2.4 Environmental Impact of AI

Environmental Impact of AI

The rapid growth of AI technologies (and cloud-based technologies in general) has sparked a lot of concern around the environmental impact of technological advancement.

There are environmental considerations at each stage of the AI development process:

Hardware: the physical resources required for generative AI hardware and infrastructure involves extensive mining and extraction of minerals, which can lead to deforestation and increased soil and water pollution. The production of hardware, like Graphic Processing Units, can also consumed large amounts of energy and water. (Hosseini et. al., 2025 ). The rapid growth of generative AI technologies will also contribute to the global increase in ewaste, which when not properly disposed of can also contribute to air, water, and soil pollution. A report from 2022 indicates that only 22% of ewaste is properly recycled (Crownhart, 2024).

Training:  Training generative AI models requires significant amounts of energy. For example, it has been estimated that creating GPT-3 resulted in carbon dioxide emissions equivalent to the amount produced by 123 gasoline powered vehicles driven for a year (Saenko, 2023 ).

Usage: using generative AI also has a substantial water footprint and significant carbon emissions. It is estimated that a ChatGPT dialogue with 20-50 prompts uses approximately 500ml of water (McLean, 2023 ). Estimates suggest that by 2027, use of AI technologies globally will account for water withdrawal equivalent to 4-6 times that of Denmark or half that of the United Kingdom (Li, Islam & Ren, 2023). However, in contrast, an analysis conducted in 2024 suggests that the carbon emissions of content creation (text and images) may actually be lower for generative AI produced content than human-produced content (Tomlinson, Black, Patterson, & Torrance, 2024). 

Data centres, though not limited to generative AI technologies, are becoming one of the largest consumers of energy, currently accounting for 3% of global energy consumption (Cohen, 2024 ). They also require large amount of water for cooling. There are more environmentally sustainable approaches to cooling down data centres, but these are substantially more expensive, which could be seen as another example of values-friction (sustainability VS profitability), as discussed in section 3.2, (Ammachchi, 2025 ).

Often, the largest environmental impact of technological development occurs in already disadvantaged communities, perpetuating existing inequities. For example, a study in the US shows that high-pollutant data centres were more likely to be built in racialized communities (Booker, 2025 ).

Many of these environmental concerns are not new, but generative AI has brought a renewed focus on the environmental impact of digital technologies and rapid technological advancement. The environmental impact is a prime example of the complexity of the generative AI conversation as it highlights tensions in values and priorities, social inequities, and the affective nature of these conversations.

See AI’s impact on energy and water usage for a review of recent research on the environment impact of generative AI technologies.