Sustainability and Artificial Intelligence: How using AI affects climate change

16-Nov-2026

Sustainability and Artificial Intelligence: How using AI affects climate change

 

by Ricardo Ayvar

 

‘And when shall we meet with everyone else, at the edge of an eternal morning, all of us breakfasted?’

- César Vallejo, La cena miserable

 

A. INTRODUCTION

Artificial intelligence (AI) has become a useful tool for people, influencing many sectors of their activities. AI systems are helping to solve complex problems around the world, bringing efficiency, innovation, and economic growth.

However, these benefits imply environmental costs, including substantial energy consumption, material resource extraction, and water withdrawal, which contribute to climate change and ecological degradation. The growth of AI systems, particularly large-scale models like the GPT series, has intensified those impacts, and their computational demands are huge, raising questions about the sustainability of their development.

This essay examines the environmental impacts of AI through three aspects: energy use, material consumption, and water usage. It also evaluates existing regulatory frameworks, proposing practical strategies for mitigating these effects.

 

B. THE MASSIVE USE OF AI

The adoption of AI has increased in recent years. The popularity of applications like ChatGPT—a dialogue-based AI chatbot—shows the rapid expansion of AI tools into mainstream use. After its launch in December 2022, ChatGPT reached one million users within just five days, an indication of the growing demand for such technologies.

In fact, in 2022, approximately 50% of global corporations used AI tools for at least one function in their operations, including human resources, marketing, risk management, and supply chain. Those aspects could be understandable as part of an efficient way to manage a business. Nevertheless, these benefits could change to issues due to the use of AI tools, like ChatGPT.

A survey conducted in 2024 revealed that India has the highest rates of ChatGPT use, with 45% of its population engaging with this tool. Other countries, including Morocco, the UAE, Argentina, and Brazil, report significant adoption rates. These statistics indicate that AI is being used not only for academic and professional purposes but also for entertainment, highlighting a potential issue of its unproductive use.

Moreover, AI’s popularity and increasing number of users have contributed to a rise in AI’s global traffic. According to Similar Web, in September 2024, ChatGPT experienced over 3.1 billion visits, emphasising its reach and influence.

In this regard, AI is increasing quickly, transforming multiple sectors, whose environmental consequences and costs remain under-explored.

 

C. ENVIRONMENTAL RESPONSIBILITY

The environmental impact of AI is triggered by its massive energy consumption. While other technologies like Google require substantial energy—approximately 0.0003 kWh per search based on over 3.5 billion searches each day that represent around 1.05 GWh— AI tools like ChatGPT consume significantly more electricity: each prompt processed by ChatGPT consumes around 2.9 Wh, which equates to approximately ten times the energy required for a Google search.

Considering the exponential increase in AI usage, the total energy consumption associated with AI tools has reached massive levels. For example, ChatGPT alone is estimated to consume 226.8 GWh of electricity annually to process over 78 billion prompts. This energy consumption is equivalent to charging 3.13 million electric vehicles, or powering over 100,810 United States homes for an entire year.

AI’s environmental impact extends beyond AI’s training process, which requires significant energy. For instance, training the GPT-3 model, which features 175 billion parameters, consumed 1.3 GWh of electricity in just 34 days. GPT-4 reportedly consumed almost 62.3 GWh of electricity over 100 days. These figures underscore the growing energy demands associated with advancing AI models.

However, energy is not the only issue related to AI. It is well known that AI’s application is allocated in data centres, which include big hardware devices around the world. In this regard, according to Justice et Paix (2019), manufacturing a 2 kg computer involves the extraction of 800 kg of raw materials and, generally, manufacturing electronic devices requires 50 to 350 times their final material weight.

In addition, data centres require water for their construction and operation. Globally, projected water withdrawal associated with AI in 2027 exceeds the equivalent of 6 times the total annual water withdrawal of Denmark. This is an important issue considering water scarcity: according to the World Health Organization and UNICEF, and Kashiwase and Fujs, approximately 2 billion people lacked access to safely managed drinking water services in 2022 and 2023.

It should be noted that world demand for data centres capacity, promoted by AI usage, is anticipated to rise extremely: by 2030, demand is expected to grow at a rate of 22% annually, reaching an annual demand up to 219 GWh.

 

D. STATUS OF LEGISLATION ABOUT AI AND SUSTAINABILITY

Currently, there is no comprehensive regulatory framework specifically addressing the environmental impacts of AI usage. Nevertheless, organisations such as UNESCO have issued non-binding recommendations on the ethics of artificial intelligence –adopted by 190 countries– focusing on the social impacts of AI, the role of AI systems in society, and the relationship with ecosystems.

Furthermore, the European Union issued a Regulation with harmonised rules on artificial intelligence, to bring environmental protection in a responsible manner according to the Charter of Fundamental Rights of the European Union. The Regulation also recognises environmental sustainability as an important consideration in AI governance, including through provisions encouraging voluntary codes of conduct addressing the environmental sustainability of AI systems.

On the other hand, in the United States, the proposed Artificial Intelligence Environmental Impacts Act of 2024 called for a study on AI’s environmental impact, based on the rapid growth in data centre infrastructure; the impact of AI on pollution, water consumption, and land-use; as well as the increases in electronic waste ('e-waste').

 

E. HOW COULD WE MITIGATE THESE RESULTS?

Considering the substantial environmental impact of AI, it would be necessary to implement mandatory regulation to reduce AI's ecological footprint, avoiding any inefficient use of AI, and focusing on the most valuable aspects of this technology. This regulation should include approaches based on behavioural economics, training and workshops, compliance programmes and economic incentives.

