How Much Electricity Do Australia's Data Centres Use?

The short answer

Australia's data centres used about 5 terawatt-hours (TWh) of electricity from the National Electricity Market (NEM) in 2025–26. That is roughly 3% of all grid consumption, spread across around 165 facilities. The Australian Energy Market Operator (AEMO) forecasts this will rise about sevenfold, to around 34 TWh a year by 2035–36, or about 13% of the grid.

Those are the official figures. But many of the numbers in the news about data centres measure something different, and mixing them up leads to claims that are either alarming or dismissive. This guide explains what each figure means and where it comes from.

How much electricity do Australian data centres use today?

According to AEMO's 2026 Electricity Statement of Opportunities (ESOO), data centres connected to the NEM used about 5 TWh in the 2025–26 financial year. The NEM covers Queensland, New South Wales, the ACT, Victoria, South Australia and Tasmania. It does not include Western Australia or the Northern Territory.

To put 5 TWh in perspective: spread evenly across a year, it equals about 570 megawatts (MW) of continuous demand.

The use is heavily concentrated. In 2025–26, New South Wales accounted for about 66% of data centre consumption and Victoria about 30%. South Australia, Queensland and Tasmania together made up the remaining 4% or so.

How much will data centres use by 2035?

AEMO's central forecast, the Step Change scenario, has data centre use reaching about 34 TWh a year by 2035–36. That is roughly seven times today's level. On the way, AEMO expects data centres to reach about 7.8% of NEM consumption by 2029–30 and about 13% by 2035–36, after which growth is expected to slow.

If 34 TWh were drawn evenly across the year, it would equal about 3,900 MW of continuous demand.

How has the forecast changed?

The forecast rose sharply in a single year. In its 2025 ESOO, AEMO forecast 21.4 TWh by 2034–35. In 2026 it forecast about 34 TWh for 2035–36, one year further out.

AEMO's 2026 forecast draws on 225 projects in development, with a combined capacity of about 67 gigawatts (GW). It forecasts data centres separately from other large loads, combining project information from developers and network businesses with an economic outlook for digital transformation and AI adoption.

AEMO describes data centre growth as one of the most significant sources of both forecast growth and uncertainty.

Where are Australia's data centres being built?

Most of the growth is expected in New South Wales and Victoria.

Grid Desk's facility registry, compiled from RenewMap's public project pages, lists 22 major facilities in those two states, with a combined announced capacity of about 7,671 MW. Of that, about 1,196 MW is listed as operating, 2,675 MW as under construction, and 3,800 MW as proposed across just two very large sites.

Does 67 GW of planned data centres mean 67 GW of demand?

No. Announced capacity is a ceiling, not a meter reading.

Compare the numbers. The pipeline adds up to about 67 GW, yet AEMO's 2035–36 forecast works out to about 3.9 GW of average demand, under 6% of the pipeline. The gap has three causes:

So when you read that a region has gigawatts of data centres "coming", ask how many are built, how full they are, and what share of their connection they use.

What are the three kinds of data centre number?

Almost every data centre energy figure falls into one of three kinds. Each is useful, but each answers a different question.

KindWhat it measuresExampleWhat it can't tell you
PublishedOfficial forecasts and statistics from bodies like AEMO5 TWh today; 34 TWh by 2035–36What any single facility uses
AnnouncedCapacity operators say they will build67 GW pipeline; 7,671 MW in NSW and VictoriaHow much will be built, used, or when
EstimatedInferences from grid data or modellingA rising "flat baseline" in regional demandWhich loads are data centres and which are not

When a figure is quoted without its kind, treat it with caution. Adding announced megawatts together and calling the total "demand" is the classic error.

Why can't anyone measure data centre electricity use directly?

Australia's public market data does not separate data centres from other customers. AEMO publishes demand for each region every five minutes, but a data centre's load is mixed in with homes, offices and industry. There is no general requirement for operators to publish their consumption. Rules proposed in 2026 would require large data centres to register as market participants, which the Australian Energy Market Commission says would give AEMO better visibility of large loads.

There is one useful clue. Unlike households, a data centre draws almost the same power day and night, every day of the week. If you take regional demand and remove the parts explained by time of day, weekends, temperature and rooftop solar, what remains includes a flat layer. That layer should grow as data centres connect.

I built an open-source tool, Grid Desk, that tracks this "flat baseline" from live AEMO data. The method has a clear limit: any other load that runs around the clock moves the baseline too. So it gives an upper bound on data centre growth, never a measurement of it.

How does Australia compare with the rest of the world?

Australia is following a global trend, from a smaller base. The International Energy Agency (IEA) projects that worldwide data centre electricity demand will more than double by 2030, to about 945 TWh a year. That is roughly as much electricity as Japan uses today. The IEA expects demand from AI-optimised data centres to more than quadruple over the same period.

In the United States, the IEA expects data centres to account for almost half of all electricity demand growth to 2030.

What would make data centre energy numbers more reliable?

Better information would help planners, investors, communities and the industry itself. Three practical steps would make the biggest difference:

  1. Report data centre consumption as its own category. If AEMO or the network businesses published total data centre use by region, even quarterly, the public debate could move from estimates to measurement.
  2. Have operators disclose actual use alongside capacity. Annual energy use per site would show how full facilities really run.
  3. Label every number. Journalists, analysts and companies can say whether a figure is published, announced or estimated.

Frequently asked questions

What percentage of Australia's electricity do data centres use?

About 3% of NEM grid consumption in 2025–26. AEMO forecasts that will rise to about 7.8% by 2029–30 and 13% by 2035–36.

How many data centres are there in Australia?

AEMO's 2026 forecast counts around 165 data centres connected to the NEM. Another 225 projects are in development.

Which state uses the most data centre electricity?

New South Wales, with about two-thirds of NEM data centre consumption in 2025–26. Victoria accounts for most of the rest.

Is AI the main reason data centre power use is growing?

AI is part of it. AEMO's forecast combines known projects with an outlook for both AI adoption and wider digital transformation, but it does not attribute a specific share of the growth to AI.

Will data centres cause blackouts in Australia?

AEMO's 2026 assessment tested demand growth arriving up to four years earlier than forecast and found grid reliability could still be maintained.

What is the difference between a data centre's capacity and its use?

Capacity is the most power a site is designed or connected to draw. Use is what it actually consumes. In 2025–26, Australian data centres used on average about 27% of their grid connection.

What is a terawatt-hour?

A terawatt-hour (TWh) is a billion kilowatt-hours. One TWh a year is roughly equal to 114 MW of power drawn continuously, all year.

Sources

Conversions from TWh a year to average MW are the author's own arithmetic (1 TWh ÷ 8,760 hours ≈ 114 MW).

About this article

Archie Maclean is a writer and technologist based in Melbourne. He is completing a Master of Space Operations at UNSW and builds open-source tools for reading public energy and satellite data.

Researched and drafted with Claude AI from my brief, building on the data and methods from my GitHub projects, including Grid Desk's data centre registry and flat-baseline analysis. I chose the topic and angle, directed a fact-check against the primary sources listed, and had every claim I couldn't source removed.