As we rush toward an AI-powered future, the technology's massive appetite for water, energy, and resources threatens the very planet we're trying to optimize.

The promise of Artificial Intelligence (AI) seems limitless, revolutionizing medicine, solving complex climate models, and optimizing energy grids. Yet behind every Large Language Model query, every AI-generated image, and every automated decision lies an uncomfortable truth: AI is becoming one of the fastest-growing threats to our environment.

Training a single large AI model emits 626,155 pounds of Carbon dioxide¹, equivalent to 125 round-trip flights between New York and Beijing. The study, led by Emma Strubell, found that the computational and environmental costs of deep learning models have been dramatically underestimated. Training large language models, for example, GPT-3, produced 552 metric tons of Carbon dioxide².

But training is only the beginning. The International Energy Agency projects that by 2026, data centers could consume 1,000 terawatt-hours annually, roughly equivalent to Argentina's entire electricity consumption. What makes this particularly alarming is that despite tech companies' renewable energy pledges, the Uptime Institute's 2023 Global Data Center Survey found that only 21% of data center energy comes from renewable sources.

The computational power needed for state-of-the-art AI doubles every 3.4 months, according to OpenAI's analysis, far outpacing any efficiency improvements in hardware or software. The computational power needed for state-of-the-art AI doubles every 3.4 months, according to OpenAI's analysis, far outpacing any efficiency improvements in hardware or software³_.

Many assume that once an AI model is trained, its environmental cost is settled. However, the reality is a cycle of recurring impact. Beyond the continuous fine-tuning of these models, the daily "inference", including every Google-style query, every AI-generated image, and every video render, demands a massive operational toll⁴.

A single AI query uses 10 times more electricity than a Google search³ and "drinks" up to 50ml of water⁵ for cooling and power. Generating just one AI video can consume 4 liters of water⁶.

This digital convenience relies on a physical world of rapidly proliferating data centers and the carbon-intensive manufacturing of specialized hardware⁷. Carbon emissions are driven by the fossil fuels burned to keep these servers running 24/7⁸. Water is depleted at an alarming rate, both through direct server cooling and the indirect requirements of electricity generation^9,10.

Finally, the "cloud" begins on the ground: biodiversity is sacrificed through the extraction of rare earth minerals, intensive land use for facilities, and the resulting habitat disruption^11,12,13. Our digital requests are, quite literally, reshaping the physical environment.

The Path Forward

This isn't a call to abandon AI, but to radically reimagine it. The European Union's AI Act is beginning to consider environmental impacts. Researchers are developing more efficient algorithms, with techniques like model pruning showing 90% reductions in computational needs without significant performance loss.¹⁴

We need mandatory environmental impact assessments for AI models⁸, water recycling requirements for data centers¹⁰, and carbon pricing for computational resources. Most importantly, we must question whether every problem needs an AI solution.

The future doesn't need faster AI. It needs sustainable intelligence, artificial or otherwise.

References:

1. Strubell, E., Ganesh, A. & McCallum, A. Energy and policy considerations for deep learning in NLP. Proc. 57th Annu. Meet. Assoc. Comput. Linguist. 3645–3650 (2019).

2. Patterson, D. et al. Carbon emissions and large neural network training. Preprint at https://arxiv.org/abs/2104.10350 (2021).

3.Electricity 2024: Analysis and forecast to 2026. (International Energy Agency, 2024).

4. Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of Bloom, a 176B parameter language model. J. Mach. Learn. Res. 24, 1–15 (2023).

5. Li, P., Yang, J., Islam, M. A. & Ren, S. Making AI less ‘thirsty’. Commun. ACM 68, 54–61 (2025).

6. Luccioni, A. S., Jernite, Y. & Strubell, E. Power hungry processing: Watts driving the cost of AI deployment? Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency 1, 85–99 (2024).

7. Gupta, U. et al. Chasing carbon: The elusive environmental footprint of computing. IEEE International Symposium on High-Performance Computer Architecture (HPCA) 27, 854–867 (2021).

8. Wu, Y., Hua, I. & Ding, Y. Unveiling environmental impacts of large language model serving: A functional unit view. Preprint at https://arxiv.org/abs/2502.11256 (2025).

9. Siddik, M. A. B., Shehabi, A. & Marston, L. The environmental footprint of data centers in the United States. Environ. Res. Lett. 16, 064017 (2021).

10. Wu, Y., Hua, I. & Ding, Y. Not all water consumption is equal: A water stress weighted metric for sustainable computing. ACM SIGENERGY Energy Inform. Rev. 5, 84–90 (2025).

11. Crawford, K. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. (Yale University Press, 2021).

12. Valero, A., Valero, A. & Calvo, G. Material bottlenecks in the future development of green technologies. Renew. Sust. Energ. Rev. 93, 178–200 (2018).

13. Shi, T., Kumar, R., Hua, I. & Ding, Y. When servers meet species: A fab-to-grave lens on computing's biodiversity impact. ACM SIGENERGY Energy Inform. Rev. 5, 34–40 (2025).

14. Paula, E., Soni, J., Upadhyay, H. & Lagos, L. Comparative analysis of model compression techniques for achieving carbon efficient AI. Sci. Rep. 15, 23461 (2025).