 

(1) Behavioural economics: Incentivising responsible AI usage

One effective way to reduce AI's environmental impact is encouraging responsible usage by people. In this regard, behavioural economics could contribute to this objective, understanding how people make decisions. The first possible alternative is to track user behaviour in order to identify users who consistently submit unproductive, redundant, or wasteful AI queries.

For instance, those who generate a high volume of trivial questions could receive reminders about the environmental impact of their actions. Such reminders would emphasise the need to minimise unnecessary queries and encourage users to be aware of the energy consumption related to their AI usage. Additionally, AI companies could develop behavioural nudges to motivate users to adopt more sustainable behaviours.

Even when defining a 'worthless question' could be difficult, the criteria applied to determine what an efficient prompt is should be designed by the AI company and included in the AI model. These criteria should be based on the following aspects:

  • Contribution to academic research or knowledge.
  • Value in terms of solving real-world problems.
  • Complex design of prompts with specific structure, ensuring valuable answers.
  • Avoiding common and simple queries that could easily be answered by search engines.
  • Discouraging the use of AI for entertainment purposes, such as jokes or memes.

These behavioural tools pursue influencing the user’s decision-making, to change their usual AI usage, for a specific and efficient way. In the education sector, these tools have been used to simplify access to AI and provide positive reinforcement.

However, these mechanisms should be implemented with appropriate safeguards to protect users' privacy and ensure that the criteria used to assess responsible AI usage remain transparent and proportionate.

Applying these criteria into AI models could help to promote a more responsible user culture, mitigating the environmental impact of AI usage.

 

(2) Workshops and training: Promoting efficient AI usage

Another action could be the implementation of workshops and training programmes where users learn how to design efficient prompts. Many people who use AI do not fully understand how to structure their queries to obtain the most accurate outputs. Often, simple prompts trigger redundant queries, causing higher energy consumption.

For this reason, AI companies should offer courses focused on efficient prompt design, ensuring that users know the ideal structure for AI systems in a way that minimises energy waste. These courses could be beneficial for businesses and organisations that operate with AI tools, and should emphasise sustainability behaviour, helping users to understand the broader environmental context and the importance of reducing unnecessary queries.

Creating good prompts is fundamental for achieving precise and useful AI content. In this regard, a well-drafted prompt can reduce ambiguity, enhance specificity, and ensure that AI's responses align with the user's goals. According to Hatton (2024), it is common that beginners make vague prompts, spending much time in the tool. For that reason, Hatton proposes a formula for an optimal prompt:

 

Perfect prompt = Context + Specific Information + Intent + Response Format

Where:

  • Context: Providing sufficient background to the AI.
  • Specific Information: Clearly defining what is being asked.
  • Intent: Stating the purpose behind the inquiry.
  • Response Format: Indicating the desired format of the answer.

Based on this approach, users can use specific prompts, reducing the number of queries needed to obtain satisfactory responses.

 

(3) Compliance Programmes

Organisations and businesses that incorporate AI into their operations can play a significant role in reducing the environmental impact of AI usage. One potential solution is to implement compliance programmes that reward employees for responsible AI usage. Companies could create a ranking system that identifies those who consistently use AI efficiently and responsibly, considering factors such as prompt quality, query relevance, among others.

Employees who demonstrate responsible and efficient AI usage could be rewarded with promotions, raises, or other incentives, supporting a culture of environmental sustainability.

In this respect, several International Organization for Standardization (ISO) standards can support responsible organisational practices, including:

  • ISO 14001: Focuses on environmental management systems, to minimise the environmental impact.
  • ISO 26000: Focuses on social responsibility, which is important for ethical AI development.
  • ISO/IEC 27001: Focuses on information security management, helping organisations ensure that information is handled securely.

 

(4) 'Green users': incentivising efficient usage

Considering the different types of users (including the premium users), AI companies could reduce the price of subscriptions for people who use AI in an efficient way, considering the reduction of environmental footprint and energy consumption. This should be applied by AI companies, requiring an assessment of the most important aspects listed above to identify who uses AI in an unnecessary way.

In this regard, the 'green user' category should include the warnings users have had in their records, as part of the behavioural tools. This category will be analysed each month at least, to identify any distortion in the AI usage that could remove those benefits.

The 'green user' category will be given as an incentive to people who use AI in a responsible manner, pursue the benefits of this tool while considering sustainable AI usage. An example of these incentives would be giving discounts on AI subscriptions. These kinds of economic incentives are effective for average people and are used by governments in tax payment and other policies.

All these tools will help to improve the sustainability levels and be aware about the risks of excessive and unreasonable use of the AI.

 

F. CONCLUSIONS

In summary, AI usage is increasing rapidly due to people’s common use of technological tools. This growth represents a huge impact on the environment, by the energy consumption, the requirements of data centres, and water withdrawal for their operation.

Bearing in mind that there is no particular regulation on these matters, I propose a mandatory regulation for AI development as part of public policy in order to mitigate the impact of AI usage on the environment. This regulation would be applied for each country in Latin America, but on a regional scope in Europe.

I propose focusing this regulation on four aspects: (i) behavioural economics, (ii) training and workshops, (iii) compliance programmes, and (iv) economic incentives.

All these measures would have a positive impact on sustainability and the balance with AI usage.

 

ABOUT THE AUTHOR

Ricardo Ayvar is an LL.M. candidate in Competition Law at University College London (UCL).