By Chris Oatey, Regional Director, and Emily Jackson, Research Manager, 2CV

It’s no secret that AI is improving rapidly, becoming an integral part of our industry. We’re seeing it utilised by clients, suppliers dedicated to it, and AI sprinkled throughout the research process. While it´s hard to argue against the efficiencies, one discussion that appears to be missing is the environmental impact of this AI usage. This article takes the form of an interview to discuss the often-overlooked environmental costs of using AI within the industry.

Chris: We might discuss the financial costs associated with AI—the cost of using a new tool, or the savings it brings through related efficiencies, but what I don’t notice is much discussion about the environmental cost. Is that just my circles, or do you notice much concern with sustainability in AI conversations today?

Emily: It’s starting to happen; there’s an increase in focus on AI’s potential to help address sustainability issues. But, you´re right, there is a gap—AI is celebrated for its efficiencies, the discussions about its environmental costs are scarce. The conversation at the moment in our industry is more about relevancy of AI synthetic data versus human interactions. But environmental impacts haven’t been adequately addressed.

Chris: Yeah, I’d agree. I recently watched a TED video about Food for the Future. They mentioned in passing how AI is being used to help combat agricultural environmental issues but didn’t explain how. This made me think, should we—and how can we—integrate AI responsibly without losing sight of our environmental responsibilities?

Emily: The implementation of AI involves various environment costs. If we just think about large language models (LLMs), there’s stats out there that talk about the energy used in the training of these models and the ongoing usage. It’s projected that by 2026, data centres could consume 6%+ of the total electricity usage in the U.S; demand globally is expected to double from 2022 levels by 2026. That’s concerning when you think about sustainability.

Chris: That’s an awful lot, and I’m glad you managed to find that statistic, I found it quite hard to find reliable sources of information. There are a few analyses out there but it seems that we’re not at the stage where facts about AI and sustainability are fully known.

Emily: Exactly, it’s not completely transparent. There’s more to it than just energy and water too, that people may not think of. The full supply chain to get to your prompts or AI-run interviews has an impact—from mineral extraction, to energy, to water consumption. For instance, mining can damage ecosystems and exacerbate water scarcity, which is vital to consider in our sustainability discourse. And then we run these machines in data centres, which consume vast amounts of electricity, also require extensive water for cooling purposes, further straining our already limited freshwater resources. Every single time we interact with AI, that’s additional energy used by the machines which means more heat and therefore more water required to cool.

Chris: That’s a great point and something that you’ve been talking about internally. How does the AI usage correlate to water consumption in practical terms?

Emily: One interaction with a large language model, like ChatGPT, can use about half a litre of water.

Chris: So, what about in research generally? You had been looking into this for a recent project we did that included AI prompting for our open ends in a survey.

Emily: Yeah, so for that study we were using an AI prompter to get richer answers within our survey. When the respondent answers, it will ask a follow-up question to get further detail. If we assume the same half a litre of water per interaction estimate, and we’re talking about thousands of respondents…well, it’s a lot of water consumption!

Chris: It’s pretty alarming, but let’s be honest—we’re all still going to use it. As we mentioned earlier, the efficiencies it can bring are too great to ignore. So instead, what tangible steps can researchers take to manage AI’s environmental implications?

Emily: Education and awareness are critical. If we train teams to use these tools effectively, focusing on efficient prompts and minimizing interactions, rather than using multiple prompts to get what we need, we can reduce our overall impact. Ideally agencies should also prioritise seeking transparency in AI platforms, ensuring they understand the energy and water consumption.

Chris: Okay, so we could consider in our initial selection criteria which of the platform providers are being clearer on the cost—though I get the impression that’s not simple for them to estimate either.

What about offsetting? That’s one of the most used mechanics at the moment for companies to try and reduce, well offset, their impact on the environment. Should we be doing that?

Emily: It’s complex. Ultimately, each entity should be accountable for their AI usage, and offsetting should be part of a broader sustainability strategy. It is a starting point, not the final solution.

Chris: It seems challenging for us all to get this right, and there isn’t a silver bullet answer we can get to in this forum. Understanding the implications of AI in our broader environmental context is essential.

Emily: Yes, it’s vital to engage in these discussions and share knowledge. Awareness is the starting point. Offsetting is a starting point. As AI continues to evolve, so too must our approach to its usage and sustainability.

While offsetting is not the only answer, it is a starting point. And at 2CV we’re donating to WaterAid for every project completed to help not only with the impact of AI, but also the wider usage of data centres within our day-to-day work.

If you’d like to find out more or discuss further, please reach out. For transparency, in this article one prompt was used to help condense the original 60-minute discussion.

This article was first published in the Q1 2025 edition of Asia Research Media