Data centres are no longer a future concept—they are rapidly becoming a critical part of Australia's digital infrastructure, and their presence is expected to grow significantly over the next 25 years. This growth will be driven largely by the extraordinary expansion of artificial intelligence (AI), cloud computing, digital services and the ever-increasing volume of information being created, processed and stored.
The case for data centres is compelling. Australia needs the computing capacity to support the growing demand for AI tools, remain connected to the rapidly expanding global information economy and ensure that we are not left behind in the international race to develop and deploy increasingly capable AI technologies.
Data centres will therefore play an important role in supporting Australia's future productivity, competitiveness and technological capability.
However, this growth also raises an important question:
How much of Australia's finite energy and infrastructure resources should be allocated to data centres, and who ultimately bears the cost?
Data Centre Growth Is Imminent
As outlined in the introduction, data centres are not new. They have been a fundamental part of the digital economy for more than 50 years. The important question is therefore not why we have data centres, but why their rapid growth is becoming such an important issue for Australia's energy system today.
The answer is largely Artificial Intelligence (AI).
The world is experiencing an unprecedented demand for AI-based products and services, and Australia is no exception. AI is increasingly being used across almost every industry, from finance and healthcare to manufacturing, education, government and professional services.
As the use of AI expands, so does the amount of computing power and data storage required to operate these systems.
This is fundamentally changing the type and scale of data centres being built.
From Traditional Data Centres to AI Infrastructure
Traditional data centres were primarily designed to house servers that stored information, hosted websites and applications, and supported IT systems for businesses and consumers. While these facilities could be large, their electricity requirements were generally modest compared with the facilities now being developed for AI and high-performance computing.
AI changes this equation.
AI systems require enormous amounts of computing power to train models, process information and deliver responses to users in real time. This requires large numbers of specialised processors, particularly graphics processing units (GPUs), operating continuously and at very high power densities.
The result is a new generation of data centres that are much larger, more power-intensive and more specialised than many of the facilities of the past.
For the purposes of this paper, data centres can broadly be considered across three levels of capability and scale.
1. Tier 1 – Basic Data Centres
Tier 1 represents the most basic level of data-centre infrastructure. These facilities typically have a single, non-redundant pathway for power and cooling and limited or no backup components.
They are designed primarily for smaller businesses, local office IT systems, testing and development environments where some downtime can be tolerated.
Tier 1 facilities have an expected annual uptime of approximately 99.671%, equivalent to up to 28.8 hours of potential downtime each year.
Their electricity requirements are relatively small, typically ranging from around 10 kW to several hundred kilowatts, and generally below 1 MW.
For comparison, these facilities are a very small load on the electricity system compared with the large-scale data centres now being developed.
2. Hyperscale Data Centres
At the other end of the spectrum are hyperscale data centres. These are enormous computing facilities designed to support cloud computing, large-scale data processing and increasingly AI workloads.
A typical hyperscale facility can contain thousands of servers and require more than 100 MW of electricity. Large facilities can require 100–500 MW of continuous power, with some of the largest AI-oriented campuses moving towards 1 GW of demand.
To put this into simple terms, a single 100 MW data centre represents a very substantial continuous electricity load, while a 1 GW facility approaches the scale of the electricity demand of a major city or hundreds of thousands of homes.
Unlike smaller data centres, these facilities can therefore have a material impact on the electricity grid and may require direct connection to high-voltage transmission infrastructure.
3. NeoCloud / AI-Native Data Centres
The next generation is often referred to as NeoCloud or AI-native data centres. These facilities are specifically designed to support artificial intelligence and high-performance computing rather than simply storing data or hosting conventional IT systems.
Their defining characteristic is extremely high computing and power density.
A traditional data-centre rack might typically require around 10–15 kW, whereas AI-focused facilities can require approximately 80–115 kW per rack, or several times the power density of older facilities.
