A Public Interest Test for Australian Data Centres

By Dr Brydon Wang •

Australia needs data centres, but capital investment alone cannot determine which projects receive priority access to constrained infrastructure. The Benevolent Infrastructure Test assesses Australian data centre projects through three signals of trustworthiness: ability, integrity and benevolence. Treating benevolence as germinal, it asks how community and Traditional Owner interests and vulnerabilities should shape a project; whether those objectives are translated into clear standards and legislative requirements; and how these data centres are able to build national capability while strengthening the energy and water systems on which they rely.

Australia is experiencing an extraordinary wave of investment in data centres. With the highest land availability per capita and robust political systems, it was ranked second to the US for data centre investment destinations in 2024 (Mandala Partners, see below). These facilities support services that modern life is increasingly dependent on: banking, healthcare, government, communications, scientific research, cloud storage, cybersecurity, and the everyday operation of Australian businesses. The growth of artificial intelligence has accelerated the demand for data centres globally because the training requirements and operational needs of large models can require unusually concentrated and continuous computing, electricity and water for cooling.

Mandala Partners, global data centre investment destinations 2024

Australia’s political stability, established legal institutions, its current proportion of renewable energy generation and proximity to growing Asia-Pacific demand have made it an attractive destination for data centre investment. The project pipeline is concentrated particularly around Sydney and Melbourne, with growing interest in Queensland, Western Australia and the Northern Territory.

A data centre is a physical facility containing racks of computer servers and the equipment needed to power, cool and connect them to the wider world. But not all data centres operate in the same way, with different facilities offering distinct services and performing varying levels of function. Some host government or critical infrastructure workloads, others provide cloud services to a gamut of customers, and yet others train or operate artificial-intelligence models for users around the world. The focus of this essay is on the very large iterations of these facilities, whose resource requirements have started to reshape electricity, water and land-use planning.

Despite servicing the virtual world, data centres consume and occupy physical resources. They compete for finite land, electricity, water, network capacity and skilled labour. But while data centres require resources, they convert these resources into computational capacity that is not directly tangible and, as such, the public is less able to see these benefits. And despite the prospect that these data centres can improve digital resilience, enable and support research and create new forms of economic activity, decision-making around where to site these facilities and why they have received the requisite approvals is not always transparent.

Infrastructure under conditions of scarcity

For much of the twentieth century, infrastructure planning operated under an assumption of expansion. Growing populations required more of the raw ingredients that make up a city, be they homes and offices, sites of manufacturing and distribution, or roads and rail bringing people and goods in from beyond the city. These building blocks consumed resources that also needed to be distributed, including energy, water and, more recently, data infrastructure. Difficult decisions were made about how improvements would be funded and where they would be located, but the prevailing assumption was that capacity could generally be expanded to meet demand.

We are familiar with debates about how projects should be delivered and where they should be sited. We are less practised at deciding how the benefits and burdens of entirely different forms of infrastructure should be compared.

People working in government will recognise the slightly absurd waiting game that follows the cancellation of a major project. Departments, proponents and consultants race to place replacement proposals before central agencies, each shiny business case jostling for position as the best use of suddenly liberated capital. But the project that ultimately succeeds in getting funded is a result of judgement about public value and whose claim should take priority when every demand cannot be met.

Infrastructure occupies a central position in this relationship because every major investment redistributes vulnerability. New transmission infrastructure may strengthen reliability for some users while altering landscapes and land uses for others; transport infrastructure can create opportunities for participation while reshaping neighbouring communities. A data centre may create national computational capability while concentrating electricity and water demands within a particular network or catchment area, creating lumpy benefits and burdens for the community.

But infrastructure governance rarely involves straightforward comparisons. We are not necessarily choosing between one transmission line and another, but between entirely different forms of public value. It is not even apples and oranges. To stretch the analogy, the exercise is often an absurd comparison of the apples, the mule needed to transport them to market and the life buoy required because the farm hand has already fallen into the dam twice. Each has value, but there is no common unit through which that value can be neatly calculated.

