File: 11 October 2026 · Source date: 11 October 2026
Electricity is not consumed only in the processor
Artificial intelligence workloads use accelerator chips as well as cooling, data transfer, storage, and backup power systems. Therefore, measuring only the training time of a model does not explain the total effect of the data center.
The distinction between education and use.
A large training run can be intense but limited in duration; the daily demands of millions of users are a constant burden. Even if efficient hardware reduces consumption per transaction, total demand may increase if total utilization grows rapidly.
Measurement at system level
The IEA reviews data center demand along with network, generation source, and regional capacity. The same calculation may have different emission and infrastructure effects depending on where and when the electricity is generated. The workload, hardware, power usage efficiency and electrical mix must be explained for the final result.
Related reading
discussion question
Should artificial intelligence services show energy consumption to the user per transaction?
Share your answer
You must log in to reply.
No response to this article yet.
Established science, active research, and philosophical speculation are clearly separated.