Australia’s new AI standards put infrastructure at the centre of data centre growth

In July, Australia announced plans for national artificial intelligence (AI) standards that could significantly affect the data centre sector. 

While much of the discussion around AI regulation focuses on software, copyright, and responsible use, the Government's announcement also addresses the physical infrastructure needed to support growth. 

This could include expectations around securing new power supply, contributing to network connection costs, managing demand during constrained periods, improving water efficiency and working with state governments and local communities on site selection. 

The final requirements are still to be confirmed. However, the direction is clear. Future developments will need to demonstrate that they can support Australia's AI ambitions without transferring disproportionate energy, water, and infrastructure costs to surrounding communities. 

National consistency could provide greater certainty for developers, but implementation will still need to reflect differences between states, including local grid conditions, regulations and water availability. 

Power availability will shape development

The challenge presented by AI data centres is not simply that they use large amounts of electricity. High-density Graphics Processing Unit (GPU) clusters create substantial, continuous demand at a grid connection point, placing sustained pressure on generation, substations and transmission networks that may not have been planned around such rapid growth. 

AI infrastructure can also create rapid changes in demand, while power-electronic loads introduce considerations such as harmonics, reactive power, and system strength.

Power availability is therefore becoming a critical development decision. Grid upgrades and new renewable generation frequently take longer to deliver than the data centre itself, meaning the connection can quickly become the bottleneck. Developers must examine the timing and quality of the available power supply before committing to a site or delivery programme. 

Renewable procurement is only part of the answer

Renewable Energy Certificates and virtual Power Purchase Agreements can help finance renewable generation and support emissions reporting. However, they do not resolve local network constraints or guarantee that clean electricity is available at the time and place it is consumed. 

A credible energy strategy must also consider how supply will be delivered and aligned with demand.

This could involve grid electricity, dedicated renewable projects, on-site or co-located generation, and battery storage. The objective should not necessarily be complete site-level energy independence. For most projects, this would be impractical and unnecessarily costly. 

Instead, developers should focus on supporting additional supply, aligning generation with consumption and demonstrating value to the electricity system. The balance will depend on network capacity, land availability, commercial arrangements and the operational requirements of the facility.

Coordinating storage, workloads and grid demand

Battery energy storage systems (BESS) can store renewable electricity, helping to reduce peak demand and support the facility in periods ranging from minutes to hours. Uninterruptible Power Supply (UPS) systems and supercapacitors can respond more quickly to demand and are better suited for short-duration events closer to the computing equipment itself. 

Whilst they differ, these should not be designed and operated in silos. 

An effective energy management system should coordinate holistically the generation, batteries, UPS equipment and computing demand involved. This allows each component to respond at the appropriate time without duplication. 

Some computing activity will likely be flexible. Non-critical workloads could be moved away from periods of network constraint, and battery charging can be delayed or reduced in times of need. 

With AI, it is slightly different. Not every AI workload can be shifted. Training programmes, customer commitments, and resilience requirements will limit the flexibility operators have in their facilities. All of this can only be found out by understanding the individual facility's workload and demands. 

Power availability will shape development

The challenge presented by AI data centres is not simply that they use large amounts of electricity. High-density Graphics Processing Unit (GPU) clusters create substantial, continuous demand at a grid connection point, placing sustained pressure on generation, substations and transmission networks that may not have been planned around such rapid growth. 

AI infrastructure can also create rapid changes in demand, while power-electronic loads introduce considerations such as harmonics, reactive power, and system strength.

Power availability is therefore becoming a critical development decision. Grid upgrades and new renewable generation frequently take longer to deliver than the data centre itself, meaning the connection can quickly become the bottleneck. Developers must examine the timing and quality of the available power supply before committing to a site or delivery programme. 

Renewable procurement is only part of the answer

Renewable Energy Certificates and virtual Power Purchase Agreements can help finance renewable generation and support emissions reporting. However, they do not resolve local network constraints or guarantee that clean electricity is available at the time and place it is consumed. 

A credible energy strategy must also consider how supply will be delivered and aligned with demand.

This could involve grid electricity, dedicated renewable projects, on-site or co-located generation, and battery storage. The objective should not necessarily be complete site-level energy independence. For most projects, this would be impractical and unnecessarily costly. 

Instead, developers should focus on supporting additional supply, aligning generation with consumption and demonstrating value to the electricity system. The balance will depend on network capacity, land availability, commercial arrangements and the operational requirements of the facility.

Coordinating storage, workloads and grid demand

Battery energy storage systems (BESS) can store renewable electricity, helping to reduce peak demand and support the facility in periods ranging from minutes to hours. Uninterruptible Power Supply (UPS) systems and supercapacitors can respond more quickly to demand and are better suited for short-duration events closer to the computing equipment itself. 

Whilst they differ, these should not be designed and operated in silos. 

An effective energy management system should coordinate holistically the generation, batteries, UPS equipment and computing demand involved. This allows each component to respond at the appropriate time without duplication. 

