‘Out of sight, out of mind’: The hidden environmental danger of AI
Italy's University of Pisa is one of the institutions trying to lower data centre water consumption. Photo: Getty
As generative AI sweeps the world, the potential for job losses and infringement on creative work have been some of the biggest talking points.
But the rapid uptake of the technology is also quietly a major environmental risk.
University of Technology Sydney Institute for Sustainable Futures research director Gordon Noble told The New Daily the growth of AI has already had an impact on energy usage and water consumption.
“It’s time for us to wake up about how much the rise of AI technologies in particular are going to impact on the environment in the future,” Noble said.
“It’s not to say that we shouldn’t use AI.
“It’s to say, just like everything else … we need to understand … the environmental impacts, and then manage those.”
AI Revolution
RMIT University Centre for Urban Research senior lecturer Ascelin Gordon said just as the Industrial Revolution had a serious long-term impact on the environment, the AI Revolution will do the same.
But while London was blanketed in soot and smog by coal-burning factories, the worldwide nature of computing and connectivity means the impacts of AI will seem far removed for many.
“Out of sight, out of mind,” Gordon said.
How does AI impact the environment?
AI is trained and run using physical data centres, which also run the internet – but AI requires much more energy than traditional internet use.
A ChatGPT request consumes almost 10 times the amount of electricity as a Google search, according to the International Energy Agency.
As AI is increasingly developed and used, the industry is predicted to consume at least 10 times more electricity in 2026 than it did three years prior.
That’s not a great sign for the world’s battle with climate change given data centres and data transmission networks already account for almost 1 per cent of global energy-related greenhouse gas emissions.
OpenAI’s ChatGPT helped open the floodgates of generative AI. Photo: Getty
Concerning water needs
Data centres also use water to cool down onsite servers; as the extra work running AI requires can cause servers to heat up faster, more water is needed to keep data centres functioning.
Microsoft’s 2024 Sustainability Report revealed the company’s emissions rose 29 per cent and it used 23 per cent more water in 2023 as “new technologies, including generative AI … create new challenges for meeting sustainability commitments across the tech sector”.
This is a particularly big issue in drought-prone regions; last year, Google received backlash for its plans to construct a data centre – expected to consume millions of litres of water per day – in Uruguay, which was facing its worst drought in 74 years.
“That’s an issue in Australia because … you tend to locate a data centre in the urban fringe of a city,” Noble said.
“And so therefore [they’re] drawing water in areas where you’ve already got high demand for water.”
Gordon pointed out the construction of data centres is also a drain on the environment, from mining metals and minerals for computer chips to the energy required while manufacturing necessary equipment.
Compounding this impact, the equipment in data centres may only be relevant for up to five years before needing to be replaced, he said.
International progress
Internationally, countries such Ireland and the US are coming to grips with data centres’ excessive demands on finite resources.
In Singapore, where data centres consume high levels of water to keep cool in the tropical climate, the government launched a world-first standard for optimising energy efficiency in data centres.
This is projected to result in data centres gradually increasing the temperatures they can operate on, which will in turn reduce cooling needs.
Google has also taken some steps to mitigate its environmental impact. In March, the company announced the heat generated by its data centre near Hamina, Finland would be rerouted to supply local households, schools and public service buildings.
Lack of reporting in Australia
In Australia, a clear picture of the full impact and future risks of AI’s environmental impact can’t be formed because there is no standardised reporting of data centres locally, Noble said.
A report Noble co-authored in 2023 that surveyed corporate sustainability managers found the majority agreed organisations can’t reach sustainability goals without significantly reducing data centre energy usage.
But more than half reported having received insufficient or no sustainability-related data from data centre service operators.
Only 22 per cent of respondents said their organisation paid enough attention to data centre energy consumption.
There is hope circumstances could soon change; Noble said proposed mandatory sustainability disclosures meant we could soon see more information come to light.
There are also numerous startups, including some based in Australia, developing methods to help data centres operate with more energy efficiency.
Tighter legislation needed
Gordon said restricting the development and use of AI would be ideal, but achieving this would be politically challenging.
Tighter legislation aimed at reducing the environmental impact of data centres may be a more viable option, she said.
The next key move would be making sure countries work together to enforce sustainability requirements.
“Because it’s such a global phenomenon, if one country has much tighter regulations, these big multinational companies can just put their data centres in other countries that don’t have such tight regulation,” Gordon said.
“It’s a really tricky co-ordination problem to try and manage this.”