Everything About AI
NextFin News - Australia's artificial intelligence data-centre boom is running into a wall of grid limits, water concerns, and community opposition - and the federal government's new national standards will not apply to dozens of already-approved projects, leaving the hungriest next-generation facilities outside the rules they were designed to enforce.
Prime Minister Anthony Albanese announced on July 15 that Australia would introduce mandatory "Australian Standards for AI," including a legal obligation for large data centres to underwrite their own new power supply, pay their full share of grid-connection costs so household bills are not affected, reduce power when the grid is stressed, and be as water-efficient as possible. Legislation is expected early next year. But an analysis of planning records and company documents published in mid-September showed that at least 27 approved-but-unbuilt data centres - with a combined capacity of at least 3.38 gigawatts - will escape the coming restrictions because the rules are not expected to be retrospective.
The tension is stark. The Australian Energy Market Operator now expects data-centre energy demand in the National Electricity Market to rise seven-fold within a decade, from around 3% of demand today to 13% by 2036 - a level the grid operator had not forecast until 2050 just a year earlier. Yet Canberra is writing the rulebook after the land rush has already begun. The question is not whether Australia wants AI investment - it does - but who pays when the bill for that investment arrives at the grid, the water system, and the household meter.
The Rules Arrive After the Land Rush
The sequence of events explains why the standards feel both ambitious and late. The federal government first issued non-binding expectations for data-centre and AI infrastructure developers on March 23, 2026 - five expectations covering energy, water, location, workforce, and innovation, announced jointly by the Industry, Energy, and Digital Economy ministers. Those expectations carried no legal force. Community and council pressure mounted through the first quarter: three Sydney councils - Ryde, the City of Sydney, and Penrith - used the New South Wales Legislative Council inquiry in March to formally call for a moratorium on data-centre approvals, with Penrith asking for a pause until energy and water impacts were properly understood.
Canberra responded on July 15 with the harder-edged framework. In a speech at the University of Sydney, Albanese framed the moment as a choice:
This world-leading framework is about Australia choosing to shape the future rather than letting the future of AI shape us. If we set our national standards high, then we can make AI stand for Australia's interests.
The Office of AI was established inside the Department of the Prime Minister and Cabinet effective immediately, with the standards to go to National Cabinet in August and legislation to parliament early next year. The government's pitch is certainty: a single national regime to replace what Albanese called a fragmented approach.
But certainty cuts both ways. Developers who secured approvals under the old, softer regime now hold permits that the new rules cannot easily reach. The mid-September analysis - verified by analysts at renewable-project mapper RenewMap and construction-analytics firm CBRE - found that the approved-but-unbuilt cohort is not marginal. Its 3.38 GW of capacity is more than double Australia's entire existing data-centre fleet, which stood at 1.5 GW of combined IT capacity across 162 facilities as of March 2026, according to the industry body Data Centres Australia. In other words, the projects that will never face the new standards are already larger than everything the country currently operates.
One of those projects is property developer Goodman's five-storey Project Apollo in Sydney's north, approved the week before the analysis was published. It will take two years to build and would be one of Australia's largest data centres by computing capacity if it existed today - and it is not subject to the incoming federal rules, though its NSW approval carried its own water and energy conditions. RenewMap co-founder Alex Thompson offered the industry's own hedge on the pipeline: even though projects have been approved, "not all of these data centres will get built."
The retrospective gap is the structural flaw at the heart of the policy. A rule that governs only future entrants while the largest incumbents-in-waiting operate under yesterday's terms does not solve the problem it was announced to fix. It creates a two-tier system: a grandfathered cohort with no obligation to underwrite new generation, and a post-2027 cohort that must. That asymmetry is not a transition detail - it is the price signal the market will actually respond to.
The Grid Math: Seven-Fold Demand in a Decade
The scale of the demand shock is what turns a planning issue into a political one. The grid operator's 2026 forecast, in its Step Change scenario, has data centres consuming one-eighth of current National Electricity Market electricity use within a decade. The load is heavily concentrated: more than 85% of forecast data-centre consumption through 2035-36 sits in New South Wales and Victoria, with about 66% of today's load already in NSW alone. RenewMap's May 2026 snapshot showed 19 sites under construction totalling 1,853 MW, with the ten largest accounting for 90% of that pipeline - clustered in western and northern Sydney and Melbourne's west.
