Sustainability, the double face of artificial intelligence

Dozens of data centres buzz around the outskirts of Dublin, the country with the most data centres per inhabitant in the world. In 2024, their powerful computers consumed 21% of the nation's electricity, more than all of Ireland's homes. The beating heart of Amazon, Alphabet, Meta, Microsoft and TikTok in Europe, the home of Guinness is now, however, starting to backtrack: Google's new data centre has not been granted permission and fears of blackouts have prompted the Irish grid operator to block any projects near Dublin until 2028.

Meanwhile, the electricity hunger of data traffic continues to rise: in its recent report 'Energy and Ia', the International Energy Agency predicts that global data centre electricity consumption will double by 2030, reaching 945 terawatt hours from 415 in 2024. The main culprit? Artificial intelligence. Asking ChatGPT a question means consuming ten times more energy than traditional search. With the advent of artificial intelligence, Google's greenhouse gas emissions increased by almost half from 2019 to 2023 and Microsoft's by 30% from 2020. In advanced economies, more than 20 per cent of the growth in electricity demand between now and 2030 is expected to depend on data centres. In the US, data processing in 2030 will consume more electricity than the production of all energy-intensive goods, including aluminium, steel, cement and chemicals.

On the other hand, says the Iea, artificial intelligence has the potential to revolutionise the way the energy sector works, leading to significant savings. Widespread industrial application of artificial intelligence and machine learning could save, according to Iea analysts, as much as 8 exajoules of energy by 2035, the same amount of energy Mexico consumes today. Widespread adoption in non-industrial sectors could save another 5 exajoules. Indeed, the Iea is already helping to improve the efficiency of electricity grids, reduce fuel consumption in shipping, and detect leaks of methane, the most potent of greenhouse gases. The Iea estimates that replacing periodic inspections of methane production and transportation facilities with continuous satellite monitoring enabled by artificial intelligence would prevent 2 million tonnes of methane emissions per year.

On the electricity grid front - a hot front after the Spanish blackout - the American start-up LineVision (backed by Microsoft) uses contactless sensors to monitor temperature, cable failure and environmental conditions on transmission lines. By analysing this data with the help of the Ia and combining it with weather forecasts, its algorithms calculate the actual load capacity of the lines. The Iea report claims that high-voltage transmission lines 'can safely carry additional capacity 20-30% above their maximum limit about 90% of the time'. The UK power grid has used this approach to 'unlock' 600 megawatts of additional offshore wind capacity per year.

Artificial intelligence can also make the use of green energy more profitable and attractive: in 2018, Google DeepMind started using machine learning to manage some of the wind farms from which the company buys energy. By combining weather forecasts and historical data from the turbines, the system was able to predict the parks' energy production up to 36 hours in advance and select how much and when to feed into the grid. Within a year, this increased the value of the energy sold by 20% and made it easier for grid operators to use.