cs.CYSep 26, 2026

Environmental Impact of Generative and Agentic AI: An in-Depth Analysis and Green Solutions

Authors: Abderaouf Bahi, Amel Ourici, Ibtissem Gasmi

Organizations: Computer Science and Applied Mathematics Laboratory (LIMA), Faculty of Science and Technology, Chadli Bendjedid University, P.O. Box 73, El Tarf 36000, Algeria · Mathematical Modeling and Numerical Simulation Laboratory (LAM2SIN), Faculty of Technology, Badji Mokhtar University, P.O. Box 12, Annaba 23000, Algeria

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

The proliferation of generative and agentic artificial intelligence (AI) systems has introduced computational demands whose environmental consequences are substantial yet underexamined. This paper examines the environmental footprint of modern AI systems across energy consumption, carbon emissions, water usage, and electronic waste over the full lifecycle of large language models, multimodal foundation models, and agentic workflows, from hardware fabrication and training through fine-tuning and inference to end-of-life disposal. This work provides a conceptual analysis, utilizing order-of-magnitude estimations based on published data, without conducting original physical measurements. We contribute a lifecycle taxonomy that crosses lifecycle phases with five impact dimensions and emission scopes; a Sustainability Assessment Framework for AI Systems (SAFIA) comprising nine indicators; a comparative analysis of traditional, generative, and agentic AI; seven open challenges; policy recommendations for regulators, cloud providers, hardware manufacturers, and AI developers; and a research roadmap to 2035. The analysis indicates that inference can rival or exceed training energy over a deployment lifetime, that agentic workflows can multiply the energy of equivalent single-pass inference by one to several orders of magnitude depending on the number of model and tool calls, that indirect water use and embodied carbon are systematically underreported, and that measurement tools, disclosure practices, and regulation have not kept pace with agentic deployment. These findings call for agent-aware energy measurement, lifecycle-based carbon and water accounting, and mandatory disclosure for large-scale AI training and deployment.

