AI’s insatiable appetite for energy

Generative AI models have been hailed as a breakthrough in computing, promising efficiencies across a roster of industries. But these efficiencies will come at a cost – AI models require vast datacentres churning billions of data sets to train and run their computations. That consumes energy and water, which is required to cool the servers. How will these needs be met, without disadvantaging others or putting the stability of energy supplies at risk?

Trying to predict how much energy will be needed is difficult as there is uncertainty about how quickly AI will be adopted, and how efficient it will become. But accurate forecasting is crucial for energy utilities, which must consider how rising demand may fundamentally reshape grid infrastructure requirements. Datacentre energy consumption may also impact their capital allocation strategies.

Another challenge for planners is that AI innovation cycles typically span two to three years, while power assets last 20 to 40 years. This raises the risk of asset stranding or financial underperformance, as efficiency gains may make infrastructure surplus to requirements at times. It also adds complexity to electricity system planning and renewable energy integration strategies. The vast quantity of water required by datacentres to cool their servers and storage units is another critical concern, especially in areas of water scarcity.

As AI and data infrastructure scale rapidly, their environmental and social impacts, particularly around energy, water, and land use, are becoming increasingly material to long-term investment outcomes. These risks can affect asset performance through regulation, resource scarcity, and reputational exposure. Engaging with these issues helps investors anticipate disruption, protect value, and support more resilient systems that benefit investment portfolios and the communities they touch.

To find out more, read the full article in EOS’s Q3 2025 Public Engagement Report.