AI and Sustainability: A Tool We Can’t Afford to Ignore or Misuse

It’s hard to go a day without using AI to some extent these days.

Whether it’s drafting an email, analysing data, or pulling together a report, AI has quickly become embedded in the way we work. In the sustainability space, the shift is especially noticeable. Tasks that once took hours, even weeks, can now be completed more efficiently with the help of AI.

In many ways, that’s a good thing, right?

Sustainability has always been a data-heavy discipline. Whether it’s tracking emissions, analysing’s resource use, or aligning with reporting frameworks, the work often involves pulling together large volumes of complex information and translating it into something meaningful. It can be time-consuming, repetitive and at times, difficult to scale.

This is where AI is starting to make a noticeable difference.

Used well, AI is proving to be a powerful tool for sustainability professionals. It can support faster data analysis, identify patterns that may not be immediately obvious, and help organisations produce clear, more consistent reporting.  From using AI in Crop Yield Forecasting (Global Green Growth Institute) to tackling illegal fishing (Ocean Mind), AI has the potential to make sustainability efforts more informed, efficient, and accessible.

Put simply, AI doesn’t just make the work easier, but it helps us understand our impacts better, and faster.

But as with most things in sustainability, there’s another side to the story.

What many people often forget about AI is that it is not an intangible, impact-free solution. While it may seem as though everything is happening behind the scenes and “in the cloud”, the reality is that AI relies on a vast network of physical infrastructure (data centres, servers, cooling systems), all which require significant resources to operate.

So what exactly is the cost of asking ChatGPT to draft an email or answer a simple question?

Due to their complexity, these systems require substantial amounts of power to operate, resulting in high energy use, particularly during the training phases (EY 2024). This high energy use often leads to considerable Greenhouse Gas emissions, especially when powered by non-renewable energy sources (which most are).

These data centres are consuming considerable amount of energy and leaving behind a significant carbon footprint.

According to estimates, ChatGPT emits around 8.4 tons of Carbon Dioxide per year (The Carbon Footprint of ChatGPT ). However these figures have been hard to provide as data is extremely insufficient in this area as AI lacks standardised measures to track energy use, water consumption and e-waste across its lifecycle (World Economic Forum).

And it’s not just energy.

As these data centres guzzle electricity to produce computational outputs, they get hot. Very hot. They require constant cooling from water intensive infrastructure. In a recent study conducted by UC Riverside, they reported that ‘during the training of  the GPT-3 language model in Microsoft’s data centre in the U.S., around 700,000l of freshwater was evaporated’ (Li, 2025). They predict that by 2027, the global AI usage demand is projected to account for 4.2-6.6 billion cubic metres of water withdrawal. That’s more than the total annual amount of water withdrawal from half of the United Kingdom.

Where is all of this water going?

It is primarily due to large demand of energy usage during the training process that in turn, causes massive amounts of heat production. Large quantities of freshwater is required to keep the temperatures of the servers under control and cool down the machinery.

Even beyond training, AI models continue to consume significant amounts of water in the inference process. This occurs when Chat-GPT is used to answering questions or generating text.

In fact, AI queries require 14x more energy than a simple Google Search (Viro AI). Why? Because large language models like GPT-4 have billions of parameters that must be activated, processed and interpreted, each time a query is run. Even when you ask simple questions.

Now you might be thinking, “I only use AI a few times a day. It cant be that bad.”

Sure, a single query or a generated report doesn’t feel resource intensive. But when scaled across organisations, industries and missions of users globally, the cumulative impact becomes much more significant.

The efficiency we gain on one side may be contributing to increased resource use on the other.

How can we reduce the environmental impact?

The question is not whether AI is useful, it clearly is. The question is how we use it in a way that aligns with the outcomes we’re trying to achieve.

We cannot avoid AI. The benefits it offers are too significant to ignore. But we need to adopt it with considerations, particularly for those of us that are actively trying to reduce our environmental footprint.

Using AI sustainably is about using it with intention.

Not every task needs to be automated, and not every process needs to be enhanced with AI. Being selective about using AI where it meaningfully improves outcomes rather than simple because it is available, can help avoid unnecessary resource use.

Match the model with the task

Not every task needs a massive model. For everyday tasks such as answering simple questions or drafting emails, smaller, efficient models can do the job with less energy.

Tools like EcoLogits allow users to explore the estimated environmental impact of different AI models. By inputting prompts, users can gain insight into factors such as energy use, water consumption, and potential climate impact. It’s a useful starting point for building awareness around how everyday AI use translates into real-world impacts.

Choosing more sustainable models

One way to address the issue of the environmental impact of AI models is for better transparency. The AI Energy Score is an initiative “to establish standardized energy efficiency ratings for AI models, helping the industry make informed decisions about sustainability in AI development”

By using the AI energy score, you can assess the different ranking of AI models and ultimately, chose a model the most energy efficient model (5 stars) for your needs. The use of the AI energy score can also assist enterprises in using these benchmarks as procurement criteria.

The AI Energy score will hopefully push for greater transparency from AI model developers in disclosing energy efficiency data.

Final thoughts

AI can support stronger reporting, better insights, and more effective strategies. It can help organisations better understand their impacts and make more informed decisions.

At the same time, it introduces a new layer of complexity, one that sits within the very systems we are trying to improve.

Navigating that complexity requires balance. It requires acknowledging both the benefits and the trade-offs, and being willing to engage with both.

Because the reality is, AI is not going anywhere. Its role in business, and in sustainability will only continue to grow.

The organisations that will benefit most are not those that adopt it the fastest, but those that adopt it the most thoughtfully.

Used well, AI has the potential to accelerate sustainability progress in ways that were not previously possible. Used carelessly, it risks becoming another overlooked source of environmental impact.

As sustainability professionals, we are used to working within trade-offs. AI is simply the next one to navigate.