The speed of the acceleration is the real story. Data-centre electricity demand grew 18% in NSW and nearly doubled in Victoria over the twelve months to mid-2026. A year earlier, the grid operator had not expected to see this level of demand until 2050; it now expects it by 2036. When a forecast horizon compresses by fifteen years in twelve months, the infrastructure that was going to be built gradually must now be built immediately - and transmission does not move at AI speed.
How that new demand is powered determines who wins and who pays. The Climate Council's June 2026 report, "Clouded Future: Managing the Risks of the Data Centre Boom," modelled the two paths. If new data-centre demand in NSW and Victoria is met mainly with gas generation rather than renewables, wholesale electricity prices could rise 26% in NSW and 23% in Victoria by 2035, with national grid emissions running roughly 14% higher than under a renewables-led path. The mechanism is simple and unforgiving: gas-fired marginal generation sets the wholesale price, and data centres - unlike households - run flat out around the clock, including through the midday solar peak when demand would otherwise be falling. The grid operator has already flagged this in its minimum-demand outlook: data centres are one reason the solar-era demand trough is filling back in, which is good for system stability but means their load is always there, always bidding.
This is where the government's "underwrite your own new power supply" obligation matters most - and where its limits show. Requiring a data centre to contract for new renewable capacity is not the same as requiring that capacity to be built in the right place, at the right time, with the transmission to deliver it. A power-purchase agreement signed in 2027 does nothing for the 3.38 GW of grandfathered capacity drawing on the grid today. And a data centre that underwrites its own supply still needs a physical connection - one that competes with households, hospitals, and electrifying industry for the same constrained network.
There is also a water dimension that the standards address but do not yet quantify. Industry estimates cited by the Climate Council suggest data-centre water demand could triple by 2030, at precisely the moment climate change is intensifying drought and water stress across the eastern states. Cooling is not optional: a facility that cannot secure water cannot run. The standards ask operators to use efficient cooling technologies and, where possible, non-potable water - but "where possible" is the kind of phrase that becomes a lobbying battleground.
Social Licence: Polls, Councils, and the Google Pause
The political economy of the boom is shifting faster than the physical economy. A YouGov survey of 1,624 Australians commissioned by the Climate Council found that 82% agree governments should insist new data centres pay for additional renewable energy and storage matching their electricity consumption; 67% agree data centres risk driving up power bills and straining the grid; 69% support strict mandatory energy and water standards. Only 49% are convinced the data-centre industry will boost Australia's economy. That last number is the one that should worry policymakers: when barely half the public believes the investment story, the social licence is thin.
Local opposition is no longer abstract. Western Sydney resident Jack Tollefsen lives in Horsley Park and works down the road from NEXTDC's S4 development, under construction and expected to provide 365 MW of capacity - a single facility roughly equal to a quarter of the country's entire existing fleet. In Victoria, a community campaign network has formed to oppose data-centre projects in Melbourne's west, arguing that approvals are being fast-tracked without public consultation. The mayors of Australia's capital cities have collectively called on state and federal governments to ensure data-centre growth is backed by sustainable infrastructure aligned with community expectations.
The Greens have pushed further. Senator Sarah Hanson-Young has led calls for a ban on data-centre construction until minimum standards are established, warning that the government risked losing community support for AI if it did not take concerns about energy, water and local benefits seriously. Her point is political arithmetic: the benefits of a data centre - a handful of permanent jobs, a line item of investment - are diffuse and corporate, while the costs - higher bills, water use, construction traffic, visual impact - are local and immediate.
Investors are already pricing regulatory risk. In March 2026, Google told the federal government it was withholding a roughly A$20 billion AI and data-centre hub investment because of the risk that the Australian Taxation Office could deem it to have a "permanent establishment" in Australia, exposing its broader operations to the 30% corporate tax rate. The company was assessing where to build its Asian hub, with Australia among the leading candidates. The tax issue is distinct from the environmental standards, but the pattern is the same: Australia is discovering that attracting hyperscale capital requires more than land and sun - it requires a stable, predictable rulebook, and every new rule announced mid-stream reminds investors that the rules can change.
The industry's response has been to argue that movement is already happening. Belinda Dennett, chief of Data Centres Australia, said her members were already taking steps to meet the government's expectations around power and water use - but drew a line at rules that do not yet exist:
Members will continue to meet those requirements until the new rules are in force - but they cannot comply with something that isn't yet established or even articulated.
