Power-hungry AI hurts climate goals
2024-09-20
| Policy Research and Advocacy Team, Friends of the Earth (HK)
"Write an article about the impacts of the AI boom on climate
progress”—in recent years, artificial intelligence (AI) has become increasingly
entrenched in many aspects of our lives. This ranges from everyday use like
ChatGPT for writing support or Midjourney for image generation to the more
technical applications in medical imaging analysis and weather forecasting. In
some of our past articles, we have also covered how AI can provide solutions to
environmental issues, such as smoothening traffic flows and optimising grid
stability.[1]

AI-generated images can
replace the need for stock photos, but getting the ‘right’ image can be costly
(Image source: WhatTheyThink)
The AI boom is bringing in convenience and productivity; the demand
for faster and greater computation power is however coming at a huge cost to
the environment, as running data centres to process all that digital
information takes a lot of electricity. What you might consider a simple
request on ChatGPT for example uses ten times as much electricity as a web
search.[2] Making
one AI-generated image could have charged your smartphone half-full.[3] This
adds up when you have millions to billions of users making requests every day.
Newer AI models demand more
computing power, and consequently more energy (Image source: The Economist)
In Google’s latest environmental report, it points to AI for the 48%
growth in greenhouse gas emissions since 2019, and for setting back the tech
company’s 2030 net-zero goal.[4]Microsoft also saw emissions rose by 30% from 2020 in its race to be an AI
leader.[5] The
International Energy Agency predicts by 2026, electricity demand will double
for data centres.[6]More specifically, the AI industry alone will use ten times as much electricity
compared to now.
Electricity is not the only concern; to keep servers cool and
running at peak efficiency, data centres are also very thirsty. In the U.S.,
data centres rank among the top 10 water-consuming industries.[7]Closer to home, research suggests data centres in China consume 1.3 billion
cubic metres of water annually,[8]and doubling by 2030, as AI becomes more commonplace.[9]Paired with climate change, this is driving conflicts with local communities in
water-stressed regions.[10]
Data centre operators like
arid regions for their abundance of low-carbon energy, but water demands put
them in conflict with local communities (Image source: Jim Todd/Reuters)
With its incredible positive potential, AI is here to stay. OpenAI
has already announced it is developing a GPT-4 successor.[11]As newer, larger generative models are expected to come with increasingly
larger environmental footprints, experts have called on data centre operators
to decarbonise operations by investing in low-carbon energy. The use of massive
‘hyperscale’ data centres can also enhance energy efficiency by consolidating
small data centres into a centralised facility.[12]Google has applied load shifting techniques to data centre operations, shifting
computing load to align with times when low-carbon energy is plentiful.[13]
The use of AI is unlocking new potential, but its hunger for
electricity is also making it harder for the world to decarbonise and meet
climate goals. As these technologies continue to mature and popularise, developers
must make sure their benefits and climate and environmental impacts are
balanced.