Published on the 07/05/2025 | Written by Heather Wright

The wildcard that is AI…

Auckland electricity distributor Vector has flagged ‘significant uncertainty’ around increasing power demands as hyperscale data centres as generative AI takes a toll on power grids around the world.

The scramble for electricity has also seen academics from the University of Sydney, Western Sydney University and University of Technology, among others, calling for Australia to lead in sustainable AI computing through building renewable-powered data centres that not only support GenAI but help stabilise the energy transmission grid, lower power prices and drive regional economic development.

“Hyperscale data centres typically request 20MVA in the first phase, which is roughly equivalent to 8500 homes.”

AI models are typically much more energy-intensive that the more traditional data retrieval, streaming and communications applications that drove data centre growth over the past 20 years.

At 2.9 watt-hours per ChatGPT request, AI queries are estimated to require 10 times the electricity of traditional Google queries, which use about 0.3 watt-hours each. Emerging, computation-intensive capabilities such as the creation of images, audio and video based on user prompts, and other AI applications, meanwhile, have ‘no precedent’ the Electric Power Research Institute (EPRI) says.

Vector, which is New Zealand’s largest electricity distributor, has forecast a 60 percent rise in demand from data centres in the coming decade and says high demand for data centres in Auckland is creating unique challenges for the region.

Microsoft opened its first hyperscale data centre in Auckland late last year with further expansion planned. AWS is also building three data centres, while telco Spark is building a large-scale data centre on Auckland’s North Shore and data centre developer and operator CDC has a third campus build underway in Auckland.

CDC shareholder Infratil has said the growth of data centres in Auckland is expected to match that of Sydney.

Vector says it’s receiving ‘many’ more customer connection requests for hyperscale data centres than it did just a few years ago and the requests are ‘complex, large and have significant impact on our future planning for the part of the network they are located in’.

“For example, hyperscale data centres typically request 20MVA in the first phase of multi-stage developments, which is roughly equivalent to 8500 homes.”

But while Vector acknowledges the need to invest in system growth given the hyperscale data centre requests, it says there is ‘significant uncertainty’ around the rate at which demand will grow to match the capacity requested since that depends on the adoption of the data centre services by data centre customers, along with the type of services run in the data centre and, most importantly, how customers choose to use the new services.”

It’s a global problem. In Australia, data centres account for around six percent of Australia’s electricity use, with that figure expected to double by 2030.

CDC is currently constructing what it says will be the largest data centre campus in the Southern Hemisphere, at the Marsden Park Industrial Precinct in Sydney’s North West. With 504MW total ICT capacity – the equivalent of 140,000 homes – it will be capable of hosting cloud, classical, high performance compute and AI related workloads. Work has also started on a new campus in Melbourne.

The International Energy Agency has forecast electricity demand from data centres worldwide to more than double by 2030, with AI ‘the most significant driver’ of the increase. Electricity demand from AI-optimised data centres are projected to more than quadruple by 2030 the IEA says in its recently released Energy and AI report (which ironically, includes the ability to use a virtual agent to use AI to explore the report).

It does, however, pay to note that data centres accounted for just 1.5 percent of the world’s electricity consumption in 2024. Local impacts, however, can be far more pronounced, the report notes, with data centres frequently clustered in specific regions, resulting in a concentration of power intensive facilities in a small area.  AI-focused data centres can draw as much electricity as power-intensive factories such as aluminium smelters.

The Electric Power Research Institute (EPRI) has also noted that AI, and more specifically generative AI models, are creating a ‘fundamental uncertainty’ when it comes to projecting data centre load.

“While AI applications are estimated to use only 10 percent-20 percent of data centre electricity today, that percentage is growing rapidly.”

There’s also, of course, the DeepSeek factor, but while the Chinese company initially sent shockwaves across the investor market with suggestions that its efficiency gains could impact demand for data centre capacity, and saw questions over whether the money pumped into the sector might have been ‘misguided’ many now believe the advent of the more efficient, cheaper models, could actually an accelerant for AI – and further data centre growth, both large-scale and distributed.

But at the same time, AI does offer the potential to optimise energy consumption and reduce waste and better forecast energy demand and supply – and history has shown demand for increased processing is often largely offset by data centre efficiency gains according to the EPRI.

Advanced energy-efficient cooling solutions, power management systems and advanced in computational and supporting hardware can help reduce overall electricity consumption, with heat reuse helping reduce the overall energy usage of an area.

Amazon’s 1.1 million square foot Doppler building has recovered up to 3200MWh of excess heat from a nearby data centre since 2016, the EPRI notes.

Spark’s planned large-scale North Shore data centre is part of a 43-hectare development with global surf park creators Aventuur. It will include an onsite seven-hectare solar farm which will supply renewable energy to the data centre, while a heat exchange system, using excess heat produced by the data centre, will be used to warm the water of a nearby surfing lagoon.

Microsoft, meanwhile is using water-free technology to cool it’s Auckland facility and has a deal with Contract Energy for renewable power (88 percent of New Zealand’s power coming from hydro, geothermal and wind”.

Australian academics, however, want Australia to go further and have suggested integrating genAI data centres into the energy system as ‘flexible, grid-responsive infrastructure’, leveraging renewable energy, demand management and policy tools to support stability and lower costs, while establishing Australia as a key provider of AI computing infrastructure in Asia Pacific.

“Co-located generative AI data centres and renewable energy facilities could enhance grid stability while generating higher economic and environmental value than traditional energy export models,” they say in Generative AI Data Centres for Renewable Energy Integration and Grid Stability: Fostering a Sustainable Economic Future.

The paper suggests hyperscale cloud providers are willing to pay premium rates for renewable-powered computing capacity, and that could be used to fund critical grid infrastructure improvements and create skilled employment opportunities in regional areas.

“The integration of AI data centres with renewable energy infrastructure offers a transformative opportunity for Australia’s economy and energy landscape.

“By colocating data centres with renewable energy sites, Australia can optimise energy use, reduce transmission losses, and stabilise the grid. These efficiencies can lead to more consistent electricity prices, potentially lowering costs for households and enhancing grid reliability,” they say.

“Crucially, this strategy positions Australia as a leader in sustainable digital infrastructure, attracting global investment and fostering innovation. The resulting high- value job creation and increased competitiveness could significantly boost economic resilience.”