Figures & tables

AbbreviationMeaning
AIArtificial Intelligence
APIApplication Programming Interface
AWQActivation-Aware Weight Quantization
CAIConstitutional AI
CO 2 eCarbon Dioxide Equivalent
CPUCentral Processing Unit
TABLE I: List of Acronyms and Abbreviations
Fig. 1: Classification of Generative AI Architectures, Model Families, and Scaling Studies by Category
CategoryCriterion
Inclusion criteria
IC1Peer-reviewed journal articles, preprints, conference proceedings, or technical reports published 2015–2026
IC2Direct measurement, estimation, or analysis of energy consumption, carbon emissions, water usage, or e-waste for AI or ML systems
IC3Studies addressing AI hardware lifecycle, data center infrastructure, or AI-specific computing infrastructure
IC4Review, survey, or meta-analysis papers on green AI, sustainable computing, or AI environmental impact
IC5Policy, regulatory, or governance analyses specifically addressing AI environmental sustainability
TABLE II: Inclusion and Exclusion Criteria
Fig. 2: PRISMA flow of the literature search.
Fig. 3: Classification of Environmental Footprint Studies on AI by Focus Area
Fig. 4: Lifecycle taxonomy of the environmental impacts of AI systems
ModelParametersHardwareEnergy (MWh)CO 2 e (t)
T5-11B11BTPU v38647
GPT-3175BV100 GPUs1,287552 a
BLOOM176BA100 GPUs43324.7 b
Values are published estimates, not measured. Values are grid-dependent and not directly comparable across rows.
TABLE III: Reported Training Energy and Emissions of Representative Models
Fig. 5: Classification of Agentic AI Patterns, Frameworks, and Benchmarks by Category
MethodDescription
FLOP-based estimationEnergy derived from FLOP counts and hardware efficiency data
Power metering (CarbonTracker)Real-time GPU and CPU power measured via NVML and RAPL, combined with grid emission factors
ML CO 2 ImpactWeb calculator using region and cloud provider data
LCA-basedOperational and embodied carbon over the hardware lifecycle
Scope-based accountingCorporate scope accounting applied to computing systems
TABLE IV: Carbon Accounting Methods for AI Systems
PhaseEnvironmental impactKey materials and substances
Raw material extractionHabitat destruction, water pollution, CO 2 emissions from miningCobalt (DRC), lithium, tantalum, rare earths
Semiconductor fabricationWater consumption, chemical waste, GHG emissionsHF, NF 3 , PFCs, ultrapure water
PCB manufacturingToxic chemical use, worker exposureLead, brominated flame retardants, nickel
System assemblyLow impact if performed in regulated facilitiesSolder, adhesives
Operational useElectricity consumption, heat generation, HVAC wearCooling fluids, HFC refrigerants
Decommissioning and end of lifeData destruction and transport energy; hazardous release if improperly handled; resource recovery if formally recycledAu, Ag, Pd, Cu, Pb, Cd, Hg
TABLE V: Environmental Impacts of AI Hardware Across Lifecycle Phases
DimensionTraditional AIGenerative AIAgentic AI
Parameter scaleSmall to mediumLargeLarge, per agent
Training energyLow to moderateVery highVery high
Inference energy per taskLowModerate to highVery high
Agentic multiplier μA\mu_{A}Not applicable1 (single pass)Greater than 1
Carbon per taskLowModerateHigh to very high
Water per taskLowLow to moderateModerate to high
TABLE VI: Qualitative Comparison of Traditional, Generative, and Agentic AI
LevelTechniqueMechanismExpected effect
ModelKnowledge distillationTeacher-student trainingSmaller, faster student model with limited accuracy loss
QuantizationReduced numerical precisionLower memory footprint and higher throughput on compatible hardware
Structured pruningRemoval of attention heads or layersFewer operations per inference with modest accuracy loss
MoE architecturesSparse expert activationLarger capacity at similar per-token compute
PEFTAdapter-only fine-tuningFar fewer trainable parameters
Efficient attentionMemory-efficient or linear attentionLower attention memory and latency
TABLE VII: Green AI Mitigation Techniques Grouped by Intervention Level
Fig. 6: Classification of Model-Level Green AI Techniques by Category
IndicatorDimensionNameUnitMeasurement methodDefault weight
I-1EnergyOperational energy intensityWh per 1,000 output tokensGPU power metering (NVML, RAPL)0.12
I-2EnergyTraining energy budgetMWh per model versionHardware logs and PUE factor0.12
I-3EnergyAgentic energy multiplier μA\mu_{A}DimensionlessInstrumented agent benchmark0.12
I-4CarbonOperational carbon intensitygCO 2 e per 1,000 output tokensI-1 ×\times grid emission factor0.12
I-5CarbonLifecycle carbon ratioEmbodied/operational fractionLCA and operational carbon accounting0.12
I-6CarbonRenewable energy fraction% of operational electricityEnergy attribute certificates0.12
TABLE VIII: SAFIA Sustainability Assessment Indicators for AI Systems
Fig. 7: SAFIA structure: the inputs feed nine indicators grouped by environmental dimension with their default weights, which are normalized and aggregated into the score of Eq. ( 2 ). Indicator I-3 captures the agentic multiplier μA\mu_{A} of Eq. ( 1 ).
Fig. 8: Research roadmap to 2035 in five non-overlapping horizons.
ActorHorizonRecommendationMechanismPriority
Regulators (EU and United States)2026–2027Mandate energy and carbon disclosure for large-scale AI training runs ( >> 10 23 FLOP) and deployments ( >> 10 million queries per day)Binding regulatory requirementHigh
Regulators2028–2029Establish an AI-specific LCA methodology standard aligned with ISO 14040/14044Standards body process (ISO, IEC, NIST)High
Cloud providers2026–2027Provide per-workload Scope 2 energy and carbon reporting via APIs for all AI compute customersVoluntary, then mandatory policyHigh
Cloud providers2028–2029Achieve real-time (hourly) renewable energy matching for AI data centersPPA restructuring and grid investmentMedium
Hardware manufacturers2028–2029Disclose per-chip manufacturing carbon, water, and material inventories via product environmental data sheetsIndustry standardization (e.g., JEDEC)Medium
AI developers2026–2027Report training compute, energy consumption, and carbon emissions in all model publications and documentationCommunity norms and journal requirementsHigh
TABLE IX: Policy Recommendations for AI Environmental Sustainability
More in this topic

Life Cycle Assessment

See all 10 papers

Explore similar work

CardsList
May 31, 2026

Evaluation of ML Resource Utilization Requires Model Life Cycle Assessment

Proper accounting of the energy requirements and environmental impact of artificial intelligence (AI) systems is necessary for researchers, developers, policy makers, and users to assess the barriers to building systems at scale. With the growing complexity of pipelines and underlying infrastructure needed to develop and deploy AI systems, previous approaches for evaluating AI efficiency which focus on the costs of a single training run or an individual inference prediction are no longer sufficient. In this position paper, we enunciate the need for applying life cycle assessment to evaluate the costs of the machine learning model development and deployment pipeline to properly account for the required resources and downstream impact. Life cycle assessments enable the incorporation of costs across the full life cycle of an AI system and its underlying infrastructure, from the embodied costs associated with the physical computing hardware through the operational costs in training and inference.

