The Office of AI and the shape of Australia's data centre future

Australia is about to receive billions in AI infrastructure investment, and where it lands will be decided in the next twelve months, not the next decade. Anthony Albanese's new Office of AI, folded into his own department, gives government one body to plan land, energy and compute together for the first time. Get the sequencing wrong and we lock in years of strained grids and congested cities. Get it right and AI investment can build regional Australia instead of overloading the metros that are already stretched.

The role of neoclouds

The capital chasing artificial intelligence is landing in Australia thick and fast, and a growing share of it is coming from neocloud providers: specialist operators supplying GPU-intensive computing capacity for AI workloads. Our leverage, as the Prime Minister put it, is physical: "The expansion of AI requires a physical, material footprint. It needs our land and energy and computing power to operate. That means we can set the terms." Not all AI infrastructure carries the same requirements. Latency-sensitive inference and many conventional cloud workloads favour locations with strong connectivity to major population and network centres. Large-scale AI training can offer greater geographic flexibility, allowing location decisions to be driven more heavily by access to firm power, transmission capacity, fibre, land and a credible pathway to approvals and delivery. Planned properly by the new Office of AI, that flexibility creates an opportunity to broaden where major AI infrastructure investment can occur without simply adding further pressure to already constrained metropolitan infrastructure.

A neocloud, in essence, is a cloud service purpose-built for AI: customers access large pools of graphics processors and other accelerated computing infrastructure rather than owning the hardware themselves. At scale, these workloads rely on vast clusters of processors at far higher rack densities than traditional enterprise computing, with corresponding demands for power and cooling. Large training workloads are generally less sensitive to end-user latency than real-time inference. That creates greater freedom to rethink where some of this infrastructure is built, but only where the required power, fibre, resilience and supporting infrastructure can be delivered.

Precincts, not projects

This combination changes the outlook. At sufficient scale, an AI campus can no longer be considered simply as a building connected to whatever infrastructure happens to be available. Power generation and firming, transmission and grid connection, water and cooling, fibre, roads and the construction workforce can all become critical elements of the development strategy. Either way, these systems need to be planned together. Mr Albanese's proposed standards make the energy relationship particularly important, with the Government signalling an expectation that large AI data centres contribute at least as much energy to the grid as they consume. This is why the largest developments should increasingly be planned as integrated infrastructure precincts rather than isolated building projects, with compute, energy, water and digital infrastructure coordinated through one delivery strategy and schedule.

The greater location flexibility of large training workloads creates an opportunity for regional Australia, but cheap land and abundant renewable resources are not enough on their own. For AI infrastructure, one of the critical location metrics is time to compute: how quickly a site can secure firm power, network capacity, diverse fibre, approvals and the infrastructure needed to energise GPUs and begin generating revenue. A cheap remote site has little value if transmission augmentation or grid connection pushes energisation back by years. Water also remains an important consideration. New closed-loop liquid and dry-cooling technologies can materially reduce potable water demand, although the outcome depends on cooling architecture and local climate and may involve trade-offs in energy use. The strongest regional locations will therefore be those where land, firm power, fibre, water strategy and approvals can be assembled into the fastest and most resilient delivery pathway.

The importance of the right sequencing

The Office has promised "greater clarity and speed for approvals"; the bigger opportunity is to identify locations where land-use compatibility, power, fibre, water and planning pathways are aligned before individual projects arrive. That does not mean pushing every AI campus into remote Australia. It means giving investors a credible map of where large-scale compute can be delivered fastest, with the least infrastructure conflict and the greatest long-term benefit.

The Office should help plan a region at a time, identifying where compute, firm power, transmission, water, fibre and land-use planning can be coordinated to create investable development pathways. Success looks like shorter timelines between investment decision and productive compute, and infrastructure that serves the region as much as the data centre. Ports, railways and pipelines were built around the industries that defined earlier eras. AI deserves the same ambition. As the Prime Minister put it: "We cannot revisit this issue after companies have built whatever they want, wherever they want." The Office of AI's opportunity is to get the planning done before the capital arrives, not after.