That is a fair description of the transition problem, and it is why the government is promising speed. But meeting expectations voluntarily is not the same as being bound by law - and the gap between the two is exactly what the grandfathered cohort will exploit.
Meanwhile, other hyperscalers are still committing. Microsoft has pledged A$25 billion to expand AI and cloud capacity in Australia, building on a A$5 billion commitment in 2023, and has affirmed its commitment to the national expectations. Amazon announced A$20 billion of investment from 2025 to 2029. Anthropic, developer of the Claude large language model, signed an agreement with the Australian government to align with the national expectations and will open a Sydney office. The capital is real. The question is whether the infrastructure and the social licence can keep pace with it.
What This Means: A Two-Tier Market and a Test for Canberra
The central judgment is that Australia's data-centre skepticism is structural, not cyclical. This is not a sentiment wave that will pass when the next investment announcement lands. Three forces have shifted permanently. First, the grid constraint is physical: transmission builds in years, data centres in months, and that mismatch does not self-correct. Second, the cost-bearing question has been politicised: households facing rising bills will not accept subsidising AI infrastructure quietly, and the 82% polling number shows the public has already formed a view on who should pay. Third, the regulatory regime has moved from voluntary expectations to mandatory standards - a one-way door. Once a government has legislated that data centres must underwrite their own power, it cannot credibly go back.
The cyclical element is the investment cycle itself. Capital is mobile, and if Australia's rules become too onerous relative to alternatives, hyperscalers can and will build elsewhere - as Google's pause already demonstrated. But the binding constraint is not capital; it is the grid and the community. Capital will go where the power is. Australia has the land and the sun, but not yet the transmission or the social consensus.
The strongest counter-thesis is that the standards are a feature, not a flaw - that regulatory certainty is exactly what quality capital wants, and that grandfathering is a pragmatic necessity. Making rules retrospective invites compensation claims and litigation that could delay projects for years; a clean transition protects the investment already approved while raising the bar for what comes next. Industry representatives and some economists argue that the market will do the disciplining anyway: as Thompson noted, not all approved projects will be built, and those that cannot secure power or water will simply not proceed. On this view, the 3.38 GW of approved capacity is an upper bound, not a committed outcome, and the standards successfully ring-fence the future.
That argument is coherent - but it rests on a bet that the grandfathered cohort will be smaller than the analysis suggests, and that voluntary compliance will hold where the law does not reach. If the grandfathered projects do get built, and if they draw on gas or unconstrained grid power, the 26% wholesale-price impact modelled by the Climate Council moves from scenario to outcome. The counter-thesis wins only if market discipline is stronger than regulatory arbitrage. History suggests arbitrage usually wins.
The forward look has three horizons. In the short term - through the National Cabinet process and into the 2027 legislative process - watch whether the government attempts any retrospective element or sticks with a pure forward-looking regime. In the medium term, the falsifying signal is the grid operator's next long-term plan: if data-centre demand is tracking toward the low end of forecasts - below roughly 7% of National Electricity Market demand by 2036 rather than the 13% in the Step Change scenario - the grid-constraint thesis weakens materially and the boom proves more self-limiting than feared. In the long term, the structural question is whether Australia can build the transmission and renewable capacity fast enough to turn data centres from grid stressors into grid assets - the International Energy Agency estimates 20-25 GW of battery storage could be installed in data centres globally by 2030, potentially making them flexible load if the incentives are right.
Three scenarios frame the outcome. The base case: the standards pass in early 2027 as announced, the grandfathered cohort proceeds largely as approved, and Australia captures a meaningful but smaller share of hyperscale investment than the headline pledges implied - with the cost borne partly by the grid and partly by consumers. The upside case: the standards attract quality capital precisely because they are credible, the grandfathered projects that cannot secure clean power are cancelled by the market, and data centres become flexible, battery-backed load that stabilises the grid. The downside case: the two-tier system produces exactly the regulatory arbitrage critics fear, gas fills the gap, wholesale prices rise toward the modelled 26% in NSW, and community opposition hardens into local vetoes that stall even well-run projects.
The kicker: Australia's data-centre rules will be judged not by what they demand of tomorrow's projects, but by what they permit of yesterday's approvals - and on that test, the country is building its AI future on a grandfather clause.
Explore more exclusive insights at nextfin.ai.
