With so much global activity, it is both logical and encouraging to see increased recognition of the role and importance of data centres as essential digital infrastructure. Australia has much to gain from taking a leading position on data centre development, not least in enabling capacity for productivity, economic growth and the renewable transition.
Therefore, the recent article in InnovationAus.com by academics Jo-An Occhipinti, Ante Prodan, and Roy Green, made important points on Australia’s potential – as a country with renewable energy abundance – to be a key player in the development of sustainable digital infrastructure.
This is aligned with a collective industry view in a 2024 Mandala Partners paper that data centres, incentivised by the sustainability goals of their customers and mechanisms such as Power Purchase Agreements, are a catalyst to renewable energy projects.
Further, these large data centre operations support grid stability by being a steady and reliable demand off-taker.
The Mandala paper – commissioned by AirTrunk, together with Amazon Web Services, CDC, Microsoft and NEXT DC – highlighted how aggregated digital processing demand in efficient data centre facilities provides a 67 per cent energy saving compared to that requirement being housed on-premises by thousands of Australian business and government users of cloud services.
With significant corporate IT yet to transition out of traditional server rooms, there is still an impressive upside to come.
One reason we commissioned the paper was to provide accurate data and insights as a baseline for commentary and policy consideration.
The Mandala paper was the first analysis of its kind to take relevant input directly from data centre developers, resulting in the stunning forecast of an additional $26 billion investment anticipated in Australian developments by 2030.
While the Green et al academic paper and media commentary make valid points on the role of generative AI data centres in renewable energy integration, grid stability, and Australia’s unique opportunity, in the interest of informed debate it is important to correct a number of inaccuracies.
Firstly, the article states that data centres currently account for 6 per cent of Australia’s electricity use – citing figures in an oft-quoted Morgan Stanley report.
However, official AEMO measurements of actual 2024 metered data centre electricity consumption puts this figure at as little as 1 per cent of national use – well behind other industries like metal manufacturing, healthcare, accommodation and food service, retail and public administration and safety.
Secondly, the paper fails to recognise that not all data centres are the same and that for the most part, the market in Australia requires dual use facilities built for both cloud and future AI demand.
This means that the way we manage data centre development must be sensitive to the particular requirements of those use-cases.
For example, the demands of time-sensitive applications and mission-critical services – including things like healthcare, transport, and emergency services – mean that data centres are located in what the industry calls ‘availability zones’, on the urban fringe, close to customers and close to the highly skilled, and security-cleared workforce that runs them.
Specifically on AI, it is correct and appropriate to draw a distinction between AI training and AI inferencing. Again, the purposes and the requirements of these two use-cases are different with consequences for where you might build them.
For example, facilities solely dedicated to AI training could – potentially – be located outside of major metropolitan centres, but facilities providing AI inference and other functions have some of the same time-sensitive and mission-critical functions as cloud services, so will need to be located in those same ‘availability zones’, close to users.
Most industry analysts believe AI inferencing is where the majority of data centre future demand will come from in Australia.
With respect to grid stability, Green et al state that data centre operators can choose low energy demand periods to batch process transactions.
While attractive at first glance, this argument fails to recognise the key principles of net neutrality whereby a data centre operator who has no visibility of its customer’s data, nor the contractual ability to make a decision about the relative importance of a customer’s transaction.
The authors claim that data centre operators ‘often’ disconnect from the grid during voltage disturbances as an economic calculation.
It is important to understand that in order to protect the operation of the many mission critical services operated from a data centre, modern facilities are designed to switch automatically to generator back-up when they sense a drop or spike in voltage from the utility.
Of course, grid stability is a factor that needs to be weighed up against the criticality of digital services that are dependent on data centre reliability, and data centre operators are proactively working with energy utilities contributing to upgrading and replacing energy infrastructure that not only benefits their services but the communities in which they operate.
Australia is in a prime position to capitalise on the boom in data centre development, leveraging its proximity to growing markets, political stability, land availability, and cost-competitive renewable energy.
But to capitalise on that potential, it is essential that policymakers have the access to well-informed and nuanced inputs that recognise not only the essential nature of data centres but also their unique characteristics and requirements.
For AirTrunk and other industry players, this means proactively engaging in key discussions, helping to raise awareness and build understanding.
Contributions like the recent commentary in InnovationAus are welcome and would be strengthened through robust industry consultation to ensure a well-rounded, fact-based discourse.
Do you know more? Contact James Riley via Email.