Toward Sustainable Generative AI: A Scoping Review of Carbon Footprint and Environmental Impacts Across Training and Inference Stages

Generative AI is spreading rapidly, creating significant social and economic value while also raising concerns about its high energy use and environmental sustainability. While prior studies have predominantly focused on the energy-intensive nature of the training phase, the cumulative environmental footprint generated during large-scale service operations, particularly in the inference phase, has received comparatively less attention. To bridge this gap this study conducts a scoping review of methodologies and research trends in AI carbon footprint assessment. We analyze the classification and standardization status of existing AI carbon measurement tools and methodologies, and comparatively examine the environmental impacts arising from both training and inference stages. In addition, we identify how multidimensional factors such as model size, prompt complexity, serving environments, and system boundary definitions shape the resulting carbon footprint. Our review reveals critical limitations in current AI carbon accounting practices, including methodological inconsistencies, technology-specific biases, and insufficient attention to end-to-end system perspectives. Building on these insights, we propose future research and governance directions: (1) establishing standardized and transparent universal measurement protocols, (2) designing dynamic evaluation frameworks that incorporate user behavior, (3) developing life-cycle monitoring systems that encompass embodied emissions, and (4) advancing multidimensional sustainability assessment framework that balance model performance with environmental efficiency. This paper provides a foundation for interdisciplinary dialogue aimed at building a sustainable AI ecosystem and offers a baseline guideline for researchers seeking to understand the environmental implications of AI across technical, social, and operational dimensions.

💡 Research Summary

This paper presents a scoping review addressing the growing environmental concerns, specifically the carbon footprint, associated with the rapid proliferation of generative Artificial Intelligence (AI). It identifies a critical gap in existing research: while the energy-intensive nature of the AI model training phase has been extensively studied, the cumulative environmental impact of the inference phase—the stage where trained models are deployed and used at scale—has received comparatively less attention. The review synthesizes methodologies and research trends in AI carbon footprint assessment, analyzing the classification, standardization status, and limitations of current measurement tools and approaches.

The analysis begins by framing environmental impact assessment along two axes: carbon footprint analysis, which focuses specifically on greenhouse gas emissions, and broader Life Cycle Assessment (LCA), which evaluates a wider range of environmental indicators. For AI systems, emissions are categorized either by responsibility (Scope 1, 2, and 3) or by timing (embodied emissions from hardware manufacturing vs. operational emissions from service use). The paper notes that operational emissions are often dominant for AI.

A core technical focus is the methodology for calculating the operational carbon footprint, typically derived as Energy Consumed × Carbon Intensity of Electricity. Energy consumption is approximated as Power × Runtime × PUE (Power Usage Effectiveness). The paper highlights significant methodological inconsistencies and sources of uncertainty at each step. Measuring “Power” is complex, varying based on whether only GPU consumption is counted or if CPU, memory, and networking are included, and on the measurement layer (software tools like NVML vs. hardware-level meters). System boundary definitions profoundly affect training phase estimates (e.g., inclusion of hyperparameter search, failed runs). For inference, factors like batch size, queries per second (QPS), and latency targets drastically alter the per-request carbon cost.

The review identifies several critical challenges in current AI carbon accounting practices: a lack of standardized methodologies, technology-specific biases, insufficient transparency (often due to non-disclosure of internal data by companies), and a failure to adopt an end-to-end system perspective that connects training and inference impacts. The variation in regional grid carbon intensity further complicates comparisons.

Based on these insights, the paper proposes four key directions for future research and governance to foster a sustainable AI ecosystem: (1) Establishing standardized, transparent, and universal measurement protocols to ensure comparability and accountability. (2) Designing dynamic evaluation frameworks that incorporate real-world user behavior and usage patterns. (3) Developing comprehensive life-cycle monitoring systems that account for both operational and embodied emissions across the entire AI system lifespan. (4) Advancing multidimensional sustainability assessment frameworks that balance model performance with environmental efficiency, moving beyond a singular focus on carbon.

In conclusion, this work provides a foundational analysis and a set of guidelines for researchers, practitioners, and policymakers seeking to understand and mitigate the environmental implications of AI across technical, social, and operational dimensions, advocating for a holistic approach to sustainable AI development.