What is certain, however, is that the rapid rise of generative AI demands astonishing amounts of computing power – along with astonishing amounts of energy.
While the potential business benefits of AI are closely examined and widely discussed, the environmental costs of this technology are, for the time being, largely overlooked.
But they will not be able to be ignored for long.
Why does AI use so much energy?
There are three main sources of AI’s heavy energy consumption, which is mainly concentrated in the enormous data centres that underpin our digital infrastructure.
The first is the sheer power used to train the AI models in the first place, which some estimate is doubling every 10 months.
The second is how much power they use every time they respond to a query. For instance, it has been estimated that an AI-enhanced Google search consumes 10 times the energy of a standard search.
The third is cooling. Keeping data centres servers from burning up requires huge amounts of cooling, often in the form of air conditioning. This often accounts for around 40% of data centre energy usage. It is also a prodigious user of water. Some estimates have put every ChatGPT as consuming half a litre of water.
Another sustainability challenge is the high demand for state-of-the-art chips needed to train large AI models. Chips are highly toxic to produce, according to US research institute AI Now, requiring huge amounts of energy to manufacture.
There is now a rising chorus of concerns around the headlong rush to deeply embed AI in many digital interactions, such as search.
It all leads to an urgent conversation that businesses need to have, says Gordon Noble, Research Director with the Institute for Sustainable Futures at the University of Technology Sydney.
“This is not a trivial issue, although I think it’s snuck up on us, and we’re only starting to wake up to how large it is,” he says.
A recent report by UTS found that few Australian companies have a clear view of the impact of their use of digital technologies on carbon emissions.
Yet CO2 emissions from software-related activities are estimated to account for 4-5% of global emissions and, by 2040, this figure may climb to 14%.
Change is coming
In October 2023, the Australian Accounting Standards Board (AASB) released an exposure draft (ED) that details required disclosure of climate-related financial information.
The ED includes three proposed Australian Sustainability Reporting Standards (ASRS Standards) that modify the baseline of the ISSB Standards with a climate-first approach, reports KPMG.
Luke Heilbuth, CEO of independent sustainability consultancy BWD Strategic, says the legislation is transformational for climate action.
“Until recently, it was unthinkable that the law would compel every major company to assess, manage and report on the implications of climate change,” he says.
The government is committed to improving the quality of climate-related financial disclosures and, to that end, the Treasury Laws Amendment (Financial Market Infrastructure and Other Measures) Bill 2024 aims to strengthen regulatory arrangements for Australia’s financial market infrastructure and to impose mandatory climate-related disclosure obligations on large businesses.
According to the Yale School of Environment, from next year the European Union’s AI Act, will also require “high-risk AI systems” (including models like ChatGPT) to disclose how much energy and other resources they use through their systems’ lifecycle.
The International Organisation for Standardisation, a global network that develops standards for manufacturers, regulators, and others, is also due to issue criteria for “sustainable AI”.
Noble says while these new standards and legislation may initially be targeted at large businesses, they are highly likely to flow on to SMEs over time.
“This will create a demand for those in the accounting industry to ‘look under the hood not just at energy consumption but climate risks more broadly,” he says. “Better information will also allow them to account for and reduce their emissions.”
Noble adds that, in the future, concepts such as data budgeting may be introduced.
Reducing climate footprint
SMEs looking to minimise the environmental impact of computing including AI and data storage need to begin by monitoring energy consumption.
From there they can measure savings made by using more energy-efficient hardware and networking equipment, implementing energy-aware job scheduling and adding renewable energy to their grid.
Other strategies could include ensuring that employees use energy-intensive AI applications judiciously or that the data storage facility they use is focused on energy efficiency. For instance, this might include introducing policies to not use generative AI or machine learning models where an equally effective non-AI tool is available.
“I also believe that as we start to really understand data, and data centres start to disclose their emissions, awareness will drive change,” says Noble.
Heilbuth says there are many benefits for a business that views sustainability as a way to build long-term value and resilience.
“Sustainability (unlike ESG) incorporates systems thinking; the wisdom to understand and accept that a business’s success is ultimately contingent on the flourishing of the society and the planet that surrounds it,” he says.