Individual AI facilities can therefore require 50 MW to more than 100 MW, while larger regional campuses can ultimately reach several hundred megawatts.
This is an important distinction. The issue is not simply that Australia will have more data centres.
The issue is that the new generation of AI data centres can consume vastly more electricity per facility and operate continuously.
Australia's Data Centre Growth
Australia's data-centre market is now entering a period of rapid expansion, with growth increasingly concentrated in large hyperscale and AI-ready facilities.
Deployable capacity is projected to increase from approximately 1,428 MW to more than 3,176 MW by 2030, while the broader development pipeline has been reported at around 6 GW of potential capacity.
Existing operational data centres across the National Electricity Market (NEM) are estimated to use approximately 1.35–1.7 GW of grid capacity, consuming around 4 TWh of electricity each year, equivalent to approximately 2.2% of total grid-supplied energy demand.
More significant, however, is the development pipeline. Between approximately 5.4 GW and 9 GW of additional maximum data-centre demand is progressing through the Australian Energy Market Operator (AEMO) transmission connection process.
These figures should not be interpreted as meaning that all proposed projects will necessarily be built or that all connection applications will translate into actual electricity consumption. They represent the scale of potential demand being considered and connected to the planning process.
The Pipeline and Future Demand
The potential scale of this growth is significant.
- Connection pipeline: AEMO data indicates up to approximately 9 GW of proposed maximum connection capacity across major data-centre projects, concentrated particularly in New South Wales (approx. 60%) and Victoria (approx. 40%).
- Projected demand: Under AEMO's central Step Change scenario in the Integrated System Plan, data-centre electricity consumption could rise to approximately 6% of NEM demand by 2030 and approach 10% by 2050, equivalent to around 34 TWh of annual consumption.
If these projections are realised, data centres will move from being a relatively small component of Australia's electricity demand to becoming a significant and strategically important source of new electricity consumption.
What Is Driving This Growth?
Artificial Intelligence and NeoCloud
The most significant driver is the rapid growth of AI. AI workloads require large numbers of high-performance GPUs operating at very high power densities.
These systems also require sophisticated cooling technologies, including closed-loop cooling systems, and can create very large blocks of new electricity demand over relatively short periods.
Digital Sovereignty
There is also an increasing requirement for sensitive government, corporate and customer data to be stored and processed locally.
This concept of digital sovereignty is encouraging investment in Australian data-centre infrastructure rather than relying entirely on overseas facilities.
Broader Economic Investment
Data-centre development also creates significant secondary investment. New facilities require construction, engineering, telecommunications, power infrastructure, transmission connections, cooling systems and supporting services.
As a result, the economic impact extends well beyond the data centre itself.
The Emerging Constraint: Electricity
This brings the discussion back to the central issue raised in the introduction.
The limiting factor may not be Australia's ability to build data centres. It may be Australia's ability to supply them with sufficient reliable electricity.
Data centres operate differently from many traditional forms of electricity demand. A large AI facility can require hundreds of megawatts of power 24 hours a day, seven days a week.
This means that the electricity must be available not only when renewable generation is abundant, but also during periods of low wind and solar output.
Consequently, the growth of data centres will need to be considered alongside Australia's broader transition towards renewable electricity.
Electricity demand from data centres has been forecast to increase substantially, potentially rising from around 1% of national electricity consumption today to more than 10% by the mid-2030s, depending on how quickly projects are developed and how AI demand evolves.
This creates a potential bottleneck involving generation, transmission, distribution and firming capacity. Australia may need to build significant additional renewable generation, storage and network infrastructure simply to accommodate the new load.
Why This Matters
The numbers demonstrate why the data-centre debate is fundamentally an energy-allocation debate.
Australia wants the economic and technological benefits that come from participating in the global AI economy. At the same time, every additional gigawatt of continuous data-centre demand represents electricity that must be generated, transmitted and delivered.
The question therefore becomes:
How can Australia accommodate the enormous electricity requirements of AI and data centres while ensuring that sufficient affordable and reliable electricity remains available for households, businesses and existing industries?
