Data centres are vital for the future and AI but their environmental footprint can be a problem

Data centres could use the equivalent of a quarter of Sydney's current annual drinking water supply within a decade, according to the state's peak water utility.

Currently data centres use approximately 3.5 billion litres of Sydney's drinking water supply per year — which is less than 1 per cent of total demand.

Sydney Water, Australia's largest water utility, is preparing for that to jump to 25 per cent by 2035.

For context, Sydney's desalination plant supplies 15 per cent of drinking water, or 91.25 gigalitres, each year.

As Australia looks to position itself as a global data centre hub, industry and governments are investing in new ways to wean these buildings off the drinking water supply.

Artificial Intelligence (AI) is driving a surge in new data centres — the buildings that house the computer servers that store, process and distribute data.

While AI is helping to solve some of the world's biggest environmental challenges, it also brings its own.

Data centres have become critical infrastructure, supporting everything from daily internet use and online services to health services and national security.

The emergence of generative AI (also known as gen AI) is driving further demand for large data centres with very high computational and energy demands.

A big part of ensuring these essential systems are kept running is temperature control — making sure the servers don't overheat.

This makes data centres power-hungry beasts that, depending on the system, sometimes require significant amounts of electricity and potable (drinkable) water.

Data centres have historically used water to transfer heat away from their servers via cooling towers or chiller systems, which can result in huge amounts of water lost through evaporation.

Australia currently hosts more than 260 data centres, about a third of which are in Sydney.

According to Sydney Water, all 89 of those located there currently draw from the public drinking water supply — but this can't last forever. 

High quality H2O

Many data centres are already investing in more water-efficient cooling systems, like closed loop, immersion and direct-to-chip liquid cooling.

Some companies — like CDC — have been using fully recycled water systems for years.

The company was founded in Canberra in 2007 during the Millennium Drought – one of the longest recorded droughts in Australian history, which started in the late 1990s.

This meant the company had to come up with a technology that required almost no operational water consumption for primary cooling, resulting in a closed-loop cooling system designed to save up to 5 billion litres of water each year.

But not all data centres operate this way, which is why Sydney Water has had to come up with solutions that won't impact the public drinking water supply.

Sydney Water senior project manager James Harrington said one of the main solutions was wastewater.

Speaking from the wastewater treatment and purification facility at Quakers Hill, Mr Harrington explained that data centres require high-quality water for cooling their systems, which could be achieved through wastewater treatment technology.

"As data centres become more prevalent, they require a lot of water for their cooling and we need to make sure that we're enabling that," he said.

Acting Sydney Water CEO Paul Plowman said the predicted rise in water demands from data centres in Sydney was based on forecasted growth.

"Those numbers equate to the amount of data centres that are forecast and the processing power that those data centres will have, because the more processing that happens the hotter the chips and the more cooling that is required," Mr Plowman said. 

He said Sydney residents should be assured they have a plan for ensuring drinking water supplies would not be impacted by the growth in data centres.

"It's really important that people understand that there's many ways to provide the water, and drinking water isn't always the answer."

Location, location, location

These issues have been on the mind of Wayne Rylands, the CEO of Sydney's Ryde Council, which takes in the Macquarie Park innovation precinct.

"We have around 11 data centres that are either already built, under construction, or planned to come in here," Mr Rylands told 7.30.

"But there's nothing stopping that [becoming] 20 or 30 or even 40.

"We're not ready for that significant use of water, significant use of power and significant loss of vegetation on some very big sites."

According to Sydney Water, the data centres located in Macquarie Park could use up to 30 megalitres per day within a decade — which represents about 2 per cent of the entire city's current daily drinking water demand.

"We're not against data centres," Mr Rylands said.

But Mr Plowman said building the centres in metropolitan areas, close to large sewers, could be key to avoiding their use of drinking water supplies.

"Macquarie Park is not too far away from the Northern Outfall Sewer. So whether that 2 per cent will actually be drawn from the drinking water system is something that we're working through with various proponents," he said.

Danielle Francis from the Water Services Association of Australia says the nation has a unique opportunity to become a resource-efficient global data centre hub.

"The growth of data centres creates a unique opportunity to help grow our national economy and productivity, but it will create extra demand for valuable water resources," she said.

"Nationally, we are working with the data centre sector to understand the rapid evolutions in technologies and how we can sustainably incorporate that to our customer mix and future demand planning."

Environmental challenges and solutions

It's hard to know how much water and energy we, as individuals, consume each time we use a generative AI tool.

Last week Google released a research report which found that the median Google Gemini text prompt used about 5 drops of water, and the energy equivalent of around 9 seconds of TV, with the tech giant citing dramatic efficiency gains over the past 12 months. 

These figures varied widely from a 2023 US research paper, which estimated that ChatGPT-3 could drink up to 500ml of water for roughly 10 — 50 medium-length responses, when considering both the freshwater needed for both cooling the servers and for electricity generation.

AI expert Professor Jon Whittle from the CSIRO welcomed Google's research as a step towards greater transparency in the industry, but said people should remember that individual use did not represent the full picture.

"When you look at the environmental impact of AI, you have to look at the full life cycle," he said.

"It's not just you using a [generative] AI prompt at home. There's also the training of the AI models, which can take a lot of energy.

"And there's also the ... e-waste."

Whilst AI has its environmental challenges, it also holds many of the solutions.

This dichotomy is on full display inside the walls of the Pawsey Supercomputing Research Centre in Perth.

Here, the most powerful research supercomputer in the southern hemisphere — called Setonix — is being used for important environmental research, including state-of-the-art climate change projections.

The building faces all the same temperature control issues as data centres, but uses a rare geothermal method to cool its system through an underground aquifer, allowing it to save more than 7 million litres per annum. 

According to executive director Mark Stickells, this geothermal cooling system is uncommon because it relies on site-specific conditions: suitable aquifers, stable groundwater flows, and a design capable of handling the sustained thermal load of a supercomputer.

“Most data centres opt for more standardised and replicable cooling methods, as geothermal requires higher up-front investment, complex integration, and long-term environmental management," he said.

Despite being ranked as one of the "greenest" systems of its kind, Setonix still requires about 1.5 megawatts of power.

"Which in comparative terms is that of a small suburb or a large shopping centre," executive director Mark Stickells said.

"But this [supercomputer] is only a fraction of the size of the systems that are training the AI models of today.

"So every time we pick up our phone to do something very cool with AI, I think we should be thinking about the infrastructure, the power, the water, the service that's needed, to provide that to you."

Watch 7.30, Mondays to Thursdays 7:30pm on ABC iview and ABC TV