The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2630
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T08:06:19.797605Z
status: published
lens: p_space
sector: climate
headline: Green artificial intelligence and its lifecycle, hardware, and measurement dimensions
dek: Context: Across the artificial intelligence (AI) lifecycle—from hardware provisioning and development to deployment, operation, and end-of-life— environmental burdens span energy, CO 2 e 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 persp…
gain_title: (none)
problem_title: A review of 103 studies found AI environmental research centres on training and deployment energy, while water use, Scope 3 value-chain impacts and end-of-life hardware are thinly covered and cloud dashboards cannot be compared.
trace_subject: (none)
gain_reading: (none)
gain_evidence: (none)
problem_reading: A review of 103 studies found AI environmental research centres on training and deployment energy, while water use, Scope 3 value-chain impacts and end-of-life hardware are thinly covered and cloud dashboards cannot be compared.
problem_evidence: Remaining gaps include inconsistent metrics, limited attention to water and Scope 3 impacts, provider-specific dashboards that hinder comparability | training and deployment dominate above baseline, while supply-chain and end-of-life fall well below
quick_read: A systematic review by University of Potsdam researchers synthesised 103 studies on the environmental burden of AI, coding them by lifecycle phase, impact type and system level. They also extracted and harmonised 164 Green AI terms. The review found that research attention is heavily skewed toward training and deployment, while upstream hardware sourcing and end-of-life are far less covered.

The review reports a Gini coefficient of 0.74 for lifecycle terms and an Intervention Bias Ratio of 3.6, which the authors read as a persistent overemphasis on compute-focused fixes. Mean coverage reached up to 39.5% for training and inference phases, against 17.1% for green end-of-life and circularity.

Measurement is a second weak point. Cloud-provider emissions dashboards use differing methods, often leave out water and Scope 3 coverage, and block cross-system comparison. The authors propose a five-phase lifecycle mapped to Life Cycle Assessment stages and a hybrid approach that calibrates estimator tools against direct power metering. They describe these constructs as candidate tools until validated.
limitation: The authors list review coverage bias, discrepancies between estimators and meters, non-comparable provider dashboards and incomplete treatment of water and Scope 3 impacts as limits, and say the proposed constructs remain candidate tools until validated.
tag: Evidence-backed problem
key_points: The review screened 1,695 records, narrowed them to 757 for title and abstract screening and 98 for full-text assessment, and included 103 articles after citation chaining added 7. | Yearly counts of energy- and measurement-focused Green AI articles rose from 6 in 2019 to 24 in 2024, with 4 counted for 2025 up to the 7 August 2025 cut-off. | Most of the 103 articles fell into lifecycle-explicit work (46, or 44.7%) or definitions (33, or 32.0%); measurement assets accounted for 21 (20.4%) and green hardware for about 13 (12.6%). | Training and deployment phases had the highest mean coverage, up to 39.5%, while circular maintenance reached 22.1% and green end-of-life and circularity only 17.1%. | Items such as emission-factor disclosure, supplier traceability, repair, reuse and recycling appeared late and rarely, mostly in 2024-2025 articles. | The authors define Green AI as a lifecycle-wide effort to minimise energy, emissions, water and material footprints, based on measured or calibrated evidence with disclosed uncertainty. | Priorities include reference workloads, cross-meter calibration protocols, inter-site reproducibility studies and standardised reporting checklists.
rundown: The search began in October 2024 and finished data extraction in August 2025. It covered Web of Science (1,035 records), ScienceDirect (318), IEEE Xplore (326) and 16 other sources, with two independent assessors screening 757 de-duplicated records and excluding 659 at title and abstract stage. Two more were dropped at full text, leaving 96, and 7 added through citation chaining gave 103.

The term analysis showed head terms such as model training, deployment, inference and power consumption dominating. Long-tail topics, including carbon-aware scheduling, power capping, conflict minerals, remanufacturing and e-waste pathways, appeared rarely but with visible growth. Recent articles show broader coverage than pre-2020 work, yet the authors say the gains do not close the upstream and downstream gap.

On measurement, the review argues indirect estimators must be calibrated against direct readings such as RAPL, rack-level meters and facility telemetry. Unresolved choices include boundary conventions, emission-factor provenance and amortisation of embodied impacts, which the authors say make cross-study comparisons fragile.

The authors say hardware strategies such as accelerator choice, placement across the edge-cloud continuum, power capping and carbon-aware scheduling can raise performance per watt. They propose governance through Plan-Do-Check-Act cycles with phase completion criteria and performance-environmental thresholds.
sources:
- peer_reviewed | Information and Software Technology | https://doi.org/10.1016/j.infsof.2026.108186 | 2026-05-24
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

4787177b1ebe08bd57d383ff224c0e8ae2166c034af74113a7315209a1791da9

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000