-1.png?w=1024)
Australia wants more data centers. But there are issues that need sorting out first.
New federal screening rules on data centers will mean future projects are judged on factors that affect household power prices, enterprise cloud bills, and how quickly Australia can keep pace in the global race for AI compute.
Australia's data center sector has been on a steep climb for years, but the last few months have made that growth impossible to ignore. The country’s data center occupancy has skyrocketed since 2005, with two-thirds of that expansion happening since 2020, according to a report from M3 Property.
The bulk of this exponential growth is concentrated in New South Wales (NSW) and Victoria, where more than 250 facilities now operate. Already, that capacity is on track to double again between 2026 and 2030, with AI workloads as the primary driver. However, these data centers draw far more power than the cloud services that came before them, raising an important question that affects everyone, from end users and businesses to governments and the companies behind these data centers.
The approval bottleneck
While the country saw an exponential rise in data centers, that growth is no longer happening on autopilot. On March 23, the Australian government released its Expectations of Data Centers and AI Infrastructure Developers.
The policy ties faster regulatory treatment to how well a data center project aligns with national priorities. These priorities include energy, water, local jobs, sovereign data goals, and domestic research capability. Although it doesn't change existing law, the policy changes the queue: projects that can't demonstrate alignment should expect slower pathways and more execution risk.
For enterprises planning AI rollouts, that's not an abstract policy story; it's a supply question. If new Australian capacity takes longer to clear approvals than AI demand is growing, the gap between what businesses want to run locally and what's actually available to lease could widen. That matters for anything latency-sensitive, anything tied to data residency requirements, or anything that depends on in-region GPU access rather than routing workloads abroad.
Who foots the bill for the build-out?
Australia currently ranks among the top 10 countries by data center size, making it the largest in the APAC region. And while that is a positive sign, there is a downside: it is becoming harder to separate it from issues with approvals.
As hyperscalers and cloud operators build out Australian sites to meet AI demand, the costs of securing power, building cooling infrastructure, and upgrading grid connections don't disappear — they just become less visible.
The new federal framework notes that new data centers shouldn't place upward pressure on energy prices and should contribute positively to the energy transition, which signals that regulators expect this cost-shifting dynamic to remain a live issue as the build-out continues.
Where this gets harder to track is in how AI is actually used day-to-day. Most of the public conversation about AI's resource footprint has centered on AI training, which is a high but one-off cost.
Less visible is inference: the computation drawn every single time someone runs a prompt through a chatbot, a coding assistant, a support tool, or an embedded enterprise feature. As businesses increasingly use AI, it has become a continuous draw on resources rather than a one-time line item.
That's the crux of the disclosure gap facing Australian buyers right now: AI usage costs accumulate continuously through inference, but there's no standardized way for providers to report the ongoing environmental footprint tied to that usage.
Put simply, this means that while AI data centers place huge stress on resources, which often flows down to businesses, these businesses now rely on these very data centers for their day-to-day activities, effectively increasing resource demand and costs.
What “renewable” claims sometimes omit
With government policies on AI making moves to protect societal resources like water and power, these data center companies now appear to be optimizing resource consumption.
A Google-authored paper reported a 33-fold reduction in energy use and a 44-fold reduction in carbon footprint for the median Gemini Apps text prompt between May 2024 and May 2025.
Outside researchers have cautioned that per-prompt figures can understate the full picture. These estimates may exclude indirect water use and location-specific grid emissions — an important caveat for Australian businesses running AI workloads on hyperscaler infrastructure. In those cases, the local energy mix and water context can differ markedly from the provider’s reported averages.
It also complicates the question of how much weight businesses should place on a cloud provider's sustainability credentials. Australia's own national framework addresses this gap by treating energy use and water management as separate considerations from emissions, effectively acknowledging that an electricity-use claim doesn't tell the whole story of a facility's local footprint.
What enterprises should watch
None of this means Australian businesses need to pull back on AI adoption. But it does mean the planning conversation needs to widen. IT and procurement teams tracking AI rollouts should track how the new approval process affects capacity timelines.
It's also worth pushing cloud and AI vendors for facility-level detail behind sustainability claims rather than relying on procurement-level renewable percentages. Also, inference costs should be factored into AI budgeting rather than treating model access as a fixed, one-time expense.
The infrastructure underneath AI is getting more expensive and more regulated at the same time — and right now, much of that is happening with less visibility than the price tags alone would suggest.