New data centre forecasts for the NEM

Hey Reader, welcome to The Energy's weekly data newsletter. This week we look at AEMO's new data centre demand forecasts and the thinking behind the wide range of scenarios.

But in breaking news, the NEM review draft report just dropped. You can find it here and get all the detail from the panel, including Tim Nelson, at our webinar today.

Scenario planning

The Australian Energy Market Operator has given itself a 40% margin of error in new forecasts on data centre energy demand.

Splitting data centres out from other large industrial loads in its latest Inputs, Assumptions and Scenarios Report, AEMO’s Step Change scenario predicts data centre energy consumption will increase by an average of 25% per year, reaching around 6% of the NEM by 2029-30, double what it had previously been forecasting.

Under this scenario, data centres would use 12% of grid supplied electricity by 2049-50, but under an alternate accelerated data centre growth sensitivity model, it could be 40% higher beyond 2030.

NEM data centre electricity consumption forecast, all scenarios.
Source: AEMO 2025 Inputs, Assumptions and Scenarios Report

The new forecasts were produced with advice from consultancy Oxford Economics Australia and will be applied in both the upcoming Electricity Statement of Opportunities and Integrated System Plan.

Oxford Economics considered a wide range of inputs in its modeling, including project information and macroeconomic drivers.

At the heart of the big gap in the forecasts is uncertainty over the level of AI uptake in Australia, and to what extent Australia can attract a segment of the highly concentrated hyperscaler market.

“Forecasting electricity demand from data centres over a multi-decade horizon involves a high degree of uncertainty. On the upside, rapid AI uptake could lead to faster than expected demand realisation, compressing ramp-up periods and bringing forward peak loads. If a significant share of anticipated projects proceed, demand could significantly exceed current estimates. On the downside, project deferrals, efficiency gains, or slower digital adoption could lead to flatter demand, particularly beyond 2030.”

Oxford Economics Australia Data Centre Energy Demand 2025

The AI effect

In his address to the National Press Club last week, Atlassian co-founder Scott Farquar made the case for Australia to be getting paid to “export megawatts as megabytes”.

He said Australia was “surprisingly cost competitive building data centres” with its talent and attractively priced green power. But he said copyright laws were a barrier to many AI companies training or hosting their models in Australia, along with the shortage of people needed to physically construct data centres and big batteries.

Oxford Economics has included a “rapid AI uptake” sensitivity in its forecast which could make the next five years particularly interesting.

Data centre consumption under rapid AI uptake sensitivity.
Source: Oxford Economics Australia based on AEMO data

But it also says much of the energy demand impact from AI-driven hyperscalers could be mitigated through advances in hardware efficiency, model compression, and smarter workload scheduling.

To date, energy consumption of data centres in Australia and globally has outstripped gains in server energy efficiency, but there’s more efficiency to come.

The Oxford Economics forecast shows energy efficiency gains offsetting most of the demand from the first wave of AI.

AI demand outlook for data centres, Step Change scenario.
Source: Oxford Economics Australia based on AEMO data

At the pointy end of these theoretical forecasts are data centre developers.

NextDC head of energy Shayne Kumar agrees growth will largely depend on AI adoption, telling last week’s Australian Clean Energy Summit NextDC expected AI inferencing to be a larger market in Australia than the more globally competitive AI training segment - something the Trump administration was working hard to capture.

Expert view

"Australia could also act as a hub for compute for Asia. We have excellent subsea infrastructure. But it will all depend on the monetisation of AI, the take up of AI, and maybe the next DeepSeek moment.

Things are going to get faster, things are going to get cheaper, but I believe in economics, and I believe in elasticity of demand. So as things get cheaper, I expect demand to increase and also productivity and broad uptake of compute.

We have cheap, firm, reliable power and the data centres will go wherever the power goes, subject to how far it is. Data centers and the energy transition are broadly interrelated - we’re both in the same game. We need more transmission, we need it faster, we need more firm generation.”

Shayne Kumar
Head of Energy, NextDC

Ask the tech bros

Another uncertainty in the mix is the extent to which hyperscalers, where the majority of AI energy demand is concentrated, will choose to invest in Australia.

Hyperscalers accounted for just 14% of consumption in FY18, according to Oxford Economics, but this escalated rapidly to 55% in FY25.

As a significant number of co-locator projects come online over the next few years, Oxford Economics forecasts hyperscalers will start to lose some market share to co-locators, holding a slim majority by 2030.

Data centre energy consumption by data centre type, Step Change scenario.
Source: Oxford Economics Australia based on AEMO data

David Scaysbrook, co-founder of hyperscaler developer Quinbrook Infrastructure Partners joked at the Australian Clean Energy Summit that “every property developer ever born is now a data centre developer” so there will be big gap between applications for connections and final demand.

And he says, the concentration of power in the world of hyperscalers could also make or break forecasts.

Expert view

"Largely it’s a story that's happening and playing out in the United States, and will continue to do that. There's no direct read across from everything you're seeing on LinkedIn and reading in the newspapers to an equivalent opportunity here in Australia and that's partly population based, it's partly policy based.

Trump's policy is for all that to happen in America. The tech bros in that consortium will try to make sure that that happens, and all the things that have been done under executive order to make it easier to develop, to build fast data centres in the United States, or tear down permitting barriers or environmental impact barriers, whatever it might be, is going to make it very hard to compete, from a capacity point of view, head to head.

The other thing to bear in mind is 40 to 50% of the entire market we're talking about is five companies, the hyperscalers. They're the ones that are not only the biggest purchasers of renewable energy on the planet, but they're the ones driving this predominantly. They speak for half of the demand, maybe more as we go forward. Most of them are active here in Australia, and it’s the decisions they make about where they locate their training."