While there is existing recognition that governments need to identify worthwhile projects, the challenge we have is the 'how?'. How do we establish institutions capable of making visible and defensible judgements between competing forms of societal need? My research responds to this challenge by connecting public confidence to the three signals of trustworthiness:

1. ability: the technical and functional capability offered by the proposed data centre, including whether it can use and contribute to electricity, water and other infrastructure systems effectively;

2. integrity: the alignment between the project's stated objectives, its actual operation and the laws, standards and public expectations applying to data centres in Australia; and

3. benevolence: whether individuals and communities can see that their interests and vulnerabilities have been recognised, and that decision-makers have a positive orientation towards them when weighing competing benefits and burdens.

Although these signals are conventionally presented as ability, integrity and benevolence (Mayer et al, 1995), my research argues that benevolence is germinal. That is, the interests and vulnerabilities identified through benevolent infrastructure design and delivery establish the objectives of the assessment. These objectives are then translated into standards, regulatory frameworks and institutional requirements that give us the measure of what is integrous decision-making. These integrity requirements then inform the technical briefs, infrastructure designs and operational capabilities through which ability is demonstrated.

The Benevolent Infrastructure Test therefore takes benevolence as its starting point, even where the assessment is presented through the more familiar sequence of ability, integrity and benevolence.

Ability: National capability and system functionality

The signal of 'ability' is concerned with whether a data centre can deliver the functions and national capability claimed for it. Ability examines whether the data centre can do so within the physical limits of the infrastructure systems on which it relies. Examples of signals of ability may include: assured access to computational capacity, research partnerships, local skills and knowledge transfer (all within the catch cry of 'jobs, jobs, jobs!'), supply-chain development and opportunities for Australian businesses.

Australia’s electricity system is already being reshaped by the scale and speed of this demand. AEMO has had to revise its estimates on data centre development time and time again as appetite for Australian resources to drive the design and development of data centres accelerates (see below). These forecasts matter because electricity and water systems are planned over years and decades, while AI investment decisions are increasingly made over months. This mismatch in timeframes has contributed to a departure from conventional approaches to connection strategies, with governments and network operators asked to accommodate very large loads and investigate what generation, flexibility, operational performance and network contributions should accompany them in very short timeframes.

Some proponents are exploring a ‘power first, grid later’ approach as network constraints increase and connection queues lengthen (ie. the Bring Your Own Power (BYOP) approach). Industry is moving quickly towards embracing behind-the-meter generation, dedicated renewables, battery storage and other firming arrangements in project design from the outset.

The debate about gas at the 2026 Data Centres Power and Water Summit exposed a difficult tension. Gas-fired generation can provide firm and responsive power while proponents wait for grid connections, transmission upgrades and renewable generation. But in the panel that I chaired, panellists approached gas from different positions that circled the critical question: was gas meeting an enduring technical need or allowing AI investment to move faster than the supporting electricity system? And would this then result in stranded assets particularly as infrastructure described as temporary may operate for decades once capital has been committed and commercial expectations have formed around it.

What emerged also from the Summit was that while industry had come up with creative and rapid solutions to the energy conundrum, the need for water produces fewer easy answers. While a project proponent can contract with a generator hundreds of kilometres away, a project is not so readily detached from the local climatic conditions and water systems, or the maturity of water infrastructure at its chosen site. Despite these constraints, moves towards providing early guidance to project proponents on the location of recycled water assets (see below), more innovative uses of closed-loop systems, immersion cooling and direct-to-chip cooling provide alternatives to the demand on drinking water. And while some facilities operate with very low levels of operational water consumption, these technologies involve trade-offs, including between water efficiency and electricity use.

Data centres do not extract and consume from an abstract national pool of water. Their effects concentrate around particular catchments, treatment systems, pipelines, industrial precincts and communities. It is estimated that data centres currently consume approximately 3.5 billion litres of Sydney’s drinking water each year. Sydney Water is thus planning for a significant increase in demand while investigating recycled-water infrastructure and other alternatives. But the policy significance of Sydney Water's approach is clear: water cannot be treated as an afterthought once location and cooling architecture have been determined because water infrastructure cannot expand as rapidly as digital infrastructure.

