Weekly headlines on AI data centres herald soaring energy footprints, eye-catching contracts for low-carbon power, alarm over electricity network congestion, and backsliding of tech companies’ net-zero commitments. Amid this emphasis on the direct impact of AI on energy and greenhouse gas (GHG) emissions, the implications of how and for what AI is used are less discussed—at least in relation to energy and climate.
A simple taxonomy of AI’s impacts on energy distinguishes direct, indirect, and systemic impacts.1 Direct impacts are from the energy consumed by AI infrastructure like data centres. Indirect impacts are from the energy consumed or saved by the AI applications that this infrastructure makes possible. As AI is a general-purpose technology, these indirect impacts occur throughout almost all forms of economic and social activity. Systemic impacts are harder to isolate and quantify but include the implications for energy demand of the structural changes wrought by AI on economic systems (e.g. from industrial to service economies) and on social systems (e.g. from physical to virtual modes of interaction).
This article sets current debates around AI’s direct energy impact in the context of evidence on its indirect impacts. It also briefly discusses the challenges and opportunities for AI governance to mitigate environmental risks. It argues that the direct energy impact of AI is problematic locally rather than globally, and that the indirect energy impact of AI is larger, more uncertain, more diffuse, and harder to regulate—and so of greater concern.