This is where the data-centre debate moves beyond technology. It becomes a question of energy policy, infrastructure investment, electricity prices and resource allocation—and ultimately, who should pay for the additional energy system required to support Australia's AI future.
Who Ultimately Pays for the Data Centre Energy Expansion?
The previous sections have established that Australia is likely to see a substantial increase in electricity demand from data centres, particularly as AI adoption accelerates.
This raises a further and very important question:
If data centres are prepared to secure their own renewable energy and pay for their own connections, does that mean Australian electricity consumers are protected from the additional costs created by their growth?
At first glance, the answer might appear to be yes.
Data-centre operators are increasingly looking to secure renewable electricity through Power Purchase Agreements (PPAs), direct investment in generation and storage, or other forms of contracted renewable energy. They may also be required to pay the costs associated with connecting their facilities to the electricity network.
This is an important point. Data centres should not be portrayed simply as large electricity consumers expecting the existing electricity system to provide their power without contributing to the cost.
However, paying for a connection is not necessarily the same as paying for all of the infrastructure and system changes that their electricity demand may require.
This distinction is central to understanding the potential cost implications for the wider electricity market.
The Difference Between "Paying for Your Connection" and "Paying for the System"
A large data centre may pay for a dedicated substation, connection assets or other infrastructure required specifically to connect its facility to the grid.
Under a conventional "causer pays" or "user pays" approach, this is relatively straightforward: the party creating the need for the asset pays for it.
The situation becomes more complicated when several large data centres locate in the same geographic area.
If multiple facilities collectively increase demand beyond the capability of the existing network, the network operator may need to strengthen upstream transmission and distribution infrastructure.
This could include larger substations, additional feeders, higher-capacity transformers, protection systems and, in some circumstances, broader transmission upgrades.
Some of these assets may be dedicated to a particular customer. Others may provide capacity that can ultimately be used by multiple customers.
This is where the debate over cost allocation becomes important.
Australia's electricity networks operate under regulated frameworks, and network charges are ultimately recovered from customers through electricity bills.
Network costs can include both transmission and distribution charges, while connection charges can also apply when a new customer joins the network. Maximum demand is an important driver of the network capacity required to supply customers.
Therefore, the relevant question is not simply:
"Will the data centre pay for its connection?"
It is:
"Will the data centre pay for all of the additional network capacity and system investment that would not otherwise have been required because of its demand?"
That is a much more difficult question.
Why Scale Changes the Argument
The scale of the emerging data-centre pipeline makes this issue particularly important.
AEMO reported that, at the end of March 2026, 11 large-scale data-centre projects with a combined maximum demand of approximately 5.4 GW were progressing through the transmission connection process.
Around 60% of this capacity was in NSW and 40% in Victoria, with most projects still in relatively early stages.
These numbers should not be interpreted as 5.4 GW of guaranteed new electricity consumption. Projects can be delayed, resized, relocated or cancelled. Maximum connection capacity is also not necessarily the same as average electricity consumption.
Nevertheless, the figure illustrates the potential scale of the challenge.
For perspective, a continuous 1 GW load operating for a full year would consume approximately 8.76 TWh of electricity.
If several large data centres are developed within the same part of the network, the issue is no longer simply one of connecting individual customers.
It becomes a question of whether the surrounding electricity infrastructure has sufficient capacity to support a new concentration of very large, continuous loads.
This is particularly important because network infrastructure cannot always be built incrementally at exactly the same pace as individual data-centre projects.
The "Clustering" Effect
The potential problem can be illustrated using a simple example.
Imagine one 100 MW data centre connecting to an area of the network with sufficient spare capacity. The required network reinforcement may be relatively limited and can potentially be attributed directly to that customer.
Now imagine ten data centres, each requiring 100 MW, locating within the same region.
The combined potential load is now 1 GW.
