Widely touted as the Fourth Industrial Revolution, Artificial Intelligence or AI holds immense potential to drive productivity gains and raise standards of living. Data centres form the key infrastructure behind this emerging technology — facilities equipped with clusters of chips that provide the vast computational power required to train and operate complex AI models. As interest in AI surges, so does the demand for data centres, as evidenced by the American tech giants Meta, Alphabet, Amazon and Microsoft collectively splashing out US$180 billion on data-centre infrastructure in 2024.
Beneath the dizzying capital spending, what has been mostly underappreciated up to now is the high energy consumption that data centres require to power servers and keep them cool. As AI models become exponentially more powerful, electricity demand has also surged in tandem. According to AI startup Hugging Face and Carnegie Mellon University, even the seemingly simple task of creating an image via generative AI uses as much energy as fully charging a smartphone! Likewise, the University of Technology Sydney estimates that a single query to an AI-powered chatbot uses ten times as much energy as a traditional Google search. Consequently, data centres currently account for 1-2% of overall global energy consumption. For context, this is comparable to the global energy consumption of the airline industry. With the World Economic Forum predicting that the computing power dedicated to AI will double every 100 days, this percentage is only set to increase.
Greater energy consumption has naturally been accompanied by larger carbon emissions, given that many data centres remain predominantly powered by dirtier fossil fuels. Unlike renewable energy, which is subjected to the volatilities of the weather, fossil fuels are reliably available round-the-clock to allow data centres to continuously maintain peak operation. According to the MIT Technology Review, carbon emissions from data centres in the US have tripled since 2018, and in the 12 months to August 2024, they were responsible for 105 million metric tons of CO2 emissions, only slightly below the emissions from domestic commercial airlines. Furthermore, environmental concerns could be more pressing in emerging markets, where regulations may be laxer, and where an increasingly number of data centres are being built.
The good news is that the world does not necessarily have to be forced into choosing between AI or the climate. Singapore, for example, is already implementing a Green Data Centre Roadmap that will push data centres to improve energy efficiency while switching to greener energy sources. Regional governments in Thailand and Malaysia are similarly pivoting towards investments in green data centres. Indeed, as we move forward, the key will be further research on how we can leverage AI to drive economic growth, while ensuring that its development does not undermine sustainability goals.