Some computing activity will likely be flexible. Non-critical workloads could be moved away from periods of network constraint, and battery charging can be delayed or reduced in times of need. 

With AI, it is slightly different. Not every AI workload can be shifted. Training programmes, customer commitments, and resilience requirements will limit the flexibility operators have in their facilities. All of this can only be found out by understanding the individual facility's workload and demands. 

Cooling decisions must consider water and energy together

Water has grown in visibility as a large factor in site selections and approvals. AI infrastructure creates concentrated heat loads. This requires cooling systems able to support dense, power-intensive equipment. Different approaches involve compromises between electricity use, water consumption and reliability. 

Cooling design inevitably has trade-offs. Evaporative cooling systems use less energy but come with higher water consumption. Dry cooling places less pressure on water use but demands more power. Closed-loop and direct-to-chip cooling are effective at managing the higher heat loads from AI equipment, but their effectiveness depends on how they are integrated into the wider cooling system. 

There is no universally sustainable cooling solution.

This is why Power Usage Effectiveness should not be considered in isolation. A low PUE does not necessarily represent the best environmental outcome. This is particularly true if it is achieved through high water consumption in a water-constrained region. 

Water strategy must also begin at site selection. The availability of recycled water, treatment infrastructure and alternative heat rejection options can determine whether a location is suitable before the cooling system is designed. 

Waste heat recovery may provide further benefits where there is a viable nearby user, but the opportunity will remain dependent on local demand. 

Site selection as a sustainability decision

The proposed standards could impact where AI infrastructure is developed. The strongest locations will combine grid capacity with access to renewable generation, water resources, land availability, and resilient fibre connectivity. Some may also provide opportunities for the site to support the network via energy storage, flexible demand strategies, or heat reuse. 

However, few sites will offer every advantage. Developers will need to evaluate the trade-offs and design around the constraints of each location. 

As a result, power, computing, cooling, and water infrastructure needs to be designed as one. Optimising one element without examining the effects on the others risks turning a success into a failure. 

Moving to an infrastructure partner

Australia should not need to choose between AI growth and responsible infrastructure development. 

However, the next generation of large data centres will need to do more than secure a grid connection and purchase renewable certificates. Projects will be expected to support additional energy supply and contribute to the wider network infrastructure. They will also need to manage their demand practically and use water responsibly. 

This doesn't mean every data centre should be designed to be self-sufficient. It does mean that large AI campuses will need to become more active participants in the systems that support them. 

If the final standards focus on measurable outcomes rather than regulatory box-ticking, investors could benefit from greater certainty whilst protecting Australia's energy and water infrastructure. 

The most successful AI campuses will look less like standalone data centres and more like integrated ecosystems. But this can only be done when systems are planned together, from the start.

Water has grown in visibility as a large factor in site selections and approvals. AI infrastructure creates concentrated heat loads. This requires cooling systems able to support dense, power-intensive equipment. Different approaches involve compromises between electricity use, water consumption and reliability. 

Cooling design inevitably has trade-offs. Evaporative cooling systems use less energy but come with higher water consumption. Dry cooling places less pressure on water use but demands more power. Closed-loop and direct-to-chip cooling are effective at managing the higher heat loads from AI equipment, but their effectiveness depends on how they are integrated into the wider cooling system. 

There is no universally sustainable cooling solution.

This is why Power Usage Effectiveness should not be considered in isolation. A low PUE does not necessarily represent the best environmental outcome. This is particularly true if it is achieved through high water consumption in a water-constrained region. 

Water strategy must also begin at site selection. The availability of recycled water, treatment infrastructure and alternative heat rejection options can determine whether a location is suitable before the cooling system is designed. 

Waste heat recovery may provide further benefits where there is a viable nearby user, but the opportunity will remain dependent on local demand. 

Site selection as a sustainability decision

The proposed standards could impact where AI infrastructure is developed. The strongest locations will combine grid capacity with access to renewable generation, water resources, land availability, and resilient fibre connectivity. Some may also provide opportunities for the site to support the network via energy storage, flexible demand strategies, or heat reuse. 

However, few sites will offer every advantage. Developers will need to evaluate the trade-offs and design around the constraints of each location. 

As a result, power, computing, cooling, and water infrastructure needs to be designed as one. Optimising one element without examining the effects on the others risks turning a success into a failure. 

Moving to an infrastructure partner

Australia should not need to choose between AI growth and responsible infrastructure development. 

However, the next generation of large data centres will need to do more than secure a grid connection and purchase renewable certificates. Projects will be expected to support additional energy supply and contribute to the wider network infrastructure.

They will also need to manage their demand practically and use water responsibly. 

This doesn't mean every data centre should be designed to be self-sufficient. It does mean that large AI campuses will need to become more active participants in the systems that support them. 

If the final standards focus on measurable outcomes rather than regulatory box-ticking, investors could benefit from greater certainty whilst protecting Australia's energy and water infrastructure. 

The most successful AI campuses will look less like standalone data centres and more like integrated ecosystems. But this can only be done when systems are planned together, from the start.