The network may need substantially more than ten individual connections. It may require major reinforcement of the local network, additional substations, higher-capacity transmission infrastructure and improved system protection.
At that point, the distinction between customer-specific infrastructure and shared network infrastructure becomes increasingly important.
The economic question becomes:
How much of that additional investment should be paid directly by the data-centre operators, and how much should be recovered through regulated network charges from the broader customer base?
This is the heart of the cost-pass-through debate.
Melbourne – A Case Study in Emerging Scale
Melbourne provides an important example of why this issue deserves attention.
Victoria is already one of Australia's major data-centre markets, and the Melbourne market is expected to grow significantly.
The Victorian Government describes Melbourne as one of Australia's largest data-centre hubs and a market expected to experience substantial growth over the coming decade.
The concentration of facilities in Melbourne's western and northern areas means that the electricity requirements of individual projects need to be considered alongside the cumulative demand created by the broader cluster.
The original analysis underpinning this paper identifies a potential 9 GW data-centre development pipeline in Melbourne and an estimated $16 billion requirement for associated network expansion and reinforcement.
These figures should be treated as pipeline and infrastructure estimates rather than guaranteed expenditure. Not all proposed data-centre capacity will necessarily be built, and not all network investment would necessarily be attributable solely to data centres.
Nevertheless, the example illustrates the fundamental issue: the infrastructure required to support a rapidly expanding data-centre cluster can be considerably larger than the infrastructure required to connect any one individual facility.
Western Sydney – Another Example
Western Sydney provides a similar illustration.
The region has become a major concentration point for hyperscale and other large data-centre developments.
As facilities cluster geographically, the challenge shifts from simply connecting individual sites to ensuring that the surrounding network can reliably supply their combined demand.
The original analysis identifies a very large pipeline of potential data-centre demand in NSW, including figures of up to 28 GW of connection interest, with approximately 13 GW described as being in more advanced connection discussions.
These figures should again be understood as maximum potential connection requirements and development pipeline figures, rather than actual electricity consumption or confirmed projects.
The significance is nevertheless clear. A data-centre pipeline of this magnitude is substantially larger than the existing demand of many parts of the network and would require careful planning of transmission and distribution capacity.
AEMO is already working with network service providers to improve visibility of data-centre connections, recognising that large data-centre loads are developing rapidly and can have material implications for network planning.
The Regulatory Response
The potential for large new loads to place pressure on the electricity system is now being recognised by Australia's energy-market institutions.
The Australian Energy Market Commission (AEMC) has been examining how the National Electricity Rules should deal with large inverter-based loads, a category that can include modern data centres and other large electronically controlled loads.
The AEMC's work reflects a recognition that very large loads are becoming more active participants in the power system and that their connection requirements may need to evolve as their scale increases.
AEMC analysis has considered 30 MW as an important threshold for large inverter-based loads, although the final application of technical requirements depends on the regulatory framework and characteristics of the connection.
This is important because it demonstrates that regulators are not simply allowing very large loads to connect to the existing grid without consideration of their broader system impact.
The objective is to ensure that new large loads contribute appropriately to the costs and technical requirements they create, while maintaining the reliability and security of the electricity system.
What Does This Mean for Electricity Consumers?
The answer is not straightforward.
It would be incorrect to assume that every dollar spent on network infrastructure for data centres will automatically appear on household electricity bills.
Equally, it would be incorrect to assume that because a data centre pays for its own connection, there can be no impact on other electricity consumers.
There are several possible outcomes.
If the infrastructure is dedicated to the data centre
The operator is more likely to bear the associated costs directly.
If infrastructure is required because of a broader cluster of new loads
The allocation of costs becomes more complicated, particularly where the infrastructure provides capacity for multiple customers.
If the infrastructure becomes part of the regulated network asset base
The way those costs are recovered is determined through the regulated network framework and ultimately affects network charges.
