick power cablesElectrical connectionsSupply electricity to equipment
Perforated cabinet doorsVentilation panelsAllow cooling air to pass through
Pipes and connectors near computing traysLiquid-cooling connectionsHelp transfer heat away from components

One useful distinction is between the front and rear of a rack.

The front often provides access to removable equipment, status indicators and ventilation openings. The rear may expose more power connections, network cabling and cooling infrastructure.

This explains why two photographs of the same rack can look surprisingly different.

The front shows how equipment is arranged. The rear often reveals more about how it is connected, powered and cooled.

Why do some server aisles have doors?

Some server halls contain transparent doors or partitions that make the rows resemble glass corridors.

These are often part of an aisle containment system, which helps control the movement of cooling air.

Servers generate heat as they operate. In a typical air-cooled arrangement, cooler air enters one side of the equipment, passes through it and leaves as warmer air.

If warm exhaust air mixes with the incoming cool air, the cooling system may need to work harder to maintain suitable operating temperatures.

To reduce this mixing, data centres commonly arrange racks around two types of aisle:

  • Cold aisle: where equipment draws in cooler air.
  • Hot aisle: where warm air leaving the equipment collects.

Depending on the design, either aisle may be enclosed.

air cooled server racks

Source: https://www.researchgate.net/figure/Typical-Data-Center-HVAC-Hot-Aisle-Cold-Aisle-Layout_fig1_255202176

A typical air-cooled server hall separates cooler supply air from warmer exhaust air. Aisle containment helps reduce mixing between the two.

The doors themselves don’t explain the cooling system. The important detail is the direction of airflow.

This also explains why server halls look so orderly. Rack positioning isn’t simply about tidiness. It helps engineers manage heat, maintenance access and airflow.

The US Department of Energy’s data centre design guide provides further technical information about airflow and cooling efficiency.

What does an AI data centre look like inside?

At first glance, an AI data centre can look much like a conventional data centre.

Both may contain long rows of racks, network cabling, electrical equipment and cooling systems.

The differences become easier to understand when you look at what is installed inside the individual racks.

Conventional servers can run websites, databases, business applications and storage systems. AI computing systems may contain large numbers of graphics processing units, or GPUs, designed to work together on demanding calculations.

These processors aren’t normally visible from the front of a closed rack. They sit inside computing trays alongside other components.

One example is NVIDIA’s GB200 NVL72 system.

In its single-rack configuration, it contains:

  • 72 Blackwell GPUs.
  • 18 computing trays, each containing four GPUs.
  • Nine NVLink switch trays connecting the computing equipment.
  • Power distribution equipment.
  • Liquid-cooling manifolds and associated connections.

The NVIDIA DGX GB hardware guide provides diagrams illustrating how these components fit together.

NVIDIA rack diagram

An NVIDIA GB200 NVL72 rack contains computing trays, networking equipment, power infrastructure and cooling connections. The rear view reveals systems that may not be obvious from the front. Source: NVIDIA.

The cooling arrangement is particularly revealing.

In this system, liquid circulates through cold plates attached to the CPUs and GPUs. The liquid collects heat from those components before returning through the cooling system.

Other components continue to use air cooling.

That makes the equipment different from a conventional server installation relying entirely on air cooling, although liquid cooling isn’t exclusive to AI facilities.

For a more detailed explanation of how cooling systems manage heat, see our article on why data centres need water.

It explains the difference between circulating cooling water and water actually consumed by a facility, including how different cooling arrangements can affect overall water demand.

What is the visual difference between a regular data centre and an AI data centre?

What you’re comparingConventional server setupHigh-density AI setup
Front of the rackHorizontal servers, storage and network equipmentHorizontal compute trays and network equipment
Computing hardwareOften general-purpose CPUs, with GPUs where neededMay contain many interconnected GPUs
ConnectionsPower and network cablingPower, networking and potentially specialised GPU interconnects
CoolingOften air cooling, sometimes liquid coolingMay use direct liquid cooling for high-density components
Appearance from outsideWarehouse, commercial building or campusOften indistinguishable from a conventional data centre

The important point is that there is no single visual feature that proves a facility is designed for AI.

A cooling pipe doesn’t automatically indicate AI computing. A large building doesn’t reveal its GPU capacity. And a dense rack of equipment can support many different workloads.

The defining differences lie in the hardware and how it operates, rather than the outward appearance of the building.

What’s in the building besides the servers?

Server cabinets may be the most recognisable part of a data centre, but they’re only one part of the facility.

Other areas can look more like industrial plant rooms or electrical substations than computer rooms.

Cooling and mechanical rooms

A mechanical plant room may contain pumps, valves, insulated pipes, heat exchangers and equipment used to remove heat from the computing systems.

For example, a chilled-water system can carry heat from inside the building to equipment that transfers it elsewhere.

Not every data centre uses the same cooling arrangement.

Some rely heavily on air cooling, while others use liquid cooling, refrigeration or a combination of systems.

It’s also important to distinguish between water circulating inside the cooling system and water consumed through evaporation or other losses.

Our detailed guide to data centre water use and cooling systems explains these differences, including why a closed-loop cooling system doesn’t necessarily mean a facility consumes no water.

From a building perspective, mechanical rooms must also accommodate maintenance access, equipment replacement, pipework and other services.

These considerations are relevant well beyond data centres. Mayfair Building Group’s commercial plumbing and maintenance services cover building systems, preventative maintenance and associated plumbing infrastructure.

The scale of a data centre’s mechanical systems is different, but reliable access and maintenance planning remain important considerations.

Electrical rooms and backup power

Another part of the building contains electrical infrastructure.

Instead of server racks, you may see switchgear, distribution cabinets, uninterruptible power supply equipment and associated electrical connections.

These systems distribute electricity and help maintain equipment operation through certain power interruptions.

Backup generators may be located outside or in dedicated areas. They can be among the most noticeable pieces of infrastructure around a data centre.

It’s worth distinguishing generators from cooling fans. Both may be large and visible, but they serve different purposes and don’t necessarily operate under the same conditions.

This difference matters when considering what a neighbouring data centre might sound like.

Our article Are Data Centres Noisy? explains the sounds produced by cooling fans, pumps, chillers and backup generators, and why the noise reaching nearby properties depends on the facility’s equipment and layout.

How the different parts fit together

A useful way to picture a data centre is as three connected systems:

  1. Computing equipment processes, stores and exchanges information.
  2. Electrical infrastructure supplies the electricity the equipment needs.
  3. Cooling infrastructure removes the heat generated as equipment operates.

The building houses and protects these systems while providing access for installation, maintenance and future changes.

This is why a data centre is more complex than a warehouse filled with computers.

Its supporting infrastructure is essential to keeping those computers operating reliably.

Can you see inside a real data centre?

Yes, although access to operational facilities is commonly restricted for security and operational reasons.

You can still explore real examples through published photographs, equipment diagrams and virtual tours.

Google offers a 360-degree tour of its data centre in The Dalles, Oregon, along with a photo gallery featuring server equipment, cabling, buildings and cooling infrastructure.

For an Australian exterior, the NEXTDC A1 Adelaide project photographs show how a data centre can be incorporated into an urban setting.

And for a detailed look at an AI rack, NVIDIA’s hardware diagrams help identify compute trays, switches, cooling connections and power equipment.

Five things to look for in data centre photographs

If you’re trying to understand a photograph, start with these questions:

  1. Am I looking at the building, a server hall or a mechanical room? Each has a different purpose.
  2. Where are the actual computers? Look for horizontal trays installed inside tall cabinets.
  3. Which side of the rack am I seeing? The rear often reveals more cabling and cooling connections.
  4. How is heat being removed? Look for ventilation arrangements, cooling pipes, containment doors and external cooling equipment.
  5. What can the photograph actually tell me? It may reveal equipment and layout, but usually won’t establish processing capacity, energy consumption or whether the facility is running AI workloads.

