Every question you ask a model costs electricity. The bill goes somewhere. Not to you.
Literal meaning: The total electricity required to build and run AI systems — from training models to answering individual queries.
Origin: The scale of AI’s energy use became publicly legible in 2019, when researcher Emma Strubell and colleagues at the University of Massachusetts measured the carbon cost of training a single large language model — comparable to the lifetime emissions of five cars. Since then, models have grown dramatically larger, and energy has moved from a footnote to a central controversy.
The electricity demand of AI infrastructure — a cost that is real, accelerating, and almost never disclosed at the point of use.
The Appeal: More compute means better models. Better models mean more accurate outputs, faster research, better tools. The energy cost feels like the price of progress — finite, purposeful, worth paying.
The Friction: The cost does not disappear — it transfers. Externalized Costs — environmental burdens placed on communities rather than priced into AI services — mean that the person asking a chatbot a question does not pay the energy bill. A community near a data centre does. Sacrifice Zones — areas bearing disproportionate environmental burdens of digital infrastructure — are often low-income or already environmentally stressed. The abstraction of the cloud makes this transfer invisible by design.
Why This Matters: Once you know this term, a simple question shifts: whose electricity is this? The speed and fluency of an AI response starts to carry weight. Not moral paralysis — but a new kind of legibility.
Related terms: Training Run · Inference Cost · Externalized Costs · Sacrifice Zones · Greenwashing · Data Center Water Consumption · Digital Colonialism
Read more:
- Energy and Policy Considerations for Deep Learning in NLP — Strubell, E., Ganesh, A. & McCallum, A. (2019). ACL
- The growing energy footprint of artificial intelligence — de Vries, A. (2023). Joule