If data-centre demand causes additional generation, storage or transmission investment
The costs and benefits extend beyond the immediate connection and become a broader electricity-market issue.
This is why the argument that "data centres will pay for their own infrastructure" is only part of the story.
The Bigger Question: Who Pays for Australia's AI Electricity Requirement?
There is a legitimate argument that data centres can help drive Australia's renewable-energy investment.
Their large and predictable electricity requirements can support new renewable generation, storage and long-term power contracts. Data-centre operators also have strong incentives to secure reliable and increasingly low-emissions electricity for their operations.
However, there is another side to the equation.
If Australia needs to build substantial additional generation, storage, transmission and distribution infrastructure primarily because of the rapid growth of data centres, then the economic question becomes who pays for that additional system and who benefits from it?
This does not mean that data centres are necessarily a burden on Australian electricity consumers. They bring investment, jobs, technology, digital capability and potentially significant economic benefits.
Rather, it means that the cost of enabling their growth needs to be transparent and appropriately allocated.
Are Data Centres Water Monsters?
The debate around data centres is not limited to electricity. Another increasingly important question is how much water these facilities require.
Data centres generate enormous amounts of heat, particularly those designed to support artificial intelligence. Thousands of high-performance processors can operate continuously, producing heat that must be removed to keep the equipment operating safely and reliably.
This has led to growing concern that the rapid expansion of data centres could create another significant demand on Australia's already valuable water resources.
But is the picture really as alarming as some headlines suggest?
The answer is more complicated than simply saying that data centres consume huge amounts of water.
The amount of water a data centre requires depends heavily on the type of cooling technology it uses, its location, the climate in which it operates, the density of its computing equipment and whether it uses potable, recycled or reclaimed water.
Understanding these differences is important when considering the potential impact of Australia's growing data-centre industry.
Why Do Data Centres Need Water?
The fundamental problem is heat.
High-density server racks, particularly those used for AI and high-performance computing, can generate enormous amounts of heat. That heat must be continuously removed to prevent equipment from overheating and to maintain reliable operation.
Traditionally, many large data centres have used chillers and cooling towers, where water plays an important role in transferring heat away from the computing equipment.
One of the main concerns with this approach is evaporative cooling.
When water is used in cooling towers, some of it evaporates into the atmosphere and is therefore permanently lost from the local water system. Depending on the design and operating conditions of the facility, evaporative losses can represent approximately 15–25% of the cooling water used, and in some circumstances can be higher.
This is fundamentally different from simply using water and returning it to the system. Evaporated water has effectively left the local water supply and must be replaced.
Why AI Makes the Issue More Important
The rapid development of AI introduces another dimension to the water debate.
AI systems require significantly more computing power than many traditional digital services. Training large AI models and processing the enormous number of requests generated by users can require thousands of high-performance processors operating simultaneously.
More computing means more heat.
More heat means more cooling.
And, depending on the cooling technology used, more cooling can mean more water consumption.
This does not mean that every AI query consumes a fixed or significant quantity of water. Water consumption varies substantially depending on the data centre, its cooling system, climate, workload and source of water.
Nevertheless, the underlying relationship is important: as AI computing demand increases, the potential demand for cooling—and therefore water—also increases where water-intensive cooling systems are used.
Not All Data Centres Are "Water Monsters"
It is important not to assume that every new data centre will place the same demand on Australia's water resources.
Cooling technology is changing rapidly.
Traditional cooling systems can rely heavily on evaporative water cooling, but newer AI-focused facilities are increasingly adopting technologies designed to reduce or eliminate ongoing water consumption.
One example is closed-loop liquid cooling.
In a closed-loop system, a fixed volume of water or another cooling fluid circulates through sealed pipes or cooling equipment, absorbs heat from the servers and transfers that heat to an external heat exchanger or radiator.
Unlike evaporative cooling towers, the cooling fluid is not continuously released into the atmosphere.
This means that once the system is filled, ongoing water consumption can be dramatically reduced.
