Generative artificial intelligence (AI) is curating your social media feed, studying your bank loan application, drafting legislation and giving you directions.
It will also soon generate millions of tonnes of waste.
Tech companies are churning through truckloads of computer parts as they race to build bigger data centres to train better generative AI models — algorithms that draw on training data to create new content.
A new study, published today in Nature Computational Science, has estimated generative AI could generate up to 2.3 million tonnes of electronic waste (also called e-waste) per year by 2030.
To put that in context, that's equivalent to discarding 13.3 billion iPhone 15 Pro phones.
"We were surprised by the magnitude of the projected e-waste," Asaf Tzachor, a co-author of the study and researcher at Israel's Reichman University, said.
"To the best of our knowledge, this is the first comprehensive study that quantifies and projects e-waste generated specifically by generative AI technologies."
Previous studies have focused on AI's energy usage (generating an image takes about as much energy as fully charging your phone) and water usage (a 50-question ChatGPT conversation may use about half a litre).
Generative AI produced around 2,400 tonnes of e-waste in 2023.
This relatively small figure will shoot up when vast data centres built to develop products such as ChatGPT are gutted of computer parts succeeded by new designs.
Meanwhile, ever-larger data centres are coming online, with $US36 billion invested in AI infrastructure globally in 2023.
These will add thousands of tonnes to the e-waste stream.
Microsoft's rush to build data centres, like this one in Wyoming, has seen its carbon emissions spike. (Supplied: Microsoft)
But it's not all bad news.
The authors found extending the life span of existing computer infrastructure, reusing reusable parts and recycling valuable materials like copper and gold could reduce e-waste generation by up to 86 per cent.
Most AI hardware replaced every three years
Although generative AI is widely available on smartphones and desktop computers, its physical infrastructure is housed largely out of sight, behind the walls of data centres.
The contents of these buildings are closely guarded secrets.
To calculate the amount of e-waste generated by these data centres, the researchers first estimated the amount of computational power that would be required to train generative AI models under different adoption scenarios.
Then they translated this demand into hardware requirements, estimating the number of servers and electronic circuits such as graphics processing units (GPUs).
This may look high-tech, but even cutting-edge components are made obsolete within three years. (Supplied: Microsoft)
They then calculated how technological improvements would reduce hardware requirements.
"Considering typical hardware life spans — often around three years due to technological advancements — we projected when this hardware would be decommissioned," Dr Tzachor said.
"This approach allowed us to estimate the amount of e-waste generated over time."
Under the "aggressive adoption" scenario, where AI was used by all internet users, 4.5 million tonnes of e-waste would be produced between 2020 and 2030.
Most of the e-waste would be clustered in Europe, North America and East Asia, which have most of the world's data centres.
Relatively simple "circular economy strategies" would have a major effect on e-waste generation, the researchers found.
About 2.8 million tonnes of obsolete AI servers could be avoided by extending their life span by one year.
Much of the world's electronic waste is shipped to countries such as Kenya, where it ends up in landfill. (Getty Images: James Wakibia/SOPA Images/LightRocket)
Dismantling, renovating and reassembling obsolete modules, such as GPUs, so they may be used for other, less intensive forms of computing could reduce e-waste by 42 per cent.
"To the best of our knowledge, widespread adoption of these practices in the industry is still limited," Dr Tzachor said.
"Challenges like data security concerns and the need for high-performance hardware can make reuse and recycling more complex."
In fact, some measures are making the e-waste problem worse.
US restrictions on the sale of advanced GPUs to countries such as China means data centres are forced to use outdated server models, which results in more e-waste.
A one-year delay in obtaining the latest chips could result in a 14 per cent increase in e-waste, the researchers calculated.
The old toasters are piling up
Generative AI e-waste accounts for a small but increasing fraction of global e-waste, which is defined as any product with a plug or a battery.
E-waste is now the planet's fastest-growing waste stream. A record 62 million tonnes of e-waste was produced in 2022, according to the UN's global e-waste monitor report.
The figure was up 82 per cent from 2010 and on track to rise another 32 per cent to 82 million tonnes in 2030.
The report states:
The 62 million tonnes of e-waste generated in 2022 would fill 1.55 million 40-tonne trucks, roughly enough trucks to form a bumper-to-bumper line encircling the equator.
Small electronic equipment such as phones, microwaves and vapes constitute the largest category of e-waste in terms of mass.
Less than a quarter of global e-waste is formally collected and recycled. In Australia, about half of e-waste is sent to recyclers.
Australians rank fourth in the world for production of e-waste per capita, generating 22 kilograms of e-waste per person per year, more than double the global average.
A recent report found most of Australia's top companies don't include e-waste in their sustainability reports.
'We're running out of resources'
Damien Giurco, from the Institute for Sustainable Futures at the University of Technology Sydney, said the study's e-waste projections appeared accurate.
"We need producer responsibility and product stewardship," Professor Giurco, who was not involved in the study, said.
"If you're putting equipment onto the market, you need to be making sure [there's] a pathway that supports reuse of materials and recycling."
Nvidia's recently launched Blackwell platform for AI data centres weighs 1.2 tonnes. (Supplied: Nvidia)
Companies importing electronic goods or building data centres in Australia are currently under no legal obligation to consider the volume of e-waste they generate.
That could change with moves to introduce a circular economy framework. A ministerial advisory group to the federal government released an interim report earlier this year recommending national and sector-based circular economy targets and specific circularity requirements for products.
Lisa McLean, a member of the ministerial advisory group and chief executive officer of research and advocacy group Circular Australia, said tech companies could be made responsible for their e-waste.
"There's more gold and silver in a tonne of iPhones than a tonne of ore from a gold or silver mine."
Corporate net-zero commitments were also pushing companies to consider the embodied carbon in their products, or the amount of greenhouse gases associated with a product's life.
"I think companies understand that this is an economic response to the crises we're facing," Ms McLean said.
"We're running out of the resources and we're not going to get to net zero without a circular economy."