Tens of billions of dollars are being poured into Australian data centres to power the AI boom. That building spree is now colliding with the electricity grid, and “AI inflation” has become a real search term and a real worry. Here is what the numbers actually say, separating the signal from the scare.
Why this is suddenly a kitchen-table topic
For most of their history, data centres were invisible infrastructure that nobody outside IT thought about. That changed the moment their electricity demand got big enough to move the grid, and therefore big enough to move power prices. In a country already living through a cost-of-living squeeze, “the AI industry is going to put up my power bill” is a story that lands hard.
The search interest is not hype. There is a genuine, measurable collision happening between the AI buildout and Australia’s energy system. But the honest picture is more conditional, and more interesting, than either the boosters or the doomers will tell you.
The scale of the boom
Start with the money, because the numbers are genuinely enormous.
- AWS committed A$20 billion (2025-2029) to expand its Australian data-centre infrastructure, alongside 11 renewable energy projects across NSW, Queensland and Victoria. It is one of the largest tech investments in Australian history. (Blackstone/industry reporting)
- Blackstone acquired AirTrunk for A$24 billion in September 2024, its largest-ever Asia-Pacific deal, inheriting more than 800MW of committed capacity and land for over 1GW of growth. (Blackstone)
- AirTrunk earmarked around A$3.4 billion for a single 354MW Melbourne facility (MEL2), announced December 2025, taking its total Melbourne investment past A$7 billion. (Forbes)
- NEXTDC signed an MOU with OpenAI for a hyperscale AI campus at Eastern Creek in Sydney.
- More than A$100 billion in Australian data-centre plans were announced between 2023 and 2025, and Australia is now the world’s second-largest data-centre investment market after the United States. (Climate Council)
This is not a rounding error in the economy. The Australian Bureau of Statistics expects information, media and telecom capex of around A$19.2 billion in FY2025-26, up more than 50% year on year, driven substantially by data centres. (ABS)
The collision with the grid
Now the electricity. This is where the “AI inflation” worry actually has teeth.
For the first time, AEMO (the market operator) has broken out data centres as a distinct category in its forecasts, which tells you they are now big enough to model on their own. The figures:
| Period | Data centre demand | Share of the NEM |
|---|---|---|
| FY2025 | ~3.9-4 TWh | ~2.2% |
| FY2030 (Step Change) | ~12 TWh | ~6% |
| FY2050 | ~34 TWh | ~12% |
That is roughly 25% annual growth to 2030. (AEMO) And the forecast may be conservative: the requested connection pipeline in NSW alone is around 44 data centres totalling 11.4 GW, which the Climate Council notes is equivalent to nearly four Eraring coal stations, a “step-change beyond” current forecasts. Data-centre electricity use nearly doubled in Victoria and rose around 18% in NSW over the prior year.
The reason this can move prices is simple market mechanics: electricity prices are set at the margin. Add large, steady new demand faster than you add cheap new supply, and the marginal price rises, especially when the marginal generator is gas.
The number everyone is quoting: 26%
The headline figure driving the “AI inflation” fear comes from the Climate Council (June 2026): data-centre demand could raise wholesale power prices by up to 26% in NSW and 23% in Victoria by 2035. (Climate Council; The New Daily)
Two things to hold onto about that number, because they are usually dropped:
- Wholesale is about 40% of a typical residential bill, not all of it. So a 26% wholesale rise is not a 26% bill rise. The retail impact is materially smaller.
- It is a scenario, not a forecast. The 26% figure is specifically the case where the new demand is met with gas rather than renewables. Meet it with renewables plus firming, and the picture changes substantially.
That second point is the whole policy fight in one sentence.
The policy response: the net-generator law
Australia did not leave this to the market. On 15 July 2026, the Prime Minister announced a binding law requiring large new data centres to be “net-generators.” (pv-magazine; RenewEconomy)
In plain terms, a large new data centre will be required to:
- Pay its full share of grid-connection costs, with no cost passed on to households or other businesses.
- Put at least as much energy into the grid as it takes out, through new renewable generation plus firming.
This is a genuinely smart bit of policy design, because it targets the exact mechanism that would otherwise cause “AI inflation.” If operators must bring their own new clean supply and pay for their own connection, then their demand stops being a cost imposed on everyone else and becomes largely self-funded. Whether it works in practice depends entirely on enforcement and on how fast that new supply actually gets built, but the design is aimed at the right target.
The honest counterpoints
Because I promised to separate signal from scare, here are the points the alarmist version leaves out.
- Data centres are efficient. Consolidated hyperscale facilities are estimated to be around seven times more energy-efficient than the on-premise servers they replace. Some of this demand is not new, it is migrated and made more efficient. (Mandala/DCA)
- They fund grid infrastructure too. Analysis for the industry estimates data centres invested around A$3.1 billion in grid infrastructure since 2020, with more forecast, some of which benefits surrounding communities.
- Water is a smaller story than the headlines. Data centres use around 5.5 GL/year today, about 0.04% of national water use, projected toward 17 GL by 2030. Real, worth monitoring, but not a crisis on current numbers. (The Conversation)
- The RBA has not called this inflation. In the sources I reviewed, the Reserve Bank has not directly named data centres as an inflation driver. The clean transmission mechanism is wholesale electricity into bills and input costs, not a broad CPI effect. Calling the whole thing “AI inflation” is a compelling headline that runs slightly ahead of the evidence.
- There is a real data gap. Australia lacks mandatory data-centre energy and water reporting, so many of these figures are estimates. We are partly arguing about “phantom demand” that nobody is required to measure.
What it means for Australian leaders
If you run a business, this is not abstract. Three practical implications:
- Budget for higher energy input costs, but don’t panic-model a doubling. The credible range is meaningful upward pressure on wholesale prices this decade, partly mitigated by policy, flowing through as a real but bounded increase in bills and input costs.
- Energy strategy is now AI strategy. If your AI ambitions involve heavy compute, where and how that compute is powered is becoming a cost and ESG question, not just a technical one. This is part of why I argue you should think in terms of an AI factory rather than a pile of AI projects: the infrastructure economics matter.
- Efficiency is a competitive lever again. The same discipline that makes an AI deployment deliver real ROI also makes it cheaper to run in a higher-energy-cost world. Wasteful AI was always expensive. Now it is expensive twice.
The data-centre boom is real, the pressure on power prices is real, and the policy response is genuinely trying to keep the cost off your bill. “AI inflation” as a tidy slogan overstates it. “The AI buildout is putting measurable upward pressure on Australian energy costs, and who pays is now a live political and engineering question” is the accurate, less catchy version. That is the one worth planning around.
Amjid Ali is a technology and AI leader based in Melbourne. If you are scaling AI workloads and want the infrastructure and cost economics thought through properly, get in touch.