What is the environmental cost of AI?

AI presents a complex paradox in the fight against climate change. On one hand, there are high hopes that AI will enhance how we tackle climate change, help us resolve some of the most challenging aspects and accelerate the transition into renewable energy. On the other, there are concerns among environmental groups and experts that the development, training and use of AI could have detrimental environmental impacts.

While there’s a growing body of evidence to support these concerns, the full extent of AI’s environmental impact is yet to be realised – AI developers offer little-to-no transparency into their internal processes and use of resources.

With many companies pledging to reduce their carbon emissions, the increasing reliance on AI could make these commitments harder to achieve. This blog explores how businesses can balance technological innovation and sustainability when it comes to digital strategy.

What is AI?

In simple terms, AI refers to the group of technologies that process information, learn from data, problem solve and mimic human thinking. AI computational power is developing at an exponential pace, doubling every three to four months. Today, AI easily outperforms human capabilities in various tasks. This rapid pace of development and the widespread uptake of AI raises several ethical concerns. One such concern is the question: what is the environmental cost?

The environmental cost of AI

There are three primary concerns relating to the environmental impacts of AI:

• Energy consumption and carbon emissions
• Water usage
• e-Waste

Energy consumption and carbon emissions

The exact energy impact of a single AI model is difficult to quantify. A recent study conducted by the University of California Riverside, found that writing a 100-word email using ChatGPT consumes as much energy as powering 14 LED lightbulbs for one hour. While this may seem insignificant, the cumulative energy impact becomes substantial when considering that ChatGPT has over 400 million active weekly users (as at March 2025). You also have to then consider the environmental cost of sending the email itself.

According to the International Energy Agency, data centres (facilities that house AI-technology) account for only 1-1.5% of the world’s electricity usage. While this is a relatively small contribution on a global scale, it’s important to remember that AI platforms were only adopted by the mainstream in the last couple of years – we’re still in the midst of the AI-boom. According to Scientific American, this contribution could increase rapidly.

Perhaps an even bigger consideration is the impact that’ll arise from the application of AI platforms. For example, Scientific American reported that ExxonMobil is using Microsoft’s Azure platform to optimise mining operations. Even though Microsoft has a commitment to sustainability and reducing its own carbon footprint, it’s unable to control how its technology is used, and in this case, the platform is indirectly contributing significantly to increased carbon emissions.

Most data centres still rely on fossil fuels. It’s important to try and reduce the impact of all emissions, no matter their contribution on a global scale. While it can be said that a single AI model isn’t going to destroy the environment, problems arise when hundreds of companies each develop their own AI models, each of which are then used by thousands. And this is the path that we as a society appear to be taking. While some AI developers are transitioning to clean energy to power their data centres, research suggests they’re unable to keep up with the pace of technology development. A 2024 report found that the energy efficiency of AI is generally improving by 26% a year, however the annual growth in demand for AI is rising by 400%.

‍

Water usage

Data centres generate a significant amount of heat and require cooling systems that rely on water. The University of California Riverside study found that asking ChatGPT to write a100-word email requires just over a litre of water to cool the necessary hardware.

The topic of data centre water consumption is complex with many AI providers outright refusing to disclose how much water they’re using. In a 2022 court case between Google and an Oregon-based media publication, Google argued its data centres’ water consumption was a trade secret. The case was ultimately decided in favour of the media publication, forcing Google to disclose the water consumption of its Oregon-based data centre, which ended up being substantial – over a quarter of the city’s water supply. Google claimed that the consumption was comparable to the annual water footprint of 29 golf courses. Whether or not that’s true, the case revealed that Google’s water consumption was on a trajectory, having increased three-fold since opening five years earlier – an indication that data centres are among the heaviest users of water in the towns in which they’re located.

The impact on water resources varies based on geography, with some data centres operating in cities with an abundance of water, and others operating in areas facing drought and water-stress.

‍

E-Waste

E-Waste is a rising problem. The rapid advancement of technology means there’s a continuous cycle of components that’ve been discarded due to becoming obsolete. It’s been predicted that AI models could produce up to 5 million tonnes of e-waste by 2030.

The issue of e-waste is twofold: first, e-Waste often contains hazardous materials, such as lead and mercury, that contaminate the environment and harm human health if not disposed of correctly. Second, valuable resources like copper, gold and silver, are not being recycled and repurposed properly. While recycling programmes and circular economy initiatives are growing, e-waste is currently expanding at a far quicker pace.

‍

How to pursue an innovative digital strategy while reducing environmental harm

While AI developers are responsible for mitigating the environmental damage created in the training and development process, organisations, businesses and individuals also need to be conscientious when using AI to ensure they’re doing so responsibly. Afterall, every bit of energy we generate has a real-world impact.

So, how do you pursue technological innovation, while also being mindful of your environmental impact?

Here are 7 things to consider –

1. ‍Be mindful of when and how you use AI: One of the most important things we can do, is to be selective with our use of AI. As a standard practice, you should avoid excessive or unnecessary queries, and reliance on AI to complete tasks that you, or another human, could easily do. Each request consumes a significant amount of energy, so being intentional and selective with your queries will help reduce impact.‍
2. Be strategic with your AI queries: Some queries require more energy than others, for example image generation responses require a significantly larger amount of energy than text responses.‍
3. Choose AI platforms that prioritise sustainability: Use platforms that are committed to carbon neutral initiatives and building energy efficient data centres. Do your research and check sustainability policies before selecting AI tools to use in your business. For example, Google Cloud and Microsoft Azure both promote their data centres as being energy efficient.
4. ‍Use pre-trained models instead of training AI from scratch: Refining pre-trained models for your specific needs drastically reduces computing power.‍
5. Prioritise green software development: If you have to develop AI, implement green coding practices such as optimising code for energy efficiency, reducing server load and removing unused features. Regularly compress AI models using techniques such as quantisation and knowledge distillation to reduce energy consumption without sacrificing accuracy.
6. ‍Reduce data redundancy and storage waste: Regularly clean up unused AI-generated content to minimise storage demands.
7. ‍Offset your carbon footprint: If your use of AI is regular, consider investing in a carbon offset program to help counteract some of the energy you are consuming.

Get in touch if you’d like to find out more about how you can embed sustainability communications in your ESG strategy.

‍