These questions help distinguish what is visible from what requires technical documentation.

Why the appearance of a data centre matters in building design

For builders, property owners and project teams, a data centre’s appearance reveals more than its architectural style.

The positioning of cooling equipment affects how heat leaves the site. Electrical infrastructure influences space requirements and service access. Building layout can affect maintenance, equipment replacement and the relationship between the facility and neighbouring properties.

For example, a cooling system located on a roof may create different construction and access requirements from one located in an external plant yard.

Likewise, generators positioned near neighbouring buildings may require different acoustic treatment from equipment positioned farther away or screened by the building itself.

These are not decisions that can be made purely from the appearance of the finished facility. They require coordination between building designers, engineers, contractors and specialist equipment suppliers.

Mayfair Building Group’s project management services explain the broader importance of planning, scheduling, procurement and coordination across commercial construction work.

For specialised facilities such as data centres, these considerations become particularly significant because the building and its technical systems must work together.

What does a data centre actually look like?

From outside, a data centre may resemble a warehouse, modern commercial building or collection of industrial facilities.

Inside, its most recognisable feature is the server hall, where rows of metal racks hold computers, network equipment and storage systems.

Beyond those racks are the electrical and mechanical systems that keep everything operating.

AI data centres can appear much the same, although their computing racks may contain more densely interconnected processors and specialised cooling equipment.

The most useful way to understand a data centre photograph is to look beyond the cabinets themselves and follow the connections between the computers, their electricity supply and the equipment that removes heat.

Once you know what those systems do, the building’s appearance starts to make considerably more sense.

power market for data centres

Why Do Data Centres Use So Much Power?

Data centres use so much electricity because they operate large numbers of servers continuously, while also powering the cooling, networking and electrical systems needed to keep those servers available. Most electricity goes to computing equipment, but removing the heat it produces adds another substantial load.

The important distinction is between how much power a data centre can draw and how much electricity it actually uses. A proposed 100-megawatt data centre does not necessarily consume 100 megawatts every hour. The figure may describe its maximum connection capacity, planned IT load or eventual capacity after several construction stages.

Understanding that difference is especially important in Australia, where data centre electricity demand is forecast to increase significantly as AI and cloud infrastructure expands.

Where does a data centre use electricity?

A data centre is more than a room full of computers. It is a building designed to run computing equipment reliably, manage the heat that equipment creates and maintain service when individual components fail.

Equipment or systemWhy it needs electricity
ServersProcess requests, operate applications and perform calculations, including AI workloads.
StorageKeep information accessible and available when needed.
NetworkingMove data between servers, storage equipment and users.
CoolingOperate fans, pumps and refrigeration equipment to remove heat.
Electrical infrastructureConvert and distribute power, with some energy lost during the process.
Building servicesOperate security systems, lighting and other supporting equipment.

According to the International Energy Agency (IEA), servers account for around 60% of electricity demand on average in modern data centres.

Cooling requirements vary considerably. The IEA estimates that cooling accounts for approximately 7% of electricity consumption in efficient hyperscale facilities, compared with more than 30% in some less efficient enterprise data centres.

These percentages describe different types of facilities. They should not be treated as a fixed breakdown for every data centre.

Why does a data centre need so much cooling?

Almost all the electricity consumed by computing equipment eventually becomes heat. That heat must be removed to keep servers operating within acceptable temperature limits.

Think of a laptop becoming warm while exporting a large video. The same principle applies to a data centre, except the heat is generated by entire rows of equipment operating continuously.

power usage by data centres

Image source: https://www.researchgate.net/figure/Estimation-of-Data-Centre-Energy-Consumption-by-2030-11_fig2_385191670

Does 1 MW of server power require another 1 MW for cooling?

No. This is a common misunderstanding.

If servers and other IT equipment consume 1 megawatt (MW), approximately 1 MW of heat must be removed from the equipment and surrounding environment under steady operating conditions.

However, removing 1 MW of heat does not necessarily require 1 MW of additional electricity.

Cooling equipment uses electricity to move heat rather than create an equivalent amount of cooling energy through electricity consumption alone.

For example, a cooling system may circulate water through equipment and release heat outdoors. When external conditions are favourable, the system may need relatively little mechanical refrigeration. During hotter conditions, compressors, pumps or fans may consume more electricity.

This helps explain why two facilities with similar computing loads can have different total electricity consumption.

Cooling requirements also depend on equipment density, airflow management, building design and local climate. For a related explanation of cooling systems, see why data centres need water.

Why do AI data centres use more power?

Artificial intelligence can increase electricity demand because some AI workloads require large numbers of specialised processors working simultaneously.

Training a large AI model may involve extensive calculations across many processors for prolonged periods. Once a model is deployed, responding to user requests also consumes computing resources.

Traditional business applications can have periods of relatively low activity. AI systems, particularly those supporting large-scale inference or training, may place sustained demands on computing infrastructure.

However, not every AI workload uses more electricity than every conventional application. Consumption depends on the hardware, workload, utilisation and efficiency.

Power density matters more than building size

One important consideration is power density: how much electricity computing equipment requires within a particular area or rack.

A rack containing high-performance AI processors may draw considerably more power than a rack of conventional servers.

Higher power density can require:

  • Greater electrical distribution capacity.
  • More concentrated heat removal.
  • Different cooling equipment or liquid-cooling systems.
  • Careful planning of equipment layout and supporting infrastructure.

This means two data centres with similar floor areas can have substantially different electricity demands.

Practical implication: Building size alone is not a reliable way to estimate data centre electricity consumption. IT capacity, equipment utilisation and cooling design are more useful starting points.

How much electricity does a 10 MW data centre use?

Before calculating annual electricity consumption, it is necessary to establish what the quoted megawatts represent.

Megawatts (MW) measure power at a particular moment. Megawatt-hours (MWh) measure electricity consumed over time.

A facility drawing an average of 10 MW continuously uses:

  • 240 MWh in 24 hours.
  • 7,200 MWh over a hypothetical 30-day period.
  • 87,600 MWh, or 87.6 gigawatt-hours (GWh), over a 365-day year.

The annual calculation is:

10 MW × 8,760 hours = 87,600 MWh = 87.6 GWh

But that result only applies when 10 MW represents the average electricity draw over the entire year.

Three different meanings of a 10 MW data centre

What 10 MW meansAnnual electricity consumption
Average power drawn by the entire facility87.6 GWh
Average IT equipment power, with annual PUE of 1.3113.88 GWh
Maximum grid connection capacityCannot be determined without actual or expected average demand

The difference matters because project descriptions may refer to different boundaries and stages of development.

A 10 MW grid connection is not proof that a facility currently consumes 10 MW. It tells us the capacity of the connection, not the average amount of power drawn from it.

What is power usage effectiveness (PUE)?

Power usage effectiveness (PUE) measures the relationship between the total energy used by a data centre and the energy consumed by its IT equipment over the same period.

The formula is:

PUE = Total facility energy ÷ IT equipment energy

For example, a PUE of 1.3 means that every 1 kilowatt-hour (kWh) consumed by IT equipment corresponds to 1.3 kWh consumed by the whole facility.

The additional 0.3 kWh covers supporting systems such as cooling, power distribution losses and building services.

Consider a facility averaging 10 MW of IT electricity demand throughout the year:

  • Annual IT electricity consumption: 87.6 GWh.
  • Assumed annual PUE: 1.3.
  • Annual whole-facility electricity consumption: 113.88 GWh.

