Australia’s governments have agreed to develop mandatory national standards for large data centres, setting the framework for federal regulation of the rapidly expanding sector while leaving some important questions around energy sourcing to be resolved.
National Cabinet agreed on 26 August that large data centres have material impacts on energy, water and land use, and that the Commonwealth will work with states and territories to develop consistent mandatory standards covering those areas, alongside skills and training requirements. The Federal government intends to legislate the standards in early 2027, with the federal framework designed to complement rather than duplicate state and territory planning and approval processes.
The agreement follows weeks of disagreement between Canberra and the Queensland and Northern Territory governments over the proposed requirement for new large data centres to underwrite additional renewable generation, backed by firming.
That requirement was not included in Wednesday’s National Cabinet statement, which instead refers more broadly to mandatory standards for data centre energy. Queensland had argued for a technology-neutral approach to energy supply, while the Commonwealth had previously said it would legislate to prevent states competing for data centre investment by offering less stringent energy requirements.
Energy compromise
The final agreement inevitably represents a compromise: Canberra has secured a commitment to national standards and retained its intention to legislate, while states and territories have avoided signing up to the specific renewable-energy model that had been at the centre of the dispute.
For data centre developers, the outcome should provide greater clarity over the direction of policy, although the detail of the standards, including how energy requirements will work in practice, remains to be developed.
The economic case has some big assumptions
The National Cabinet agreement comes as the data centre industry seeks to demonstrate the broader economic benefits of its expansion. Industry research commissioned by Data Centres Australia and prepared by Mandala Partners estimates that Australia’s data centre capacity could reach 3.9GW by 2030, generating AUD 5.6 billion in annual economic activity and supporting 23,040 ongoing jobs. It also identifies a potential additional 1.9GW to 2.9GW of AI export capacity, which it estimates could generate up to $4.1 billion in annual economic activity.
The AI export figures are scenario estimates rather than forecasts, based on Mandala’s assessment of Australia’s competitiveness against regional markets. The report itself cautions that the figures represent an opportunity rather than a guarantee, with the outcome dependent on energy, workforce, planning, investment and policy settings.
How much of the AI investment stays in Australia?
One question receiving less attention in the enthusiasm around data centre investment is how much of the headline capex translates into Australian economic value.
Tim Toohey, head of macro and strategy at Yarra Capital Management, argues that estimates based on headline data centre investment can overstate the domestic economic impact because a substantial proportion of the equipment is imported and much of the economic rent may accrue to offshore owners and intellectual-property holders.
Toohey estimates that servers and GPUs can account for around 70% of the cost of an AI data centre and argues these are effectively entirely imported. He consequently estimates that more than 80% of the cost of a typical AI data centre could be imported, although the analysis is based on a composite estimate rather than a single published Australian benchmark.
The argument delivers an important qualification to headline investment figures. Economic activity recorded in Australia does not necessarily indicate where ownership, profits and intellectual property ultimately reside. That is particularly relevant to a sector dominated by global hyperscalers and international capital. For policymakers, the question is therefore not simply how much investment Australia can attract, but how much value, tax revenue, intellectual property and long-term capability the country can retain.
Energy remains critical
The other major question is how the additional electricity, transmission, firming and water requirements associated with data centre growth will be accommodated. Mandala’s research argues that Australia could provide substantially lower-emissions AI compute than more carbon-intensive regional markets. It estimates that hosting 1.9GW of export compute in Australia would produce 3.2 million tonnes of CO2 in 2030, compared with 11.9 million tonnes if the same capacity were hosted in Indonesia. However, the comparison covers only location-based Scope 2 emissions from grid electricity and excludes embodied emissions, backup generation, refrigerants and additional firming requirements.
That leaves the detail of the national framework particularly important. The Commonwealth has secured agreement that data centres will be subject to consistent mandatory standards, but the most contentious element of its original proposal, that is, requiring large facilities to bring forward additional renewable generation backed by firming, has not made it into the National Cabinet statement.
The standards to be developed will therefore determine how effectively Australia can balance competing objectives: attracting investment and AI infrastructure, maintaining reliability and affordability in the electricity system, managing water and land-use impacts, and ensuring that data centres pay an appropriate share of the infrastructure required to support them.
The policy challenges remain around what conditions should be attached to data centre development and, with legislation planned for early 2027, governments now have an ever-closing window of opportunity to establish those conditions before the next wave of AI infrastructure is committed. Getting this part right will determine how much of the economic and strategic value ultimately remains in Australia. Getting it wrong will highlight how much of the associated cost is borne by the wider economy.