Artificial intelligence and energy demand: a geographical perspective

Artificial intelligence is closely linked to energy. Its fundamental infrastructures, data centers, require large amounts of energy. This strong dependence on geographically concentrated energy sources makes AI a territorially selective and, at the same time, ambivalent technology: it can contribute to energy efficiency and ecological transition, but it can also generate new socio-environmental imbalances. The growing global energy demand from data centers, which, according to the International Energy Agency, could more than double by 2030, significantly impacts the spatial organization of digital infrastructures.
The location criteria of data centers vary depending on the function they serve within artificial intelligence systems, distinguishing two main phases: training and inference. The training phase requires high energy consumption but does not impose strict constraints on network latency. As a result, data centers dedicated to this phase tend to be located in peripheral areas. In the inference phase, which involves real-time interaction between artificial intelligence systems and users, priority is given to network speed and reliability, factors that drive the placement of data centers closer to major urban centers. These choices perpetuate environmental and energy injustices, where some territories specialize in attracting large digital infrastructures, often with high environmental impacts, while bearing the negative consequences. The presence of hyperscale data centers can indeed trigger conflicts over the use of energy and environmental resources, highlighting the increasingly central role of territory in the dynamics of the digital economy.

Domenico de Vincenzo, Intelligenza artificiale e domanda di energia: una prospettiva geografica in "RIVISTA GEOGRAFICA ITALIANA" 1/2026, pp 111-129, DOI: 10.3280/rgioa1-2026oa22422