The result is calculated as:

87.6 GWh × 1.3 = 113.88 GWh

There is an important limitation: PUE must be applied to the correct boundary.

If a reported power figure already includes cooling and other facility systems, applying PUE again would double-count those loads.

The Australian Government’s data centre energy-efficiency guidance explains PUE and the importance of measuring electricity consumption across different facility systems.

Does a data centre use its maximum power all the time?

Not necessarily. A data centre can operate around the clock without drawing its maximum electrical capacity every hour.

Electricity consumption changes for several reasons:

  • Computing workloads increase or decrease.
  • Servers may operate below their maximum processing capacity.
  • New equipment may be installed in stages.
  • Customer demand may take years to reach mature levels.
  • Cooling electricity demand changes with external conditions.

Even when activity is relatively low, servers, storage and network equipment may still need to remain available.

Worked example: Connection capacity versus actual demand

Consider a hypothetical data centre with a maximum whole-site grid connection capacity of 20 MW.

If its average electricity draw is 30% of that capacity over a year, its average demand would be 6 MW.

6 MW × 8,760 hours = 52.56 GWh annually

If the average draw subsequently increases to 60% of connection capacity, average demand becomes 12 MW.

12 MW × 8,760 hours = 105.12 GWh annually

ScenarioAverage site demandAnnual electricity use
30% average use of connection capacity6 MW52.56 GWh
60% average use of connection capacity12 MW105.12 GWh
100% average use of connection capacity20 MW175.2 GWh

These are illustrative mathematical scenarios, not reported utilisation rates for an actual data centre. The percentages describe average whole-site draw relative to connection capacity, not CPU or server utilisation.

The example demonstrates why a facility’s electricity consumption can double without any change to its grid connection rating or building footprint.

For Australian communities evaluating developments, this is an important distinction. A building can look unchanged from the outside while its internal equipment and electricity demand increase substantially.

How much electricity do Australian data centres consume?

Oxford Economics Australia and the Australian Energy Market Operator (AEMO) published their Data Centres Energy Demand Outlook in Australia on 25 August 2026.

Under the central Step Change scenario, the report estimates and forecasts national data centre electricity consumption as follows:

Financial yearNational data centre electricity consumption
2025–265.2 TWh
2029–3015.8 TWh
2035–3634.3 TWh

The 2025–26 figure is an estimate; the later figures are projections, not guaranteed future electricity consumption.

The same report provides slightly lower figures for the National Electricity Market (NEM), which should not be confused with the national totals.

What is driving the projected increase?

An important finding is that new buildings are not the only source of growth.

Oxford Economics and AEMO identify the ramp-up of existing data centres and expansions of existing facilities as major drivers of near-term electricity demand.

Over time, newly developed facilities are expected to become increasingly important.

This means electricity forecasts cannot be understood simply by counting how many data centres are being built. The amount of equipment operating inside existing buildings also matters.

The report also highlights uncertainty about which proposed projects will proceed, when they will connect to the electricity network and how quickly they will reach mature operating loads.

Can an efficient data centre still use more electricity?

Yes. Improved efficiency does not automatically mean lower total consumption.

Consider a hypothetical facility that initially averages 5 MW of IT demand with an annual PUE of 1.4.

Its average whole-site demand is:

5 MW × 1.4 = 7 MW

Later, IT demand doubles to 10 MW, while its PUE improves to 1.2.

Its average whole-site demand becomes:

10 MW × 1.2 = 12 MW

MeasureInitiallyAfter expansion
Average IT power5 MW10 MW
Annual PUE1.41.2
Average whole-site power7 MW12 MW
Annual whole-site electricity61.32 GWh105.12 GWh

The facility has become more efficient at supporting each unit of IT electricity consumption, yet its overall electricity use has increased by approximately 71%.

This is why PUE alone cannot establish whether a data centre’s total electricity demand is rising or falling.

A meaningful assessment should consider both efficiency and the absolute quantity of electricity consumed.

Do backup generators contribute to normal electricity demand?

Backup generators are generally designed to supply electricity during interruptions to the normal power supply. They do not ordinarily operate continuously simply because they have a high power rating.

A generator’s rated capacity describes the power it can deliver when operating, not its contribution to everyday grid electricity consumption.

Generators may operate during testing or outages, while other equipment such as uninterruptible power supplies supports continuity during power disturbances.

Some backup arrangements therefore have electricity or fuel requirements separate from normal grid consumption.

This also matters when considering the effects of a facility on neighbouring properties. Cooling equipment may operate regularly, while standby generators generally operate intermittently. See are data centres noisy? for a related explanation.

How can you assess the electricity demand of a proposed data centre?

When reviewing a proposed Australian data centre, five questions can help distinguish a headline capacity figure from a meaningful electricity estimate.

  1. What does the quoted MW figure describe? Is it grid connection capacity, average facility demand, IT capacity or another measure?
  2. Does the figure include cooling? Confirm whether electricity for pumps, fans, refrigeration and electrical losses is already included.
  3. Is the figure a maximum or an expected average? Maximum capacity alone cannot establish annual electricity consumption.
  4. Which stage of development does it represent? The initial building may operate at a fraction of the planned completed campus capacity.
  5. What assumptions support the annual estimate? Look for operating hours, expected average load, utilisation assumptions and annual PUE where relevant.

If an announcement gives only maximum connection capacity, annual consumption cannot be determined reliably without additional operating assumptions.

For broader context on Australian developments, see who is building data centres in Australia?.

The bottom line

Data centres use large amounts of electricity because they bring substantial computing capacity into concentrated locations and maintain the systems required to operate it reliably.

Servers account for most of the demand in a typical modern facility, while cooling, networking, storage and electrical infrastructure add to the total.

AI can increase electricity demand by raising computing intensity and equipment power density. However, the actual electricity used by a facility depends on its workload, equipment, operating conditions and stage of development.

When assessing a data centre, the most useful question is not simply how many megawatts it is rated for, but how much electricity the entire facility is expected to consume over a given period.

Sources and calculation notes

This article draws on published research and government technical guidance. The worked examples are illustrative calculations developed to explain how electricity consumption, connection capacity and PUE relate to one another. They are not measurements of a specific facility.

Calculation assumptions: A full year is taken as 365 days or 8,760 hours. Average electrical loads are assumed constant in the worked annual examples unless otherwise stated. PUE examples use assumed annual ratios, not measured values from a named facility. Electricity figures may differ from actual grid imports where on-site generation or other supply arrangements apply.

Source review date: 6 November 2026. Future electricity consumption figures are scenario forecasts and may change as demand, technology and development schedules evolve.

inside an Australian data centre

Who Is Building Data Centres in Australia? Major Developers, Construction Companies and Projects

Updated: November 2026 | Australian data centre construction industry overview

Australia’s data centre construction industry is expanding rapidly, with major projects being developed by NEXTDC, AirTrunk and CDC Data Centres, alongside significant infrastructure investments from Microsoft and Amazon Web Services (AWS).

But when people ask who is building Australia’s data centres, there are two different questions: who owns and develops these facilities, and which construction companies are actually building them?

Companies such as NEXTDC and AirTrunk typically develop and operate data centre infrastructure, while construction businesses including Multiplex, A W Edwards and Kapitol deliver major building works.

For construction contractors, subcontractors, suppliers and property owners, knowing the difference matters. An announced multibillion-dollar investment does not necessarily mean construction has commenced, a head contractor has been appointed or subcontracting opportunities are available.

Below, we examine confirmed Australian data centre projects, the contractors delivering them, how these facilities are constructed and what their development means for the Australian building industry.

Who Is Actually Building Australia’s Data Centres?

