Green artificial intelligence and its lifecycle, hardware, and measurement dimensions
Abstract
Context: Across the artificial intelligence (AI) lifecycle — from hardware provisioning and development to deployment, operation, and end-of-life — environmental burdens span energy, CO2e emissions, water use, and embodied/material footprints. Cloud-provider tools can improve operational visibility but remain heterogeneous and only partially cover water and value-chain effects, limiting comparability and reproducibility. Objective: Addressing these multidimensional burdens benefits from a lifecycle-oriented perspective linking phase-explicit structuring with system levers such as hardware context, placement, energy mix, cooling, and scheduling, together with calibrated measurement across facility, system, device, and workload levels. Methods: This systematic review synthesizes 103 studies and, within the lifecycle-focused analysis, extracts and harmonizes 164 Green AI-related terms. Included studies were coded along three dimensions: lifecycle phase, environmental impact category, and system level. The identified concepts were analytically related to Life Cycle Assessment (LCA) stages to structure the lifecycle perspective. Measurement approaches were compared by distinguishing estimator-based reporting from metered or telemetry-based reference data, including calibration procedures and reporting completeness. Results: The review derives a synthesized definition of Green AI that provides a conceptual reference point for the lifecycle, hardware, and measurement analyses. It synthesizes a five-phase lifecycle structure and analytically relates it to the four established LCA stages. It further synthesizes a class-based hardware perspective across the edge–cloud continuum and a hybrid measurement logic that relates estimator-based approaches to metered or telemetry-based reference data to support calibration and more consistent reporting. Important gaps remain, including inconsistent indicators, limited treatment of water and Scope 3 impacts, provider-specific dashboards that hinder comparability, and the lack of reproducible, calibrated measurements across hardware tiers. Conclusion: By integrating definition, lifecycle structuring, hardware context, and calibrated measurement, this article offers an analytically grounded synthesis of Green AI relevant to researchers and practitioners. It supports more structured discussion of green model and infrastructure choices, context-aware scheduling, transparent reporting, and circularity-oriented future work.
| Kategorie | Journalbeiträge |
| Autoren | Rojahn, Marcel; Grum, Marcus |
| Zeitschrift | Information and Software Technology |
| Datum | 10/2026 |
| Volume | 198 |
| DOI | https://doi.org/10.1016/j.infsof.2026.108186. |
| Keywords | Green artificial intelligence, Life Cycle Assessment, Systematic literature review, Environmental impact measurement, AI hardware |
| ISSN | 0950-5849 |
| BibTex |
@article{ROJAHN2026108186,
title = {Green artificial intelligence and its lifecycle, hardware, and measurement dimensions},
journal = {Information and Software Technology},
volume = {198},
pages = {108186},
year = {2026},
issn = {0950-5849},
doi = {https://doi.org/10.1016/j.infsof.2026.108186},
url = {https://www.sciencedirect.com/science/article/pii/S0950584926001758},
author = {Marcel Rojahn and Marcus Grum},
keywords = {Green artificial intelligence, Life Cycle Assessment, Systematic literature review, Environmental impact measurement, AI hardware},
abstract = {Context:
Across the artificial intelligence (AI) lifecycle — from hardware provisioning and development to deployment, operation, and end-of-life — environmental burdens span energy, CO2e emissions, water use, and embodied/material footprints. Cloud-provider tools can improve operational visibility but remain heterogeneous and only partially cover water and value-chain effects, limiting comparability and reproducibility.
Objective:
Addressing these multidimensional burdens benefits from a lifecycle-oriented perspective linking phase-explicit structuring with system levers such as hardware context, placement, energy mix, cooling, and scheduling, together with calibrated measurement across facility, system, device, and workload levels.
Methods:
This systematic review synthesizes 103 studies and, within the lifecycle-focused analysis, extracts and harmonizes 164 Green AI-related terms. Included studies were coded along three dimensions: lifecycle phase, environmental impact category, and system level. The identified concepts were analytically related to Life Cycle Assessment (LCA) stages to structure the lifecycle perspective. Measurement approaches were compared by distinguishing estimator-based reporting from metered or telemetry-based reference data, including calibration procedures and reporting completeness.
Results:
The review derives a synthesized definition of Green AI that provides a conceptual reference point for the lifecycle, hardware, and measurement analyses. It synthesizes a five-phase lifecycle structure and analytically relates it to the four established LCA stages. It further synthesizes a class-based hardware perspective across the edge–cloud continuum and a hybrid measurement logic that relates estimator-based approaches to metered or telemetry-based reference data to support calibration and more consistent reporting. Important gaps remain, including inconsistent indicators, limited treatment of water and Scope 3 impacts, provider-specific dashboards that hinder comparability, and the lack of reproducible, calibrated measurements across hardware tiers.
Conclusion:
By integrating definition, lifecycle structuring, hardware context, and calibrated measurement, this article offers an analytically grounded synthesis of Green AI relevant to researchers and practitioners. It supports more structured discussion of green model and infrastructure choices, context-aware scheduling, transparent reporting, and circularity-oriented future work.}
}
|