Is AI Good or Bad for Sustainability? It's More Complicated Than You Think

Is AI Good or Bad for Sustainability? It's More Complicated Than You Think

Every time someone types a prompt into ChatGPT, a data centre somewhere hums a little louder. That is not a metaphor. It is electricity, water, and heat, and the numbers are getting bigger fast.

So here is the uncomfortable question nobody in the AI space actually wants to answer honestly: is using AI bad for the planet?

The honest answer is not "yes" or "no". It is "it depends, and most people making claims about it are either overconfident or selling something".

We have done some proper digging into this. Not the usual "read a blog post and paraphrase it" kind of research. We pulled data from the International Energy Agency's 2025 and 2026 energy reports, Stanford's AI Index and Foundation Model Transparency Index, Lawrence Berkeley National Laboratory's data-centre projections, peer-reviewed research from ACM and Nature Sustainability, Google and Microsoft's own environmental reports, the UK Competition and Markets Authority's Green Claims Code, and the Greenhouse Gas Protocol's guidance on avoided emissions. We also looked at what the Advertising Standards Authority has actually ruled on sustainability claims.

The short version: the data is messier than either side admits. But the signal is clear enough to act on.

The Footprint Nobody Talks About

Let's start with the bit that makes AI companies uncomfortable. Data centres already account for about 1.5 percent of global electricity demand. The International Energy Agency projects that could reach 3 percent by 2030. In the US, data centres used 4.4 percent of national electricity in 2023. By 2028, that could be 12 percent.

Training a single large model is genuinely staggering. Stanford's AI Index estimated GPT-4's training emissions at roughly 5,184 tonnes of carbon dioxide equivalent. Grok 4? Try 72,816 tonnes. To put that in perspective, that is roughly the annual carbon footprint of 15,000 British homes. For one model.

Then there is water. Data centres use it for cooling, and the electricity powering them uses it too. Berkeley National Laboratory estimated US data centres consumed about 66 billion litres directly in 2023, with another 800 billion litres consumed indirectly through electricity generation. In water-stressed regions, that matters.

And here is the kicker. Stanford's 2025 Foundation Model Transparency Index found that ten major AI companies, including Google, Amazon, OpenAI, Anthropic and xAI, disclosed absolutely zero information about model-level energy, carbon, or water use. Zero. So when a tech company says their AI is "sustainable", they are asking you to take their word for it. They are not showing you the receipts.

But Here's Where It Gets Interesting

Because the same technology that burns through electricity can also save far more of it elsewhere. The IEA estimates that existing AI applications could help avoid approximately 1.4 billion tonnes of carbon dioxide emissions by 2035. That is several times the projected emissions from data centres themselves.

In transport, AI-assisted route optimisation can improve vehicle efficiency by 5 to 10 percent. Google's fuel-efficient routing feature reportedly enabled more than three million tonnes of avoided emissions in 2025 alone. In buildings, AI-optimised heating and cooling can save around 10 percent of energy. Predictive maintenance can catch equipment failures before they waste power or leak resources.

Winnow, a food-waste platform, uses computer vision to track what gets thrown away in commercial kitchens. One case study with a Lisbon Marriott reported a 50 percent reduction in food waste over nine months. That is a real, measurable, physical reduction. Not a vibes-based sustainability claim.

The pattern is clear. AI helps when it changes something physical for the better. It hurts when it just adds computation without producing an equivalent real-world benefit.

The Greenwashing Problem

This is where it gets messy. A business can reduce staff time with AI and call it "sustainable" without reducing a single kilowatt. A digital process can eliminate paper while increasing travel. An efficient model can still drive higher total electricity use if you deploy it at enormous scale.

Google reported a 12 percent reduction in data-centre emissions during 2024. Sounds great. They also reported a 27 percent increase in data-centre electricity consumption. Both statements are true. Neither one tells the whole story on its own.

The UK's Competition and Markets Authority has rules about this. The Green Claims Code requires environmental claims to be truthful, accurate, and supported by evidence. The Advertising Standards Authority has ruled against companies using vague terms like "sustainable" without substantiation. AI does not get a special exemption.

So when a company says "our AI-powered platform is green", ask them: compared to what? Measured how? Over what period? Including what hardware, what electricity, what water? If they cannot answer, the claim is marketing, not evidence.

What This Means for Small Businesses and Web Designers

You probably are not training frontier models. You are using ChatGPT to draft copy, maybe generating some images, possibly running analytics. Your individual footprint is small. But scale changes the maths.

If you are a web designer using AI tools daily, here is the practical question: is the AI actually replacing something wasteful, or is it just adding a layer of computation on top of work you would do anyway?

Using AI to write blog content that replaces a two-hour drive to a client meeting? That is probably a net win. Using AI to generate fifty image variations you will never use? That is just burning electricity for the sake of options.

The same principle applies to the websites we build. A well-structured, fast-loading site that serves users efficiently is sustainable by design. An AI-heavy site with bloated scripts, endless personalisation, and chatbots nobody asked for? That is computation dressed up as innovation. We have written about whether AI tools will replace traditional websites, and the sustainability angle only reinforces the point. Simpler, faster, more deliberate wins every time.

The same logic applies to how we think about what actually threatens web designers. It is not AI taking your job. It is the assumption that more technology automatically means more value. Sometimes less computation, used more deliberately, is the better answer.

The Question Nobody Asks

Before adopting any AI tool, ask yourself: what physical outcome will change? Is AI actually necessary here, or would a spreadsheet, a rules engine, or simply doing nothing solve the problem with less cost and less energy?

Because here is the thing. A small model that prevents equipment failure, reduces unnecessary journeys, or cuts food waste has a compelling environmental case. A frontier model used to generate routine emails nobody reads and decorative images nobody clicks? That is just expensive electricity with a confident tone.

The question is not whether AI is good or bad for sustainability. It is which AI, performing which task, on which infrastructure, compared with which alternative, and with what measured result. Everything else is noise.

And if you want to know where web design is actually heading, the future of web design is less about bolting on AI and more about building things that work properly. Sustainable design is just good design, with smaller bills.

Key Takeaways