Several major construction companies have publicly documented experience delivering Australian data centres.

Construction companyConfirmed projectDeveloper/operatorDocumented involvement
MultiplexS4 Sydney, Horsley Park, NSWNEXTDCEarly contractor involvement and design-and-construct delivery
A W EdwardsSYD3, Western Sydney, NSWAirTrunkExpanded construction contract announced July 2026
KapitolM3 Melbourne, West Footscray, VICNEXTDCConstruction delivery, prefabrication and systems coordination

These are documented examples rather than an exhaustive list of Australian data centre builders. Other contractors may be involved across different projects, expansion stages and specialist packages.

The distinction is particularly important on hyperscale developments, where one campus may be delivered in multiple stages over several years.

A construction company appointed to deliver the initial building may not necessarily deliver every subsequent expansion.

1. Multiplex – Building NEXTDC’s S4 Sydney Data Centre

Multiplex is delivering NEXTDC’s S4 Sydney data centre at Horsley Park in Western Sydney.

According to Multiplex’s March 2026 construction announcement, the project is located at 16 Johnston Crescent, Horsley Park, and comprises two four-storey buildings designed by HDR and Greenbox Architects.

Multiplex identifies the development as a 350MW data centre project, scheduled for completion in mid-2027.

The contractor’s involvement began well before construction commenced. Multiplex provided early design advice from 2022, followed by formal Early Contractor Involvement (ECI) in 2024.

This allowed the builder to contribute to design development, construction planning and technical feasibility before the main delivery stage.

What makes this project significant?

The S4 project highlights how data centre construction differs from conventional commercial building work.

Multiplex’s published project information identifies several notable construction features:

  • Extensive use of precast and modular construction elements.
  • Large-capacity cranes to handle substantial building components.
  • An on-site 330kV electrical substation.
  • Two four-storey buildings incorporating complex technical infrastructure.
  • Early coordination between construction, engineering and operational requirements.

In June 2026, Multiplex also reported installing exceptionally large luffing tower cranes to support construction.

Construction industry analysis: These details illustrate why early contractor involvement can be valuable on large data centre projects. Structural design, electrical infrastructure, equipment access and installation sequencing need to be coordinated before expensive specialist systems arrive on-site.

A decision affecting the building structure can also affect electrical distribution, cooling equipment, equipment replacement access and the construction program.

For subcontractors, that means a data centre project may require closer coordination between structural, mechanical and electrical teams than a conventional warehouse or office development.

Sources: Multiplex – S4 Sydney project and Multiplex – construction commencement, March 2026.

2. A W Edwards – Building AirTrunk’s SYD3 Expansion

A W Edwards, an Australian subsidiary of Bouygues Construction, is another confirmed builder involved in major Australian data centre development.

In July 2026, Bouygues Construction announced that A W Edwards had secured an expanded contract worth approximately €1.2 billion to continue delivery of AirTrunk’s SYD3 hyperscale campus in Western Sydney.

The broader SYD3 campus is designed for more than 400MW of capacity.

This is a significant confirmed construction appointment, rather than simply an announcement of future investment.

Why staged construction matters

Hyperscale data centres are often delivered progressively, with different buildings and technical systems completed as additional computing capacity is required.

That approach can create distinct procurement and construction opportunities.

For example, a campus expansion may involve:

  • Additional structural and civil construction.
  • New electrical distribution infrastructure.
  • Cooling and mechanical plant installation.
  • Data hall fit-outs.
  • Fire protection and building management systems.
  • Testing and commissioning of new infrastructure.

Not every package will necessarily be tendered separately. Contract structure, supplier prequalification and existing commercial relationships determine how work is allocated.

Construction industry analysis: An expanded construction contract provides stronger evidence of active delivery work than a general investment announcement.

However, it does not establish which individual subcontracting packages remain available. Businesses pursuing work must still investigate the appointed contractor’s procurement arrangements.

Source: Bouygues Construction – AirTrunk SYD3 expanded contract, July 2026.

3. Kapitol – NEXTDC Data Centre Construction in Melbourne

Kapitol has documented experience delivering data centre projects across Melbourne, including work associated with NEXTDC’s M1, M2 and M3 facilities.

Its NEXTDC M3 project in West Footscray provides a particularly useful example because the contractor has publicly documented construction methods, procurement challenges and technical delivery requirements.

The first stage of M3 was completed in September 2022.

Kapitol’s published project account describes a development including data halls, generators, cooling plant, electrical switch rooms and supporting infrastructure.

What the M3 project reveals about data centre construction

Kapitol identifies several construction methods used to manage the project’s complexity.

Prefabricated plant and equipment

Components including pump houses, chiller modules, generator modules and pipe sections were manufactured off-site.

This approach allowed equipment to be prepared before installation, with lifting arrangements and construction sequences coordinated in advance.

Structural design improvements

Kapitol describes changes to electrical switch-room construction that reduced work at height and improved construction efficiency.

The project also incorporated substantial precast structural elements.

Ongoing testing and commissioning

Testing was coordinated throughout delivery rather than left entirely until the end of construction.

This matters because data centre electrical, mechanical and control systems must operate together. Completing the building structure is not the same as demonstrating that the facility’s critical infrastructure is ready for operation.

What construction businesses can learn

The M3 case provides three practical lessons.

First, procurement planning can affect construction progress when major components have long manufacturing and delivery periods.

Second, prefabrication is not simply a way to reduce on-site labour. It requires careful design coordination, transport planning, lifting arrangements and quality control.

Third, testing and commissioning should influence the construction program from an early stage.

These lessons are drawn from Kapitol’s published project account, not from an independently conducted site inspection.

Source: Kapitol – NEXTDC M3 Data Centre Portfolio.

Which Companies Are Developing Australia’s Data Centres?

The companies commissioning Australian data centres are not always the businesses physically constructing them.

Understanding the principal developers and operators helps identify where construction demand originates.

NEXTDC

NEXTDC develops and operates data centres across Australia.

Its facilities include established sites, developments undergoing expansion and new projects at different planning and construction stages.

S4 Sydney is an example of an active development with a publicly identified construction contractor.

By contrast, NEXTDC’s A1 Adelaide facility officially opened in September 2024.

That distinction matters: A1 Adelaide represents completed infrastructure, not a future construction opportunity simply because it appears in a list of NEXTDC investments.

Source: NEXTDC – A1 Adelaide opening.

AirTrunk

AirTrunk develops and operates hyperscale data centre campuses designed for major computing requirements.

The SYD3 expansion demonstrates how the operator and head contractor can have separate responsibilities.

AirTrunk is the data centre business behind the campus, while A W Edwards has a documented construction delivery role.

The same operator may appoint different contractors across different campuses or construction stages.

CDC Data Centres

CDC develops and operates major data centre infrastructure in Australia and New Zealand.

In May 2026, CDC announced a 555MW customer contract, with capacity expected to become operational during FY2028 and FY2029.

This is an important indicator of future infrastructure demand, but it is not equivalent to an announcement that 555MW of capacity is already operational.

Nor does the customer contract identify the construction contractor for every associated development.

CDC did not publicly name the customer in that announcement, so identifying it as a particular technology company would be speculation.

Source: CDC – 555MW customer contract.

Are Microsoft and Amazon Building Data Centres in Australia?

Both Microsoft and Amazon are making major investments in Australian computing infrastructure.

However, investment figures need to be interpreted carefully.

Microsoft’s A$25 billion Australian investment

In April 2026, Microsoft announced plans to invest A$25 billion in Australian cloud and AI infrastructure by the end of 2029.

The company described this as capital and operational expenditure associated with expanding its computing infrastructure.

That expenditure should not be confused with the value of construction contracts.

