Are AI’s energy demands spiralling out of control?

 By the CogX R&I team

July 09, 2024

You've likely seen the alarming headlines... AI guzzling electricity like entire countries and warnings about overloaded power grids. Sometimes it certainly feels like we're in the midst of an AI energy crisis. But how concerned should we really be about AI's rising emissions?

Well, it's actually a more complex issue than those eye-catching headlines suggest.

The first thing people bring up when discussing AI's climate impact is its voracious appetite for energy and they're not wrong to do so. Training and running large AI models is undeniably energy-intensive. A recent study by Goldmand Sachs showed a single AI query on platforms like ChatGPT can use up to 10 times more electricity than a regular Google search. What's more, generating just one image using a powerful AI model can consume as much energy as charging over 500 smartphones.

However, not all AI tasks are equal in terms of energy consumption. Text generation, for instance, is much less energy-hungry than generating images. Using smaller, specialised models instead of giant all-purpose ones can significantly reduce energy use. But even with increased use of these smaller models, if you process hundreds of queries per second, the carbon footprint adds up quickly.

The International Energy Agency (IEA) estimates the total electricity used by data centres could double from 2022 levels to a staggering 1,000 terawatt-hours (TWh) by 2026, equivalent to Japan's annual energy consumption. But while pinpointing AI's exact share of this energy use is challenging, the trend is abundantly clear: our computing needs are skyrocketing, and AI is a major driver.

So what can be done to solve this? While some believe that this challenge could spark a revolution in greener, less carbon-intensive technologies — companies will need to find ways of slashing their energy bills and reduce their greenhouse gas emissions — the jury is still out on whether companies, especially Big Tech firms, can innovate quickly enough to meet their climate pledges.

It's not just about energy use

Microsoft, which pledged four years ago to reach zero greenhouse gas emissions by 2030, is now seeing its emissions rise. Since 2020, the tech giant’s total carbon emissions have risen 30%, due in large part to the company's AI activities.

Interestingly, it's not necessarily the direct electricity use of AI that's boosting Microsoft's emissions, at least on paper. The company claims to meet its electricity needs, including for AI, with renewable energy credits and agreements. Instead, the culprit seems to be the massive infrastructure buildout required to support AI development. 

Microsoft is pouring $50 billion into expanding its data centres between July 2023 and June 2024 alone. Building these facilities requires carbon-intensive materials like steel, cement, and semiconductor chips.

Google is facing a similar situation. Their emissions jumped a staggering 48% since 2019, again largely due to data centre expansion for AI. Their 2030 net-zero goal, just like Microsoft's, is starting to look like a stretch, being deemed ‘extremely ambitious’ by their own environmental reporting team. 

The data centre boom

Driven by the surging demand for AI, countries worldwide are racing to expand their data centre capacities. Globally, investments in data centre infrastructure have ballooned to over $22 billion in the first half of 2024, with the US and EU attracting the highest levels of funding. Notably, the appetite for data centres within Europe is growing exponentially, increasing its share from 6% in 2022 to 20% in 2023.

Image from Linklaters

Countries in Asia are also racing to expand their data centre capacities. In Southeast Asia, Singapore plans to increase its data centre capacity by more than a third, while Malaysia is set to host its first Google data centre. China, meanwhile, plan to double their data centre racks by 2030, a move that's projected to consume a staggering 6% of the nation's total electricity demand by 2026.

This expansion is not just about power but also about water. Cooling systems in data centres require vast amounts of water. In China alone, annual water consumption by data centres is estimated to reach over 3 billion cubic metres by 2030, enough to cover the annual residential water usage of Singapore.

AI’s energy appetite is a challenge, not a crisis

AI isn't inherently bad for the environment. In fact, it has the potential to become a powerful weapon in our fight against climate change. Imagine AI optimising energy grids, developing cutting-edge sustainable materials, or even predicting weather patterns for maximised renewable energy production.

But unlocking this future potential hinges on our ability to harness AI's power while ensuring it doesn't come at the cost of the planet.

The key ultimately lies in the choices we make today. AI's thirst for energy doesn't have to be a climate catastrophe. Instead, it can be the spark that supercharges the shift to renewables and green data centres.

By acknowledging the environmental cost of AI and taking proactive steps to mitigate it, we can ensure this powerful technology helps us build a sustainable future, not hinder it. 

The alternative — a world choking on the ‘cloud’ — is simply not an option.