But a more interesting aspect of ability is its commercial aspects. It asks more than the question of 'can the technology perform' but whether the business case actually holds water. If Australian data centre proponents are locking in land, grid connection and water allocations based on projected AI compute demand (ie the demand for tokens) and that demand curve is subsequently undercut by cheaper, capable open-weight alternatives (what Scott Galloway has described as 'AI dumping', where free or near-free models undercut Western AI pricing structures), the community is left holding the infrastructure commitments after the capital has moved on (a classic stranded asset scenario).

This raises a second, related question. My research on offshore development and China's recent underwater data centre (following Microsoft's earlier Project Natick trials) suggests a different way of framing 'ability' altogether: if we're able to site these facilities such that they do not compete with the community for scarce land and freshwater, does that 'ability' fundamentally change the public interest calculus, and in particular, the benevolence signal, even before questions of stranded demand arise?

Integrity: Aligning public purpose, standards and performance

The Australian Government has announced stronger expectations for large data-centre developers, including underwriting new power supply, bearing connection costs and addressing energy and water use. These developments are significant. But even with these measures in place, a project may satisfy technical requirements, fund its augmentation and reduce potable-water consumption while contributing relatively little to Australian capability or the community required to host it. Such technical compliance provides us with a baseline for us to measure against but it does not necessarily establish the standard we can use to determine and compare public value.

Integrity requires consistency between the commitments used to justify a data centre and the way the facility is designed and operated over its project lifecycle. Claims about renewable energy, temporary generation, water efficiency, sovereign capability and national benefit should be tested rather than assumed. Such a public interest test could examine the operating life, water use and emissions intensity of energy generation, as well as the milestones for grid and water connection, responsibility for stranded assets and how the facility will function during events of system stress.

One of the ways we can interrogate the public interest of data centres is to treat their output (tokens) as a rough proxy to test against. The idea of exporting ‘tokens’ is, perhaps, an imperfect analogy. Tokens are units through which the inputs (prompts) and outputs of many generative AI systems are processed, and they are now being used to assess the productivity and value of data centres. Importantly, questions are being raised as to how Australia can continue to benefit from the tokens generated across the operating life of the data centre as it consumes Australian energy, water and land. However, tokens are not a reliable measure of every service performed by a data centre or a ready-made unit of taxation. Despite this imprecision, the analogy captures the genuine concern that Australia may supply the physical resources through which computational value is produced without retaining a proportionate share of that value.

This concern echoes current debates dissecting the overall contribution from the LNG growth wave (2007 to 2018) within the latter half of the mining boom. The Petroleum Resource Rent Tax (PRRT) has been cited in debates about what can occur when the design of public returns remains unresolved while export infrastructure is being established. Accumulated deductions and uplift arrangements can defer the return received by the Australian community even when export volumes are substantial. The lesson for data centres is not that tokens should be taxed like gas, but that if Australia is to become one of the largest exporters of tokens, whether Australian households and businesses should pay global market prices for these generated tokens or receive a clear benefit by having lower domestic prices.

Integrity therefore requires the public return to be identified at the outset of a decision-making challenge. Such a public return should be expressed through clear standards and commitments capable of being assessed over the operating life of the facility.

Benevolence: Community benefit and the distribution of vulnerability

Every major infrastructure decision redistributes vulnerability. Data centres may strengthen national resilience and economic capacity while concentrating burdens within particular communities and infrastructure systems. Benevolence, as a signal of trustworthiness, requires visible care for those exposed to a decision and an honest accounting of how competing needs are weighed.

Benevolence is not demonstrated by adding a community benefit package after the principal decisions about location, scale, electricity and water have already been made. Instead, it requires affected individuals, communities and Traditional Owners to be identified early enough for their interests and vulnerabilities to influence the objectives and design of the project. Community benefit should also be examined over the operating life of the facility. This is because data centres are highly capital intensive and may employ relatively few people once construction is complete (repeating some of the challenges with gas infrastructure investment). Short-term construction activity and investment should not be falsely compared and served as an answer to decades of land, electricity and water use.