Importantly, the NSW Government’s March 2026 list of projects progressing through its Investment Delivery Authority process included a Microsoft Datacenter (Australia) proposal at Honeman Close in the Blacktown local government area.

This provides a project-specific development reference.

It does not, by itself, establish that a head contractor has been appointed or construction has commenced.

Sources: Microsoft – Australian investment announcement and NSW Government – data centre development announcements.

Amazon Web Services’ A$20 billion investment

Amazon announced a planned A$20 billion investment between 2025 and 2029 to expand, operate and maintain its Australian data centre infrastructure.

The commitment includes infrastructure development and ongoing operational expenditure.

For construction suppliers, the headline figure indicates the scale of AWS’s infrastructure program, but it does not reveal the builder appointed to each project.

Identifying actual construction opportunities requires more detailed information about sites, project stages, contracting arrangements and procurement.

Source: Amazon – Australian data centre investment.

Why Data Centre Construction Is Different From a Normal Commercial Building

Although data centres may resemble large industrial buildings externally, their technical requirements make them substantially different from many conventional commercial developments.

The building shell is only one part of the project.

Data centres must accommodate extensive systems supporting computing equipment, including electrical distribution, backup power, cooling, controls and fire protection.

building data centres in Australia

Electrical infrastructure

Large data centres can require substantial incoming electrical capacity, specialist switchgear, transformers, uninterruptible power supply systems and backup generation.

These systems require careful integration with the building layout, maintenance access and operational requirements.

Mechanical cooling

Computing equipment produces heat that must be removed to maintain operating conditions.

Depending on the facility’s design, cooling infrastructure may include chillers, air-handling equipment, pumps, pipework and heat-rejection systems.

Different cooling approaches also have different implications for energy consumption and water requirements.

Our guide to why data centres need water explains the relationship between data centre cooling systems and water use.

Structural and equipment installation requirements

Generators, chillers, transformers and other large components can create significant structural, access and lifting requirements.

Kapitol’s M3 example illustrates how prefabrication and installation planning can influence construction methods.

Multiplex’s S4 project similarly demonstrates the importance of planning for unusually large construction components.

Testing and commissioning

A completed building is not necessarily an operational data centre.

Electrical, mechanical and control systems must be tested to demonstrate that they operate as intended, both individually and together.

Commissioning requirements can have a substantial influence on construction sequencing and project handover.

Noise and neighbouring properties

Cooling equipment, ventilation systems and backup generators can also create noise considerations.

For surrounding properties, the most relevant evidence is generally the project-specific acoustic assessment and the conditions applying to its development.

Our guide to whether data centres are noisy discusses common operational noise sources.

Where Are Australia’s New Data Centres Being Built?

Sydney and Melbourne remain important centres for major data centre construction, although development also extends to other Australian locations.

Western Sydney

Western Sydney includes significant development activity involving NEXTDC, AirTrunk and other infrastructure businesses.

In March 2026, the NSW Government announced that 15 data centre projects, collectively valued at A$51.9 billion, would progress through its Investment Delivery Authority process.

The published list included proposals involving NEXTDC, Microsoft, STACK Infrastructure and other developers.

However, a project progressing through this process should not automatically be described as approved, under construction or operational.

Each development must be assessed against its own planning and construction milestones.

Melbourne

Melbourne has established data centre infrastructure and ongoing construction activity.

Kapitol’s documented NEXTDC projects demonstrate how existing campuses can undergo staged construction, expansion and technical upgrades.

For suppliers, this means opportunities can arise from expansion works as well as entirely new developments.

Adelaide and other locations

NEXTDC’s A1 Adelaide facility demonstrates that major computing infrastructure is not confined to Sydney and Melbourne.

Its September 2024 opening is evidence of an established operational facility, rather than proof of current construction at that site.

National development activity should therefore be assessed project by project rather than inferred from general investment announcements.

How Can Contractors and Suppliers Find Data Centre Construction Opportunities?

For construction businesses, the most useful question is often not simply which companies are developing data centres, but which projects have reached a stage where construction packages may be procured.

A practical investigation should follow five steps.

Step 1: Identify the developer and exact project

Start with the facility name, address and developer.

Large operators may have multiple facilities with similar naming conventions, and one campus may contain several buildings or expansion stages.

Step 2: Establish the development stage

Check whether the project is:

  • Proposed or undergoing early feasibility work.
  • Subject to a planning application.
  • Approved but not yet under construction.
  • Preparing for construction or undergoing early works.
  • Under active construction.
  • Being fitted out, tested or commissioned.
  • Operating or undergoing expansion.

These are materially different stages and should not be grouped together.

Step 3: Identify the appointed head contractor

Look for announcements from the developer and construction company.

A confirmed contract announcement is generally more reliable than assumptions based on previous relationships.

A builder that delivered one NEXTDC facility, for example, should not automatically be assumed to hold the contract for every future NEXTDC project.

Step 4: Investigate specialist packages

Depending on the development, opportunities may exist in electrical, mechanical, civil, structural, hydraulic, fire protection, controls, communications and commissioning works.

However, some major systems may be procured directly by the developer or through specialist contracting arrangements.

The head contractor’s scope should be confirmed before assuming every trade package is available through it.

Step 5: Check procurement requirements

Data centres are complex facilities, and contractors may face demanding requirements relating to safety, technical capability, quality assurance, commissioning and delivery experience.

Suppliers should confirm the relevant qualification and procurement process rather than approaching every organisation associated with a project.

What Do Major Data Centre Announcements Actually Tell Us?

A useful way to assess the Australian market is to separate four common announcement types.

AnnouncementWhat it confirmsWhat it does not necessarily confirm
Investment commitmentA company intends to allocate capital or expenditureA specific construction contract
Planning application or government progressionA proposed development has entered or advanced through a formal processFinal approval or construction commencement
Construction contract awardA contractor has been appointed for a defined scopeThat every trade package is still available
Facility openingA facility or stage has reached an operational milestoneThat further stages are not planned

This distinction is important because the Australian market contains projects at very different stages of development.

For instance, NEXTDC’s S4 Sydney has a publicly confirmed construction contractor, CDC’s 555MW announcement concerns future contracted capacity, and Microsoft and AWS have announced major infrastructure expenditure programs.

All three types of information matter, but they answer different questions.

The most useful indicators of immediate construction activity are project-specific appointments, construction commencement announcements and verified site progress, not investment headlines alone.

Commonly Asked Questions

Who are the biggest data centre builders in Australia?

Multiplex, A W Edwards and Kapitol are examples of construction companies with publicly documented Australian data centre work. The examples in this article do not establish a definitive ranking by annual revenue, project value or completed data centre capacity.

Is NEXTDC a construction company?

NEXTDC is primarily a data centre developer and operator. It works with construction partners to deliver facilities. Multiplex’s involvement in S4 Sydney and Kapitol’s work on Melbourne facilities are documented examples.

Who is building AirTrunk’s SYD3 data centre?

A W Edwards is a confirmed construction delivery partner for AirTrunk’s SYD3 campus in Western Sydney. Its parent company, Bouygues Construction, announced an expanded contract in July 2026.

Are Australian data centres being built because of AI?

AI is one important driver, alongside cloud computing, enterprise computing needs and digital infrastructure demand. Recent announcements from Microsoft, Amazon and major operators identify growing computing demand as a reason for investment.

Can smaller Australian contractors work on data centre projects?

Potentially, yes. Specialist subcontractors and suppliers may be involved in civil, structural, electrical, mechanical and other packages. Actual opportunities depend on the project’s stage, contracting model and supplier qualification requirements.

Does an announced data centre mean construction has started?