Alternative Cooling Technologies
Recycled and Reclaimed Water
One of the most straightforward approaches is to avoid using drinking- quality water wherever possible.
Instead, data centres can use treated wastewater, recycled water or other reclaimed sources for cooling. This reduces competition between data-centre operations and the water required for households and other essential community uses.
Closed-Loop Liquid Cooling
Direct-to-chip and other liquid-cooling technologies can transfer heat much more efficiently than conventional air cooling. Because the cooling fluid can circulate within a closed system, ongoing water consumption can be substantially reduced.
This technology is particularly relevant to AI data centres because modern GPUs generate far more heat than traditional computer equipment.
Air Cooling
In cooler climates or during cooler periods of the year, outside air can sometimes be used to assist with cooling. This can reduce the amount of mechanical cooling—and therefore water—required.
The suitability of this approach depends heavily on local climate and the design of the facility.
The Australian Water Question
The water issue is particularly relevant to Australia.
Australia is a country where water availability varies considerably between regions and where drought, population growth and climate variability can place significant pressure on water resources.
This raises an obvious question:
If Australia is preparing for a substantial expansion in data-centre capacity, how do we ensure that the water required to operate these facilities does not compete with the water required by Australia's population, agriculture and other essential industries?
The answer is unlikely to be simply to prevent data centres from using water.
Instead, the focus should be on where the water comes from, how efficiently it is used and whether potable water is genuinely necessary.
A data centre using recycled wastewater in a region with adequate reclaimed-water infrastructure presents a very different resource challenge from a facility relying heavily on drinking-quality water in a region already experiencing water stress.
This makes the location and design of data centres important considerations when assessing their overall environmental impact.
Managing Water Use Through Better Design
The emerging approach is therefore increasingly focused on water efficiency rather than simply water consumption.
Regional governments and regulators are beginning to place greater emphasis on efficient cooling technologies and the use of non-potable water sources. In NSW, for example, government planning and policy discussions around data-centre development are increasingly considering water availability and infrastructure requirements alongside electricity and land requirements.
This is an important development because the water impact of a data centre is not determined solely by the size of the facility.
Two data centres of identical computing capacity could have very different water footprints depending on their cooling systems and water sources.
The Bottom Line
The concern about data centres becoming "water monsters" is therefore not without foundation—but it needs to be put into context.
Large AI data centres can generate enormous heat loads, and facilities using evaporative cooling can consume significant quantities of water. As AI computing demand grows, this issue deserves serious consideration.
However, technology is also providing solutions.
- Recycled and reclaimed water can reduce reliance on drinking water.
- Closed-loop and direct-to-chip cooling can significantly reduce ongoing water consumption.
- Air cooling can reduce water requirements where climate and facility design allow.
- Better planning and regulation can ensure that water-intensive facilities are not developed in locations where they would place unacceptable pressure on local water supplies.
The key issue is therefore not whether data centres use water—they do.
The more important question is how much water they use, what type of water they use, where that water comes from, and whether their water consumption is sustainable in the region where they operate.
Just as with electricity, Australia will need to balance the benefits of data centres and AI against the finite resources required to support them.
The objective should be to ensure that Australia's AI future does not come at the expense of the country's fundamental requirement for secure, affordable and sustainable water supplies.
Conclusion
Australia should embrace the growth of data centres and the enormous opportunities that AI can deliver—but we must be equally clear about who pays for the resources and infrastructure required to support them.
The objective should be simple: data centres should pay a fair and transparent share of the additional electricity, network and water infrastructure they create, without unintentionally transferring those costs to households and businesses that did not create the demand.
This is not a debate about data centres versus Australian consumers. It is about finding smarter ways to accommodate Australia's AI future while protecting the affordability and reliability of our essential resources.
If we have the ingenuity to build AI, surely, we have the ingenuity to solve the challenges it creates.
Perhaps, in the end, we should just ask ChatGPT for the answer.