Inference
Inference (AI)
Use phase of an AI model, when it answers requests, as opposed to training. It now dominates consumption.
What is Inference (AI) (Inference)?
Inference means running an already trained model to produce a response. Distinct from training, the creation of the model, a one-off but highly intensive event. In 2026, inference represents most of the energy consumed over a model’s life (around 60 to 90%), because it repeats with every request and its volume is exploding. The real environmental issue of AI is therefore aggregate use, rather than training.
Frequently asked questions
What is Inference (AI) (Inference)?
Use phase of an AI model, when it answers requests, as opposed to training. It now dominates consumption.
Why does Inference matter for carbon accounting?
Inference means running an already trained model to produce a response. Distinct from training, the creation of the model, a one-off but highly intensive event. In 2026, inference represents most of the energy consumed over a model’s life (around 60 to 90%), because it repeats with every request and its volume is exploding. The real environmental issue of AI is therefore aggregate use, rather than training.
Have a question about how this applies to your business? A senior consultant will answer.