No. An announcement may relate to a proposed development, planning process, investment commitment, customer agreement or actual construction commencement. These milestones must be distinguished.

The Bottom Line: Identifying Who Is Building a Data Centre

Australia’s data centre construction sector involves a combination of major infrastructure developers, cloud companies, head contractors and specialist trades.

NEXTDC, AirTrunk and CDC are among the businesses developing data centre capacity. Microsoft and AWS are making substantial infrastructure investments. Multiplex, A W Edwards and Kapitol are examples of contractors with confirmed delivery experience.

But the company funding or operating a facility is not necessarily the company physically constructing it.

For anyone assessing a proposed data centre, three questions provide the clearest starting point:

  1. Who is developing or operating the facility?
  2. Which construction contractor has actually been appointed?
  3. What stage has the specific project reached?

Together, those answers provide a more reliable understanding of Australian data centre construction than a list of investment announcements or operator names.

Sources and research methodology

This article draws on publicly available announcements and project information from NEXTDC, Multiplex, AirTrunk’s construction partner Bouygues Construction, Kapitol, CDC Data Centres, Microsoft, Amazon and the NSW Government.

Project details have been compared to distinguish development intentions, confirmed construction appointments, operating facilities and future capacity commitments.

Information reviewed: November 2026. Construction schedules, appointments and project status may change. This analysis is based on published records and does not claim first-hand project involvement or private interviews.

data centre noise

Are Data Centres Noisy?

Yes, data centres can be noisy. Cooling fans, chillers, pumps and backup generators all produce sound. However, how much noise reaches neighbouring properties depends on the equipment, building layout, distance and acoustic controls.

A facility can be loud inside its plant areas without creating the same noise level at a nearby home. Equally, a relatively steady hum can become noticeable in a quiet neighbourhood, particularly overnight.

For residents and building owners, the useful question is: what will people hear at their property during normal operation, hot weather and generator testing?

What makes noise at a data centre?

Data centres contain computing equipment that produces heat and supporting systems that keep it operating. Much of the noise relevant to surrounding properties comes from that supporting infrastructure.

In an explanation published by UNSW, acoustics specialists Marion Burgess and John Davy identify cooling equipment and backup generators as significant noise sources. They also distinguish noise within a facility from the sound reaching surrounding areas after distance and noise controls are considered. unsw.edu.au

Noise sourceWhat it doesWhen it may be heard
Cooling fans and air-handling unitsMove air through cooling and ventilation systemsDuring normal operation, with output varying by demand
Chillers and cooling towersHelp remove heat from the facilityDepending on cooling design, weather and computing load
PumpsCirculate cooling liquidsWhile the associated cooling circuits operate
Backup generatorsSupply power when the normal supply is unavailableDuring testing, maintenance and power interruptions

The mix varies between facilities. A description of one data centre’s equipment should not be treated as a noise forecast for another.

Do data centres make noise all night?

Data centres generally operate around the clock, so their cooling and ventilation systems may continue overnight.

However, continuous operation does not mean every fan, pump or cooling unit runs at full output all the time. Operating conditions can change with equipment demand, outside temperature and control settings.

A proper assessment should therefore consider several operating scenarios. A daytime visit while a facility is lightly loaded does not necessarily reveal what neighbours will hear during a warm night or after the site expands.

For anyone reviewing a proposal, useful questions include:

  • What equipment operates overnight?
  • What is the predicted noise at the nearest homes?
  • Does the assessment include hotter weather and higher cooling demand?
  • Does it cover the completed development or only the first stage?
  • How will actual performance be checked once the facility opens?

Why can a hum be more noticeable than expected?

Noise is not only about volume. Its character, duration and timing also matter.

A distinct tone or recurring drone may attract attention differently from a broad wash of background sound. This is why a single decibel figure does not always provide a complete description of someone’s experience.

EPA Victoria’s industrial-noise guidance considers factors including sound character, frequency content, duration and how often noise occurs. Its guidance also identifies circumstances where a detailed assessment by an acoustic specialist is appropriate. Those Victorian provisions are not South Australian rules, but they illustrate why industrial noise assessment goes beyond a simple loud-or-quiet judgement. epa.vic.gov.au

If the concern is a persistent hum, describe the hum when reporting it. Its timing and character can help an investigator decide what needs measuring.

How loud are data centres in decibels?

There is no single decibel figure that reliably describes every data centre.

A measurement beside a generator, inside a server hall or at a neighbouring bedroom represents a different situation. Comparing those readings without their locations and operating conditions can be misleading.

When someone presents a noise figure, ask:

Where was it measured?
At the equipment, the site boundary or a nearby property?

What was operating?
Normal cooling equipment, generator testing or a combination of systems?

What does the figure represent?
An average over a period, a maximum level or a prediction?

What else was audible?
Traffic, wind and other industrial activity can affect the measurement.

A useful acoustic report connects its figures to a clearly defined location and operating scenario. An unexplained claim that a facility produces “X decibels” tells a neighbour very little.

Are backup generators running constantly?

Backup generators are generally intended for interruptions to the normal electricity supply, with testing and maintenance needed to confirm they are available when required. They should not automatically be treated as equipment running continuously throughout normal operation. unsw.edu.au

Nevertheless, their operating arrangements deserve attention.

For a proposed development, ask how frequently testing is planned, how long it will last and whether multiple generators may operate together. Testing and emergency operation should be clearly distinguished in the project documents.

That distinction makes discussions more useful. Residents need to understand both the everyday sound environment and the less frequent operating conditions.

How can building design reduce data centre noise?

Noise management works best when it influences the design before equipment locations and building layouts are fixed.

Cundall acoustics principal Victor Lindstrom explains that plant placement, building screening and assessment of the combined noise from a campus can affect both the initial development and its future expansion. His published guidance also highlights the need to coordinate acoustic performance with heat rejection. Cundall

Position equipment carefully

The design team can compare alternative locations for cooling plant and generators, considering neighbouring buildings and the paths sound may take.

An early layout review is an opportunity to ask whether the building itself can provide useful screening.

Select suitable noise controls

Potential measures include quieter equipment, acoustic louvres, barriers, enclosures and silencers. Their suitability depends on the particular source and installation. unsw.edu.au

Rather than simply specifying an “acoustic screen”, the project team should establish what performance is needed and demonstrate how the proposed treatment achieves it.

Keep cooling performance in the design discussion

Acoustic treatment needs to work alongside the cooling system. Cundall’s project experience shows why acoustic modelling and airflow analysis may need to be considered together. Cundall

For more background on the cooling process, read our guide to why data centres need water.

Check the finished installation

A practical project brief should identify how noise performance will be verified after installation, who will investigate unexpected results and how any required improvements will be managed.

This turns a design prediction into something the operator can check against the completed facility.

What noise rules apply in South Australia?

South Australia’s Environment Protection (Commercial and Industrial Noise) Policy 2023 provides a framework for assessing commercial and industrial noise. The applicable assessment needs to consider the particular site and relevant requirements. legislation.sa.gov.au

A noise limit quoted for a project in Sydney, Melbourne or overseas should not be assumed to apply to an Adelaide development.

For a specific proposal, ask the project’s acoustic consultant to identify the applicable criteria and explain how they have been applied. The assessment should make clear which locations, operating conditions and periods have been considered.

What should nearby residents ask to see?

If a data centre is proposed near your property, start with the project’s acoustic assessment and supporting planning documents.

Look for answers to these questions:

  1. Is your property represented? Check whether the assessment includes your location or explains why another assessment point represents it.
  2. Are overnight conditions covered? Daytime results alone may not address your main concern.
  3. Is generator testing explained? Look for timing, duration and simultaneous operation assumptions.
  4. Is future expansion included? Establish whether the report covers the whole development.
  5. Are noise controls specific? Look for defined measures rather than general commitments.
  6. Is there a verification process? Ask how predictions will be checked after the facility begins operating.

