Tech giants’ eyewatering investments in genAI data centres threaten to consume energy faster than renewable sources can produce it – with a new report warning of a “continued dependence on fossil fuels” that threatens new government emissions reduction targets.
Energy-hungry genAI wasn’t a factor back in 2020, when Microsoft’s CEO promised the company would be “carbon negative” by 2030 – but its emergence destroyed those plans, with Microsoft since admitting that its emissions are up 29.1 per cent from its 2020 baseline.
Indeed, a newly released Stand.Earth analysis found the firm’s North American data centres used 12,400 gigawatt hours (GWh) during 2023 alone – more electricity than the entire state of South Australia used last year – but that’s only the beginning.
Microsoft, for example, recently switched on Fairwater, a Wisconsin data centre it calls “the world’s most powerful AI datacentre”; is investing $45 billion ($US30 billion) in UK AI capacity; and finalised a $26 billion ($US17.4 billion) deal with AI datacentre builder Nebius.
These and other investments are expected to grow the power consumption of Microsoft AI data centres a further 600 per cent from 2024 to 2030 – enough energy to power almost 10 million households, equivalent to nearly every home in Australia, for a year.
Energy companies, in turn, are relying heavily on fossil fuels to keep up, with Stand.Earth warning Microsoft’s “current approach masks a continued dependence on fossil fuels, with rapidly increasing energy demand on track to drive a lock-in of fossil fuel infrastructure.”
“The findings exemplify a clear connection between AI development and fossil fuel build out,” it said, with “dramatically increased pollution” meaning “the company’s pursuit of AI dominance is threatening to undermine [its] climate leadership position.”
'Ambitious' emissions reduction targets now seem even more so
Microsoft’s surging energy consumption – and the fact that other companies are doing exactly the same thing – dilutes the federal government’s newly announced emissions reduction target, which it spruiked as a “credible contribution” to fighting climate change.
Its Net Zero Plan and “ambitious, achievable” new goal – to reduce greenhouse gas emissions by 62 to 70 per cent below 2005 levels by 2035 and reach Net Zero by 2050 – are “a significant step up from our 2030 target” of 26 to 28 per cent below 2005 levels, it said.
National emissions are already down by 29 per cent from 2005 levels, with future reductions guided by billions in funding and plans for six industry sectors – including an electricity and energy sector plan that’s predicated on cutting electricity emissions by 85 per cent by 2030.
With that sector producing 34 per cent of Australia’s carbon emissions – and the ‘built environment’ of buildings and homes an additional 5 per cent – the government hopes the plans outline a “stable and navigable investment environment” for renewables investment.
“We know every effort to tackle emissions today will help avoid the worst impacts of climate change in the future,” Minister for Climate Change and Energy, Chris Bowen, said in joining Prime Minister Anthony Albanese and Treasurer Jim Chalmers in announcing the contentious targets.
The plan “builds on what we know are the lowest cost actions we can deliver over the next decade while leaving room for new technologies to take things up a gear.”
But new technologies didn’t get the memo
Hunger for AI computing capacity, however, threatens to derail those ambitions – with the Climate Change Authority recently warning AI buildouts will “[place] additional pressure on the renewables buildout.”
A recent ITU report warned that data centres’ energy consumption increased by 12 per cent annually from 2017 to 2023 – four times as quickly as general electricity consumption – and that AI giants’ use, in particular, grew 150 per cent annually through 2023.
Tech giants are continuing the growth trajectory: Amazon, for one, recently committed to invest $20 billion in its Australian data centres by 2029, albeit with a companion investment in renewables, while OpenAI is driving the $750 billion US Stargate AI project.
Each of these facilities, and the many more to come, relies on AI-optimised processors from Nvidia, which is laughing as it rides the AI surge all the way to the bank – and will surge demand further after committing to invest up to $150 billion in genAI pioneer Open AI.
A recent Bain & Company analysis found AI’s computational needs are growing more than twice as quickly as Moore’s law, requiring $750 billion ($US500 billion) in new data centres every year to keep up – demanding 200 gigawatts (GW) globally by 2030.
Can our infrastructure even keep up?
Local institutions are joining the rush, with La Trobe University recently switching on an AI supercomputer, Monash Uni planning a green system, ResetData launching its AI-F1 system, the ADF launching an AI computer, and Sovereign Australia AI setting records.
Yet for all the benefits of AI, each new facility exacerbates energy generators’ challenges to provide enough energy – and threatens the government’s ambitions to cut emissions.
Finding balance is proving tricky: a year after it was slammed for falling behind its carbon emissions reductions targets, for example, in June the NSW Government launched a tech-focused Investment Delivery Authority to fast-track the approval of new data centres.
Ten data centre applications, from the likes of Microsoft, Amazon and AirTrunk, have been approved since 2021, with Reuters noting that the facilities will attract $6.6 billion in investment – and that cooling them will consume 2 per cent of Sydney’s total water supply.
Sydney Water believes data centres could consume up to a quarter of Sydney’s water by 2035.
