Sustainability & AI
Model training
The compute-intensive process of fitting an AI model to data; the stage where much of an AI system's energy use is concentrated.
Definition
The process of fitting a model's parameters to data by optimising an objective — the phase of the AI life cycle where computation, and thus energy demand, is concentrated.
Quick reference
At a glance
- Subject
- Sustainability & AI
- Editorial status
- Editorial draft
- Definition status
- Established
- Last updated
- 21 August 2026
- Also known as
- training · model fitting
References
This source provides part of the technical or institutional basis for the definition.
This source supports the explanation of how the term is applied, measured or governed in practice.
This source supports the wider sustainability significance and context described in the entry.
Overview
What it means
Training runs many passes over data to adjust millions or billions of weights, using specialised accelerator hardware. Landmark analyses (Strubell et al. 2019; Patterson et al. 2021) showed training emissions vary enormously with model scale, hardware and grid carbon intensity — and that design choices can cut them dramatically.
How it is used
For sustainability assessment, training is the first accounting boundary: the IEA and researchers increasingly treat it separately from inference. The EU AI Act requires general-purpose model providers to document known or estimated training energy consumption.
Why it matters
Training is where AI's footprint is most measurable and most concentrated — and where siting, hardware and timing decisions have the largest marginal effect. It is the natural starting point for any organisation disclosing AI-related emissions.
Review
Help keep this definition useful and accurate.
Submitted reviews stay private until an editor decides whether to accept and attribute them.
Endorse this definition
Confirm what works
Endorse the current wording when it is accurate and useful. Suggested changes use the separate editorial form below.