For an existing concern, keep a record of the dates, times, duration and character of the sound. Note where you heard it and whether windows were open or closed. These observations give the operator or investigating authority a clearer starting point.

What this means for commercial building projects

Data centres make the relationship between building services and neighbouring properties particularly visible. The same coordination issue arises on smaller commercial projects with substantial cooling, ventilation or standby-power equipment.

A useful early brief asks the building designer, mechanical engineer and acoustic consultant to agree on equipment locations, access requirements and noise objectives before construction constrains the options.

Mayfair Building Group provides commercial building, maintenance and project management services in Adelaide. For projects involving substantial mechanical services, early coordination helps the building team understand the space and construction requirements of specialist designs.

Typically asked questions

What we get asked.

Are all data centres noisy to live near?

No single answer applies to every location. What reaches a nearby home depends on the facility’s equipment, layout, operating conditions and noise controls. A site-specific assessment is more useful than the building’s size alone.

Are data centres noisier at night?

They are not necessarily producing more sound at night, but overnight operation needs to be assessed separately. A daytime visit cannot establish what a neighbouring property will experience throughout the night.

Does liquid cooling make a data centre silent?

No. Liquid cooling still involves supporting equipment and a way to release heat. The whole cooling and power system needs to be considered when assessing noise.

Can data centre noise be reduced after construction?

Potential improvements should be assessed by an acoustic specialist alongside the mechanical design team. Available space, equipment access and cooling requirements can constrain the options, which is why early acoustic input is valuable.

How far away do you need to be to avoid the noise?

There is no universal distance that guarantees a particular result. The appropriate assessment considers the source, surroundings, screening and the location where people may be affected.

a new data centre

Why Do Data Centres Need Water?

Data centres use water mainly to help remove heat from servers and other computing equipment.

Every time a server processes data, it produces heat. At the scale of a data centre, thousands of servers may be running continuously, so that heat has to be collected, moved away from the equipment and released outside the building.

Water is useful because it can carry heat efficiently. In some cooling systems, water also evaporates, and that evaporation helps remove heat from the facility. When this happens, the lost water has to be replaced.

However, not every data centre uses large amounts of water. Some systems circulate water through sealed loops. Some use air cooling, refrigeration, dry coolers or direct-to-chip liquid cooling. Some use recycled or non-potable water instead of drinking water.

The important question is not simply:

“Does the data centre use water?”

A better question is:

“How does the facility reject heat, how much replacement water does it need, and where does that water come from?”

data centre water consumption

Image source: https://www.mordorintelligence.com/industry-reports/study-of-data-center-water-consumption-in-australia

Why do data centres need so much cooling?

Think about how warm a laptop can get when it is working hard. A data centre is that same basic problem at a much larger scale.

Servers, processors, memory, power supplies and networking equipment all turn electrical energy into heat. If that heat is not removed, equipment can slow down, shut down or fail.

Fans can move warm air away from the servers, but the heat still has to leave the building. Moving heat from one part of a room to another does not solve the problem.

That is why data centres rely on mechanical services such as air handling units, chilled-water systems, pumps, heat exchangers, chillers, cooling towers, refrigerant systems and liquid cooling equipment.

The exact design depends on the facility, the climate, the computing load and the operator’s priorities around energy use, water use, reliability and cost.

How water can cool a data centre

Water can play two different roles in a data centre cooling system:

  1. Collecting heat from equipment
  2. Releasing that heat outside the building

Those two jobs are not always done by the same water circuit.

In a conventional chilled-water and cooling-tower arrangement, the process may work like this:

  1. Air carries heat away from the servers.
  2. Cooling equipment transfers that heat into a chilled-water loop.
  3. A chiller transfers the heat into a separate condenser-water loop.
  4. The condenser water carries the heat to a cooling tower.
  5. At the cooling tower, some water evaporates and releases heat into the outside air.

That does not mean water is flowing over exposed computer equipment. The water is moving through controlled mechanical systems.

In newer high-density systems, liquid cooling may collect heat closer to the source. For example, direct-to-chip cooling uses liquid-filled cold plates attached to hot components. The liquid remains contained inside the cooling system.

Either way, the heat still has to go somewhere.

Where does the water actually go?

In many cooling-tower systems, most of the water is recirculated. But recirculation does not mean the system can run forever without a fresh supply.

Water can leave the system in several ways:

RouteWhat happens
EvaporationSome water becomes vapour and carries heat into the atmosphere.
BlowdownSome water is deliberately drained to manage dissolved minerals.
DriftSmall droplets can escape with the airflow, although equipment is used to reduce this.
Leaks or overflowFaults, poor control or ageing equipment can cause avoidable losses.

The replacement water added to the system is called make-up water.

Evaporation leaves minerals behind. If those minerals are not managed, they can build up and contribute to scale, corrosion or poor performance. That is why cooling-tower systems need water treatment, monitoring and controlled discharge.

Water that evaporates has not disappeared from the planet. But it is no longer immediately available to the local water supply from which it was taken. That distinction matters when communities are assessing pressure on local water resources.

Does a closed-loop system still use water?

Yes, it can.

“Closed loop” means liquid circulates within a contained circuit. It does not automatically mean the whole building has no water consumption.

For example, a server cooling loop might be closed, while another part of the facility still uses a cooling tower that evaporates water outside.

So the key question is:

Does the final stage of cooling evaporate water?

If the system ultimately rejects heat through evaporation, it will need make-up water. If the system rejects heat through dry coolers, air cooling or other non-evaporative methods, ongoing on-site cooling-water demand may be much lower.

Microsoft has described newer data centre designs that use closed-loop liquid cooling and consume zero water for cooling during normal operation. That is an important design shift, especially for high-density AI workloads. But it is not the same as saying every data centre, or every Microsoft facility, uses no water.

Water may also still be needed elsewhere on site, including for amenities, cleaning, landscaping, commissioning, maintenance or fire systems.

What data centre cooling experts and operators say

Industry guidance is increasingly clear on one point: data centre water use has to be assessed locally.

Uptime Institute, a well-known data centre authority, has warned that general statements about data centre water use can be misleading because the impact depends on the local water supply, the climate, the cooling system and competing demands in the watershed. In water-stressed areas, Uptime Institute says operators should consider dry cooling or minimal-water cooling strategies.

ASHRAE’s data centre guidance also shows that cooling design is not one-size-fits-all. Its technical work covers environmental limits for air-cooled and liquid-cooled IT equipment, including allowable temperature and humidity ranges. In practice, those limits affect how much mechanical cooling a facility needs and whether water-based cooling is appropriate.

Large operators are making the same point in their own sustainability guidance:

  • Google says its cooling decisions consider the local environment and balance energy, carbon-free electricity and responsibly sourced water. In higher water-risk locations, Google says it looks at alternatives such as air cooling or recycled water.
  • Microsoft has introduced data centre designs for AI workloads that use closed-loop cooling and consume zero water for cooling during operation.
  • The US Department of Energy highlights cooling-tower management as a major water-efficiency opportunity, particularly through water treatment, cycles of concentration, leak control and blowdown management.
  • Berkeley Lab separates direct on-site water use from indirect water use associated with electricity generation, which is important when comparing “zero water for cooling” claims with a facility’s broader water footprint.

The practical takeaway is simple: a credible water-use claim should explain the cooling system, the water source, the local conditions and the measurement boundary.

A statement such as “the data centre uses water” is too broad.

A statement such as “the facility uses evaporative cooling, supplied by potable mains water, with peak summer make-up demand of X litres per day” is much more meaningful.

