11. Outputs and Interpretation
Primary outputs, intended use, and interpretation guidance.
11.1 Primary Outputs
The methodology produces the following outputs for each supported model and billing configuration:
- Energy per token, expressed in kWh. The estimated electricity consumption per token processed as input or generated as output during inference.
- Carbon per token, expressed in gCO₂e. The estimated greenhouse gas emissions per token, comprising operational emissions derived from energy consumption and regional carbon intensity, and amortised embodied emissions of server hardware.
- Water per token, expressed in litres. The estimated water consumed in cooling the inference hardware and in generating the electricity drawn by inference, reported as IT cooling water consumption and electricity generation water consumption.
11.2 Intended Use
These outputs are intended to support organisational sustainability reporting, AI model selection decisions, deployment configuration optimisation, and comparative analysis of environmental impact across models and cloud providers.
11.3 Interpretation Guidance
Outputs are estimates derived from benchmarked and modelled data, reflecting typical energy consumption under the workload conditions and hardware configurations tested. Actual consumption in production may vary due to data centre utilisation, cloud provider optimisations, workload characteristics, and real-time grid conditions.
Per-token values should be interpreted as indicative estimates suitable for planning, reporting, and comparison. They are not direct measurements of a specific customer's inference workload.