Artificial Intelligence's Impact on the Environment: Latest Research

As demand for increasingly powerful artificial intelligence (AI) models surges, so does the need for energy, water, and hardware. This has resulted in a dual narrative: while AI can significantly contribute to solving climate challenges, it also consumes vast resources that may worsen them if left unchecked. Understanding and addressing this balance is now an urgent global concern.

Image Credit: earthphotostock/Shutterstock.com

The Carbon Cost of Generative AI

The rapid advancement of generative AI has raised substantial concerns regarding its environmental impact, particularly in terms of energy consumption and carbon emissions.

According to a report in arXiv, the training of GPT-3 consumed approximately 700,000 liters of freshwater and emitted 552 metric tons of CO₂-equivalent, comparable to the emissions from over 500 round-trip transatlantic flights per passenger.¹

Another study published in Procedia CIRP conducted a life-cycle assessment of image-generation models such as Stable Diffusion and found that operational emissions over one year can exceed the embodied emissions of the underlying hardware.²

While model training is energy-intensive, recent studies suggest that most emissions occur during inference, with a contribution of between 65–90% of a model’s total life-cycle carbon emissions.³

This is due to the deployment of these models across millions of devices and platforms, where repeated queries such as generating text, images, or search results, accumulate massive compute demand. For instance, large language models embedded in virtual assistants or customer support tools may process billions of prompts each day, each leading to small but significant energy costs.

The increasing demand for large-scale inference has intensified pressure on data centers, which require significant energy for both computation and cooling. Moreover, water usage for AI model cooling is projected to scale dramatically, potentially reaching the consumption levels of medium-sized urban areas unless mitigation strategies are implemented.³

AI’s Expanding Environmental Footprint

Beyond generative tools, AI is also used in finance, targeted advertising, and surveillance. These applications rely on continual, large-scale computation, yet often go unmeasured in energy audits.

As models become larger and cloud-based systems proliferate, infrastructure demands intensify. Many of these server farms are still powered by non-renewable energy sources, particularly in regions lacking clean grid infrastructure, which increases carbon emissions.

Another concern is the environmental cost of AI hardware. The production of such hardware necessitates extensive mining of rare earth elements, often under ecologically destructive and geopolitically sensitive conditions.⁴

These operations can cause deforestation, soil degradation, water contamination, and the displacement of local communities. Moreover, the short lifecycle of AI accelerators, driven by rapid model scaling, contributes to e-waste and resource depletion. These issues illustrate how AI’s environmental impact extends well beyond electricity use, affecting land, water, and mining systems.⁵

Green AI: Solutions Born from the Problem

AI itself offers tools to mitigate environmental challenges. Google’s DeepMind’s wind energy project, launched in 2023, uses AI to forecast wind supply more accurately, improving renewable grid reliability.⁶

AI-driven satellite imagery analysis now detects illegal deforestation in real-time, aiding enforcement efforts across the Amazon and Congo Basin.

A recent study in Environmental Science & Technology also highlights AI’s role in carbon accounting and methane leak detection using hyperspectral imaging and drone-based analytics.⁷ On the efficiency front, models such as TinyML and edge AI minimize compute demands by operating on local, low-power hardware. These innovations exemplify how AI, if responsibly designed, can become a tool for sustainability rather than a threat.⁸

Policy and Industry Response to AI

Several major tech firms have begun to act in response to environmental concerns. In 2024, Google and Meta pledged to achieve net-zero emissions across their AI operations by 2030, with intermediate transparency goals published annually.^9,10 Some companies now provide carbon labels for AI models, detailing energy use during training and inference to inform users and developers.

On the policy front, 2024-2025 saw the release of multiple white papers, including the European Commission’s ‘Sustainable AI Framework’, calling for standardized environmental disclosures and responsible computing practices.¹¹ These include carbon-aware scheduling, energy audits, and stricter life cycle assessments of AI hardware. Such guidelines aim to align innovation with climate obligations.