Why not just use air cooling?

Data centres can use air cooling, refrigeration and dry heat rejection. Some do.

The reason water is still used in many facilities is that cooling design involves trade-offs.

Evaporative cooling can reduce electricity use because evaporation is an efficient way to reject heat. But it increases local water demand.

Dry cooling can reduce or avoid water consumption, but it may require more electricity in some climates or operating conditions.

Liquid cooling can remove heat very efficiently from high-density equipment, but the system still needs a way to reject that heat outside the building.

So “less water” and “less energy” are not always the same design outcome. The best solution depends on the local climate, available water sources, equipment density, power availability, operating temperatures and community expectations.

Do data centres have to use drinking water?

No.

Where suitable supplies exist, data centres may use recycled water, reclaimed water, industrial water, groundwater or other non-potable sources.

However, switching water sources is not as simple as connecting a different pipe. Water quality affects treatment, filtration, corrosion control, scaling risk, maintenance and equipment life.

A proposed recycled-water supply also needs enough capacity and reliability to support the facility during peak demand periods.

Rainwater collection may also contribute to a site’s supply, but it is rarely a complete answer by itself. Its usefulness depends on roof area, rainfall patterns, tank size, storage capacity and daily cooling demand.

It is helpful to separate two different goals:

  1. Reducing total water demand
  2. Reducing the use of drinking water

A project can do one without doing the other.

Australian example: groundwater cooling at Pawsey

The Pawsey Supercomputing Research Centre in Perth provides a useful Australian example.

CSIRO has described Pawsey’s groundwater cooling system, which pumps groundwater from an aquifer through an above-ground heat exchanger and then returns it underground. This allows the groundwater to carry away heat without relying on conventional cooling-tower evaporation for that cooling duty.

That does not mean groundwater cooling is suitable for every site.

Local geology, groundwater quality, environmental approvals, aquifer behaviour and long-term monitoring all matter. But the Pawsey example shows that water can perform a cooling role without necessarily being consumed through evaporation.

Image source: groundwater cooling at Pawsey

How much water does a data centre use?

There is no single daily figure that applies to all data centres.

Water use depends on the facility, the computing load, the cooling system, the weather, the source of water and what the operator includes in the measurement.

One common metric is water usage effectiveness, or WUE. It usually expresses water use in litres per kilowatt-hour of IT energy.

For example:

  • IT equipment uses 100,000 kWh in a day.
  • The assumed water intensity is 1 litre per kWh.
  • The reported water use within that boundary would be 100,000 litres.

That is only an example of the calculation. It is not an industry average and should not be treated as a forecast for a particular building.

When reading a data centre water-use claim, ask:

  • Is the number measured or estimated?
  • Is it an annual average or a peak-day figure?
  • Does it include drinking water only, or all water sources?
  • Does it measure withdrawals, consumption or both?
  • Does it include cooling only, or all site water use?
  • Does it include water used to generate electricity?
  • Is it for one facility, one campus or a whole company?

Peak demand is especially important. A modest annual average may hide much higher water use on hot days, when cooling demand and pressure on local infrastructure can both increase.

What about water used to generate electricity?

On-site cooling is only part of the picture.

Electricity generation can also use water, depending on how and where the electricity is produced. Some power stations use water for their own cooling systems.

This means a facility with low on-site water use may still have an indirect water footprint through the electricity it consumes.

It also means overseas estimates should not be copied into an Australian context without checking the electricity mix, climate, cooling design and reporting boundary.

A strong water assessment should look at both:

  1. Direct water use on site
  2. Indirect water use linked to electricity supply

What this means for data centre design in Australia

For proposed data centres in Adelaide or elsewhere in Australia, water demand should be considered early in the design process.

By the time the building layout, plant rooms, roof space, service routes and external infrastructure are fixed, some water-saving options may become harder or more expensive to adopt.

Useful questions for the project team include:

  • Cooling method: Will the facility use evaporative cooling, dry cooling, refrigeration, liquid cooling or a combination?
  • Peak demand: What happens during hot weather?
  • Water source: Will the site use potable mains water, recycled water, groundwater, rainwater or another supply?
  • Local capacity: Can the local water network support the proposed demand?
  • Treatment: What water quality is required for the cooling equipment?
  • Monitoring: Can the operator separate cooling water use from amenities, landscaping and other uses?
  • Maintenance: Is there enough access to service pumps, filters, heat exchangers, chillers and treatment equipment?
  • Resilience: What happens if a cooling unit, pump, water supply or power source becomes unavailable?
  • Reporting: Will the operator publish annual, seasonal and peak-day water-use information?

These questions need input from mechanical engineers, hydraulic consultants, electrical designers, water authorities, sustainability advisers and the building team.

They are also easier to resolve before construction begins.

What building owners should look for in water-use claims

Not all data centre water claims are equally useful.

A credible claim should tell you:

  • What type of cooling system is used
  • Whether water evaporates as part of normal operation
  • Whether the water is potable, recycled or from another source
  • Whether the figure is daily, annual, average or peak
  • Whether the number is measured or modelled
  • Whether indirect water use from electricity is included
  • Whether the facility has leak detection and water monitoring
  • Whether the operator reports local water impact, not just company-wide totals

Be cautious with broad claims such as:

  • “Water efficient”
  • “Closed loop”
  • “Sustainable cooling”
  • “Zero water”
  • “Low impact”
  • “AI-ready infrastructure”

Those terms may be accurate, but only if the details support them.

For example, “closed loop” may describe the server cooling circuit, while another part of the system still consumes water. “Zero water for cooling” may refer to normal cooling operations, while the site still uses water for other purposes.

The details matter.

Key takeaway

Data centres need cooling because computing equipment produces heat.

Water is often used because it can move and release heat effectively. But water use varies widely depending on the cooling system, the climate, the water source and the facility’s operating conditions.

The most important question is not whether water appears anywhere in the design.

The most important question is:

How much water is consumed, when is it needed, where does it come from, and what alternatives were considered?

For commercial building owners and project teams, water demand should be part of the early design conversation, especially for facilities with high cooling loads.

Mayfair Building Group provides commercial building, maintenance and project management services in Adelaide. For projects involving complex building services, early coordination between builders, designers and service providers can help avoid costly design constraints later in the project.

Commonly asked questions

Read the common questions we get asked below.

Do all data centres need water for cooling?

No. All data centres need heat removal, but not all rely on evaporating water. Some use dry cooling, refrigeration, air cooling or closed-loop liquid cooling systems with very low ongoing cooling-water consumption.

Is liquid cooling the same as water consumption?

No. Liquid cooling describes how heat is collected or transported. Water consumption depends on how the system ultimately rejects heat and whether water leaves the system through evaporation, discharge or other losses.

Can cooling water be reused?

Yes. Many systems recirculate cooling water many times. However, systems with cooling towers still need make-up water to replace evaporation, blowdown, drift and losses.

Can data centres use recycled water?

Yes, where suitable recycled-water supplies are available. The water still has to meet the quality and reliability requirements of the cooling system.

Does AI use water?

AI runs on computing equipment that needs electricity and cooling. The water impact depends on the data centre, the workload, the cooling system, the local climate and the electricity supply. A single litres-per-question figure cannot describe every AI interaction.

What is WUE?

WUE stands for water usage effectiveness. It is a data centre metric that usually expresses water use in litres per kilowatt-hour of IT energy. Always check what the figure includes before comparing one facility with another.

Why does data centre water use matter locally?

Because the impact depends on the source of water, the timing of demand and the capacity of local infrastructure. Annual use is useful, but it does not always show what happens during the hottest or driest periods.