The government wants AI data centres to minimise water use and maximise energy efficiency, but those rules will mean little without public, facility-level measurement of their real impact.
Two lines from the recent announcement by Australia’s Prime Minister on artificial intelligence (AI) should be read side by side.
First: “Last week, the Department of Employment and Workplace Relations published real-time, data-driven analysis of AI’s impact on the labour market.”
Second: “Which is why our rules will require data centres to minimise their water use, maximise their energy efficiency, and pay for any additional water infrastructure required.”
Put together, they point to a gap the government may not have intended. On employment, it has established a monitoring framework – an independent, data-driven way to track the relationship between AI and jobs, and one that carefully notes its findings are not yet definitive. On water and energy, no comparable framework has been announced.
Requiring data centres to minimise their water use is welcome. But minimise against what baseline? And the rule skips the prior questions entirely: how much water will these data centres use, where will it come from and how much is sustainably available in the first place?
Efficiency alone is not enough. A highly efficient data centre can still consume a very large amount of water and electricity, and could still mean more water drawn from a catchment that may have none to spare.
Requiring operators to pay for any additional water infrastructure begs the questions against what baseline, boundary, time period — it presupposes a measurement and reporting system that has not yet been announced.
The promise that data centres will be “net generators” of energy rather than net users raises similar obvious questions: netted over which boundary and time period, and verified by whom?
Here is the awkward fact. Data centre water and energy use is not absent from Australia’s official statistics but it cannot be seen. It is buried inside broader industry aggregates. As the Australian Bureau of Statistics (ABS) set out in a recent spotlight on data centres, the businesses that run them sit inside the Information Media and Telecommunications division, their construction inside “commercial building not elsewhere classified”, and their energy and water use inside the national Energy and Water Accounts.
It is counted – but it cannot be separately identified by facility, operator or catchment. We can require an operator to use less, but we cannot yet say how much any of them use, where, or how that use behaves when the catchment runs dry, which is precisely when it matters most.
Data centres use water in two different ways: directly for cooling, and indirectly via the energy they consume. The first is the water used on site to cool the servers; a direct, local claim on the host city’s supply. The amount consumed is determined by the cooling technology used.
The second is the water used to generate the electricity the data centre needs. This water comes from wherever the power is made, often in another catchment or another state.
The direct use occurs at the facility and can, in principle, be measured and attributed there. The indirect use is recorded, if it is recorded at all, against the electricity producer rather than the data centre. And for understanding scarcity, the critical measure is water consumed – evaporated or otherwise not returned to the same water system – not merely the volume withdrawn and partly returned. Confuse withdrawal with consumption, or direct use with electricity-related use, and headline figures become incomparable and potentially misleading.
There is a deeper asymmetry here and it is the one that should trouble the Treasury. We treat the national economic accounts – gross domestic product (GDP) and its components – as serious infrastructure, reported quarterly by the Australian statistician. We treat the environmental cost of the same activity as an afterthought.
With AI, the gap is especially stark. Data centre investment and operating activity do enter the economic accounts, but the broader productivity gains attributed to AI remain hard to isolate, and the associated water and energy demands cannot be identified independently at the facility and catchment levels where their effects are felt. One side is recorded in established economic accounts; the other is scattered through company disclosures and broad environmental aggregates.
This is not an argument against data centres. It is an argument against flying blind. You cannot weigh a benefit against a cost when only one side is on the books.
We do not need a new conceptual framework. The System of Environmental-Economic Accounting (SEEA), the international statistical standard and an extension of the System of National Accounts (SNA), already provides the way to record the water and energy an industry uses alongside the value it creates.
Australia already compiles SEEA-based water and energy accounts. What is missing is the industry detail, the spatial resolution and the mandatory reporting needed to identify data centres and connect their demands to particular grids, water supply systems and catchments.
The legislation itself is the place to fix this. It could require each large facility to report its annual and peak electricity demand; the source and timing of any new generation and firming; water withdrawal and consumption by source; its cooling technology and its water and energy efficiency; and how it plans to operate during drought or grid stress. Regulators and utilities could collect the facility-level information, with the ABS compiling consistent industry and regional aggregates – at the site, water supply system and catchment level, not simply the “region”. That would provide timely data, reduce the burden of a stand-alone survey and give the “framework” some substance.
Over time, Australia’s digital economy statistics could also be extended to identify data centre activity more clearly and to improve estimates of AI-related output and productivity – so that, eventually, the benefit is counted as carefully as the cost.
The labour market work the Prime Minister cited is the proof of concept. Water and energy deserve the same independent, data-driven monitoring.
There is still time to do this in the right order. National Cabinet will consider the framework next month, and legislation is planned for early next year. The measurement and reporting requirements should be designed around the substantive obligations, not bolted on after facilities have been approved and the water and energy have already been committed.
I set out the underlying problem that Australia is expanding AI data centres faster than it is measuring their water and energy use in The Conversation last month.
Republished from Global Water Forum
Michael Vardon
Dr Michael Vardon is an Associate Professor at the ANU Fenner School of Environment and Society (ANU) where he researches and teaches environmental accounting. His knowledge of environmental accounting spans the collection of basic data, account compilation, analysis and indicators as well as applications to public policy.