Benevolence requires going beyond these simplistic comparisons of numbers into the granular details of benefits that look to long-term local procurement, apprenticeships and transferable skills (jobs, jobs, jobs!), continued engagement with Traditional Owners and other adjacent infrastructure improvements. Commitments made to the community in relation to these proposed benefits need to be monitored and enforceable, including complaints and dispute handling processes that meet the expectations of all parties and allow benefit-sharing commitments to continue flowing rather than be caught up in red tape and judicial case handling.

Critically, we are seeing movements to show regulation of content in copyright law reform. But the regulation of the internet, AI and data centre infrastructure requires us to consider the regulatory frameworks at the content layer, logical layer (platform and software) and the infrastructure layer (pipes and wires).

Again, the assessment of benefit to the community cannot be viewed through just one layer and allowing legal reforms in that layer to supersede the need to consider the community benefits and value may sit with the other two layers of the internet. That is, we must pay close attention to how law reform to benefit the community must occur alongside the regulation we develop to guide infrastructure development, as well as regulation to ensure clear returns on platform revenues and computational services (logical layer) enabled by the infrastructure layer. On this point, it is worthwhile re-emphasising the physical nature of data centres. As we build infrastructure that connects us virtually, communities continue to be formed in parks, libraries, main streets, sporting fields, cafés and public squares. Investment in digital connection should remain attentive to the physical places and systems through which collective life is sustained.

This essay proposes the Benevolent Infrastructure Test, which will be developed in greater detail in a subsequent paper. The test requires transparent examination of how decisions are made about the allocation of scarce resources, including industrial land, skilled labour, enabling infrastructure, water capacity and government attention. It asks who controls and benefits from the capability created, what the project contributes to the systems it depends on, and how its benefits and burdens are distributed among affected communities.

Australia needs data centres. Strong projects should be able to demonstrate why they deserve priority, with capital investment assessed alongside the capability to be generated, the systems strengthened and the value shared. The proposed Benevolent Infrastructure Test would make that judgement visible and permit examination of these decisions to ensure they do not arise from knee-jerk reactions particularly in the face of scarcity. This is what benevolent infrastructure should require.

Author's note: The images presented in this cornerstone essay are from the 2026 Australian Data Centres Power and Water Summit, Sydney, 10-11 June 2026.

About the Author

Dr Brydon Wang is an energy and infrastructure adviser, lawyer and scholar whose research examines trustworthiness in artificial intelligence, automated decision-making and urban development. He is an Adjunct Associate Professor at the Centre for Policy Futures, University of Queensland, and chaired the second day of the 2026 Australian Data Centres Power and Water Summit.

Related research

• Brydon T Wang (2023). An Updated Model of Trust and Trustworthiness for the use of Digital Technologies and Artificial Intelligence in City Making. MAB '23: 6th Media Architecture Biennale Conference, Toronto, Canada, 14 - 23 June 2023. New York, NY United States: Association for Computing Machinery. https://doi.org/10.1145/3627611.3627618

• Wang, Brydon (2021). The seductive smart city and the benevolent role of transparency. Interaction Design and Architecture(s) 48 100-121. https://doi.org/10.55612/s-5002-048-005

• Brydon Timothy Wang (2022). The role of trustworthiness in automated decision-making systems and the law. PhD Thesis, School of Law, Queensland University of Technology. https://doi.org/10.5204/thesis.eprints.231388

• Brydon T Wang and Mark Burdon (2021). Automating trustworthiness in digital twins. Automating cities: design, construction, operation and future impact. Edited by Brydon T. Wang and C. M. Wang. Singapore, Singapore: Springer Singapore. 345-365. https://doi.org/10.1007/978-981-15-8670-5_14

• (For how integrity is translated into enforceable standards, technical briefs and accountable decision-making) Brydon Wang and Mark Burdon (2021) Augmenting Superintendent Discretion: Trustworthiness and the Automation of Construction Contracts. ANU Journal of Law and Technology 2(1) https://anujolt.org/article/24468-augmenting-superintendent-discretion-trustworthiness-and-the-automation-of-construction-contracts

• (For more information on tokens and AI processing) Wang, Brydon T. (2024). Prompts and large language models: a new tool for drafting, reviewing and interpreting contracts? Law, Technology and Humans 6 (2) 88-106. https://doi.org/10.5204/lthj.3483