Jared Fernandez, Clara Na, Yonatan Bisk +2

Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, USA · Department of Civil and Environmental Engineering, Carnegie Mellon University, Pittsburgh, PA, USA · Wilton E. Scott Institute for Energy Innovation, Carnegie Mellon University, Pittsburgh, PA, USA

Energy-Efficient MLResponsible AILife Cycle Assessment
AskPDF
Apr 13, 2026

Environmental Footprint of GenAI Research: Insights from the Moshi Foundation Model

New multi-modal large language models (MLLMs) are continuously being trained and deployed, following rapid development cycles. This generative AI frenzy is driving steady increases in energy consumption, greenhouse gas emissions, and a plethora of other environmental impacts linked to datacenter construction and hardware manufacturing. Mitigating the environmental consequences of GenAI remains challenging due to an overall lack of transparency by the main actors in the field. Even when the environmental impacts of specific models are mentioned, they are typically restricted to the carbon footprint of the final training run, omitting the research and development stages. In this work, we explore the impact of GenAI research through a fine-grained analysis of the compute spent to create Moshi, a 7B-parameter speech-text foundation model for real-time dialogue developed by Kyutai, a leading privately funded open science AI lab. For the first time, our study dives into the anatomy of compute-intensive MLLM research, quantifying the GPU-time invested in specific model components and training phases, as well as early experimental stages, failed training runs, debugging, and ablation studies. Additionally, we assess the environmental impacts of creating Moshi from beginning to end using a life cycle assessment methodology: we quantify energy and water consumption, greenhouse gas emissions, and mineral resource depletion associated with the production and use of datacenter hardware. Our detailed analysis allows us to provide actionable guidelines to reduce compute usage and environmental impacts of MLLM research, paving the way for more sustainable AI research.

Marta López-Rauhut, Loic Landrieu, Mathieu Aubry +1

LIGM, CNRS, Univ Gustave Eiffel, ENPC, Institut Polytechnique de Paris, France · Université Paris-Saclay, CNRS, LISN, Orsay, France

Life Cycle AssessmentResponsible AIEnergy-Efficient ML
AskPDF
Aug 7, 2026

Towards Sustainable Artificial Intelligence: A Comprehensive Review and Comparative Analysis of Deep Learning Models' Carbon Footprint

Artificial Intelligence (AI) and Machine Learning (ML) have become powerful tools for supporting and automating complex human tasks. Despite their benefits, growing attention has been directed toward their environmental implications, primarily due to their high energy demands and associated carbon emissions. This concern is particularly relevant in light of the increasing deployment of large-scale models, especially Deep Learning (DL) architectures, which provide advanced predictive capabilities but require substantial computational resources. This paper presents a systematic review of research on Green AI, Green DL, and optimization techniques aimed at reducing the environmental impact of AI models. In addition, we examine and compare several carbon measurement tools for estimating emissions generated by AI algorithms. To complement the review, we conducted an empirical evaluation using a CPU-based experimental setup, in which six DL models were implemented for a multi-label classification task. The objective was to quantify and compare their overall carbon emissions and to determine which stages of the DL lifecycle contribute most significantly to the total footprint. The results show that the training phase is the primary source of emissions. Moreover, the findings reveal that increased architectural complexity does not systematically translate into proportional accuracy gains, highlighting the importance of carefully balancing predictive performance and environmental cost. These results reinforce the need to integrate sustainability considerations into model selection and AI system design.

Samar Garrab, Sarra Boughriou, Manel BenSassi

Royal Military College of Canada, Ontario, Canada. · Concordia University, Quebec, Canada. · ´Ecole Nationale des Sciences de l’Informatique, Tunisia.

Energy-Efficient ML
AskPDF