Agir
6.6 billion cubic metres of water per year. That staggering figure is what data centres that house AI systems are projected to gulp down by 2027. Why water, you ask? Because these data centres need water to cool down and massive amounts of electricity to stay operational. For comparison, Mauritius in 2021, consumed 604 million cubic metres of water, which includes 2,6 million cubic metres from renewable water resources. Microsoft’ s water consumption increased by 34 %, while Google’ s rose by 20 % due to AI-related activities. In fact, training these AI models is yet another water-intensive process. GPT4, the latest ChatGPT model, consumes 700,000 litres of water over a two-week period, the equivalent of the water footprint of manufacturing 320 Tesla cars and emitted as much
CO 2 as 123 gasoline-powered cars.
In 2022, data centres, cryptocurrencies and AI accounted for almost 2 % of the global electricity consumption, reaching 460 TWh. The International Energy Agency estimates that it will reach between 620 and 1,050 TWh by 2026. If you’ re wondering what that number represents? Let’ s just say, France in 2023 consumed 445 TWh of electricity. And it doesn’ t end there. ChatGPT consumes 10 times more than a standard Google search. As of December 2024, ChatGPT was said to receive over 1 billion messages daily. The AI model BLOOM for instance? It emits 25 metric tons of CO 2
. Ten times more than the annual emission of an average French citizen.
Let’ s call it how it is. AI has taken over the world. No, really. Those Netflix recommendations based on your preferences? AI. Your spam filter recognising junk emails so your inbox stays spotless? Yep, AI again. That face recognition unlocking your phone and automatically tagging people on Facebook? You guessed it. AI. And then of course. The ultimate AI mammoth of a tool. The one that’ s gotten all of us writers on our knees, wondering if we’ re going to be out of jobs. Yes, you, ChatGPT. And while this field of science that can reason, learn and act in a way that would normally require human intelligence has certainly exceeded the scale of data that humans can analyse, it is both the hero and the villain in the global sustainability story. It also aids in optimising energy usage, predicting climate trends, reducing waste and accelerating the transition to renewables. CTrees and Planet Labs are two such companies that use AI to monitor degradation and provide actionable data for better conservation efforts.
The AI challenge is clear: can it optimise itself to become sustainable? If done fast enough, it clearly isn’ t all doom and gloom. For instance, the Abu Dhabi National Oil Company generated $ 500 million in value and reduced carbon emissions by a million tonnes through their AI powered energy-saving efforts. That figure is the equivalent of taking off 200,000 gasoline-powered cars off the roads. Smarter grid management, which is the process of optimising the flow and distribution of electricity within an electrical grid, a network that supports generation, transmission, distribution, storage and control of electricity, could integrate 2,600 GW of backlog renewable energy( solar, wind and more) back to the power grid of the United States as from 2025. The US Department of Energy has launched an AI programme called AI4IX to speed up this process, which would help predict energy demand and automate approvals. In a nutshell: the US alone has got the potential to unlock a massive clean energy source waiting to be utilised, that could cut building emissions by 14 to 25 %, with predictive AI for wind farms that could boost efficiency by 20 %.
AI pollutes. Yet AI fixes. And now it’ s a race. Will AI optimise itself before it drains the world of its resources? Time is ticking. So the next time you’ re using ChatGPT to write your wedding vows“ from the heart”, remember AI is out here trying to save the planet all while guzzling water and spitting CO 2
.
Priorities.
56