On 23 March 2026, the Australian Government placed sustainable water use at the heart of its data centre investment framework for the first time. The Expectations of Data Centres and AI Infrastructure Developers, jointly announced by three ministers, sets out five national expectations intended to drive how data centre developments are assessed and approved across Australia. Expectation 3 is dedicated entirely to water.
What Expectation 3 actually requires
The water-specific provisions are substantive. Data centres are expected to engage early and proactively with water utilities, communities and First Nations peoples on location and water sourcing. They must adopt efficient cooling technologies, use non-potable water where possible, apply circular water opportunities, cover their share of infrastructure and delivery costs, build resilience against drought and climate disruption, and report transparently on water usage and efficiency.
Read together, these provisions touch on almost every concern the water sector has raised about data centre growth. The critical qualification is how they operate. The Government has not legislated them. Proposals aligned with the Expectations will be prioritised in Commonwealth regulatory assessments; those that are not will face a slower process. This is an incentive, not a requirement — and that distinction matters.
The scale of what is coming
The numbers make clear why this policy moment matters. In NSW alone — home to 90 operating data centres and the largest concentration of data centre investment in Australia — the state's Investment Delivery Authority has just endorsed 15 data centre projects collectively worth $51.9 billion. Data centre construction investment in NSW has been growing at 65 per cent per year on average, and the sector now represents 12 per cent of all non-residential building investment in the state. A further $40.7 billion in proposals was rejected as premature or speculative — a striking reminder of how much phantom demand is circulating in the system, and how difficult that makes infrastructure planning.
At the water level, the pressure is already being felt. According to the Water Services Association of Australia's Data Centres and Water in Australia (December 2025), Australian utilities have been receiving applications from data centre proponents seeking between 5 and 40 million litres per day — up to 20 times the consumption of the largest existing single drinking water customer. Sydney Water estimates that data centres could represent 15 to 20 per cent of Sydney's total water supply by 2035. The NSW Data Centre Consultation Paper, released the same week as the federal Expectations, is candid about the planning challenge this creates: it acknowledges that Sydney Water's demand forecasts "differ drastically" from those in an industry-commissioned report, and that precise water demand data from data centres is currently not available during the planning phase at all. Without reliable data, infrastructure planning becomes guesswork.
The enforcement gap
The central problem is the gap between aspiration and obligation. The concern is not with developers who are already efficiency leaders. It is with the rest of the market — those who will meet the minimum necessary to secure approvals, then make different decisions once operational scrutiny moves on.
This is the retrofit problem. Cooling systems are architecturally integrated into data centre design. Shifting from evaporative to liquid or immersion cooling in an existing facility is not a matter of swapping equipment — it is an architectural shift that effectively cannot happen post-construction. Design decisions made at approval stage lock in water consumption for 20 to 30 years.
Both the Water Services Association of Australia and the NSW Government's own consultation paper point toward minimum Water Usage Effectiveness (WUE) standards as the logical next step. And notably, the state level is now moving in that direction. The NSW Consultation Paper explicitly proposes that data centres "could be designed to achieve Power Usage Effectiveness (PUE — the energy equivalent of WUE) and WUE thresholds consistent with international benchmarks," pointing to the EU and Singapore as reference models. It also proposes smart water meters at all data centre sites to improve monitoring, and a collaborative process to align water demand forecasting assumptions between industry and government.
This is meaningful: NSW is actively consulting on the very performance standards the federal framework stops short of mandating. If those consultations result in binding WUE thresholds built into State Significant Development approvals, NSW would effectively have done what the national Expectations left undone. Australia already requires data centres that store or process federal government data to achieve a five-star NABERS energy rating — demonstrating that mandatory efficiency standards for this sector are not a new concept. The logical next step is extending that model to water, and applying it to all new builds rather than just those handling government data. Environmental groups and industry alike have recognised that consistent, enforceable standards would benefit investors and developers by providing certainty and accelerating approvals.
A welcome signal, but the clock is ticking
Taken together, the federal Expectations and the NSW Consultation Paper represent the most significant regulatory attention Australian data centre water use has ever received. The federal framework sets the national tone; NSW is now working out what that means in the jurisdiction where most of the growth is actually happening.
But the window to influence outcomes is not at the point of operation or construction — it is now, at the policy and planning stage, before cooling systems are designed and sites are locked in. The NSW consultation is open until May 8. Every approval granted without a binding water efficiency condition is a decision that will shape urban water demand for a generation. The water sector has both the expertise and the standing to shape what comes next. The question is whether it moves quickly enough to do so.