Rethinking AI Development for a Greener Future

Researchers are increasingly advocating for Green AI, which prioritizes environmental efficiency and model performance. Academic groups are now publishing carbon footprints associated with their models and actively developing computationally efficient architectures. Innovations such as sparse transformers, quantization techniques, and low-rank adaptation are gaining momentum for their significantly lower computational and energy demands.⁵

In parallel, carbon-aware scheduling is being adopted, aligning training workloads with periods of high renewable energy availability to minimize carbon intensity. Cloud service providers are also exploring geographic scheduling to optimize for cleaner grids. Moreover, regulatory bodies, funding agencies, and scientific publishers are beginning to mandate sustainability disclosures for large-scale AI projects. These include detailed reporting on energy consumption, hardware use, and lifecycle emissions.

Download your PDF copy now!

Collectively, these developments mark a growing consensus within the research and policy communities: the future of AI must be environmentally accountable by design and practice, ensuring that innovation proceeds in alignment with global climate objectives.

Can AI be Part of the Climate Solution?

AI raises a complex dilemma in environmental discourse. Its growing development and deployment can strain ecosystems, yet it also holds significant potential to support climate solutions.

The fundamental question is not whether AI will be used but how and to what ends. A firm commitment to sustainability and ethical responsibility must match this promise.

As reliance on these systems expands, transparency, energy awareness, and environmental safeguards must become foundational principles. The aim is not to resist innovation, but to ensure its trajectory aligns with the ecological boundaries we cannot afford to ignore. Achieving this will require collaboration between scientists, lawmakers, and technology leaders.

Long-term success of AI depends on current efforts to build systems that account for both their capabilities and consequences. Technology can only play a constructive role in addressing environmental pressures when developed with care and accountability.

References and Further Reading

1. Li P., Yang J., Islam M. A., & Ren S. (2023). Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models. arXiv preprint. https://doi.org/10.48550/arXiv.2304.03271
2. Berthelot A. et al. (2024). Estimating the Environmental Impact of Generative‑AI Services Using an LCA‑Based Methodology. Procedia CIRP 122, 707‑712. https://doi.org/10.1016/j.procir.2024.01.098
3. Luccioni A. S., Jernite Y., & Strubell E. (2023). Power Hungry Processing: Watts Driving the Cost of AI Deployment? arXiv preprint.  https://doi.org/10.48550/arXiv.2311.16863
4. Bainton, N., & Holcombe, S. (2018). A critical review of the social aspects of mine closure. Resources Policy, 59, 468–478. https://doi.org/10.1016/j.resourpol.2018.08.020
5. Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2020). Green AI. Communications of the ACM, 63(12), 54–63. https://doi.org/10.1145/3643049
6. Google DeepMind. (2019). Machine learning can boost the value of wind energy. DeepMind Blog. https://deepmind.google/discover/blog/machine-learning-can-boost-the-value-of-wind-energy/
7. Chen, Y., Zhang, T., Li, K., & Li, J. (2024). AI-enabled hyperspectral imaging for real-time methane leak detection and carbon accounting. Environmental Science & Technology, 58(25), 10941–10955. https://doi.org/10.1021/acs.est.3c08511
8. Xu, Y., Shen, C., & Annavaram, M. (2022). Edge AI: On-device intelligence for edge computing. ACM Transactions on Computer Systems, 40(4), Article 36
9. Meta. (2023). Mapping our path to net zero. Meta Sustainability Blog. https://sustainability.atmeta.com/blog/2023/07/25/mapping-our-path-to-net-zero/
10. Google. (2024). Accelerating sustainability in AI: Our commitment to net-zero by 2030. In Google Sustainability Reports. https://sustainability.google/operating-sustainably/net-zero-carbon/
11. European Commission. (2025). Sustainable AI Framework: Towards environmentally responsible artificial intelligence. Directorate‑General for Communications Networks, Content and Technology. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.