Power-hungry prompting and dirty data centres: the AI dichotomy for planet‑friendly businesses

“Not another AI blog post…”, we hear you. Bear with us - this one’s a little different.

Rather than practical tips on crafting the perfect prompts or the magic of “vibe coding”, we want to share a tension we’re increasingly seeing in our work: AI can unlock real productivity and creative benefits, but it also raises a moral and ethical dilemma for organisations trying to put the planet first.

The dilemma is this: for all its promise, AI has a hidden cost - and that cost is environmental.

Remember those cutesy plastic‑wrapped AI “action figure” images that did the rounds on social media a while ago? To generate them at scale, tonnes of carbon will have been [belched out into the atmosphere](https://guides.library.queensu.ca/c.php?g=740510&p=5344286#:~:text=Artificial Intelligence models consume an,training%2C operation%2C and maintenance.), and huge volumes of freshwater will have been used to keep data centres cool.

So widespread was the trend that OpenAI’s CEO, Sam Altman, publicly referenced the need to restrict usage for this purpose — with “melting” doing some heavy lifting in the message:

And it’s not only environmental impact. There are serious privacy concerns wrapped up in how these tools are trained and how people use them - which is new territory for many teams.

“The cloud” is still a physical thing

This is only one example of a broader truth: digital has hidden costs, and they need to be measured. We’ve written before about the importance of understanding and reducing digital impact - including how organisations can start to measure a digital carbon footprint and what the emerging website sustainability guidelines mean in practice.

At this point, you’d be forgiven for thinking we’re about to argue that everyone should abandon AI immediately - partly to stop the world burning, and partly to feel better about ourselves.

That isn’t our view.

AI is here, and it’s here to stay. It’s also proving surprisingly difficult to measure the true environmental impact of any single use case. The type and size of a model, the kind of output you generate, where the data centre is, which grid it’s connected to, and even the time of day a request is processed can materially change the footprint.

So, rather than planting a flag, we’d rather this post acted as a signpost: a short, honest primer and a curated set of resources to help you explore the issue further - whether you want to feel more worried, or more reassured.

What does the literature say?

Here are three thoughtful pieces that represent the range of perspectives we’ve found useful. (NB: bold emphasis in quotes below is ours.)

1) MIT Technology Review — a sobering, systems‑level view

A deep‑dive that frames the rapid growth of AI as a serious ecological challenge, while making the key point that impact is hard to measure cleanly:

“If you’ve seen a few charts estimating the energy impact of putting a question to an AI model, you might think it’s like measuring a car’s fuel economy or a dishwasher’s energy rating: a knowable value with a shared methodology for calculating it. You’d be wrong.

In reality, the type and size of the model, the type of output you’re generating, and countless variables beyond your control … can make one query thousands of times more energy‑intensive and emissions‑producing than another.”

Read: https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

2) Andy Masley - a “cheat sheet” that challenges personal‑use framing

A pragmatic analysis that argues we risk focusing on the wrong lever — that there are often bigger wins elsewhere (flights, cars, heating, supply chains), and that personal AI abstinence may be a comparatively small part of the climate equation:

“If everyone in the world stopped using ChatGPT, this would save around 3GWh per day. If everyone in the world who owns a microwave committed to using their microwaves for 10 fewer seconds every day, this would also save around 3GWh per day.

…There must be much more effective things to campaign for that would have a lot more climate impact… Campaigning to stop ChatGPT is exactly like campaigning to use microwaves for 10 fewer seconds.”

Read: https://andymasley.substack.com/p/a-cheat-sheet-for-conversations-about

3) HRW Healthcare - a more reassuring take on individual use (with a call to aim higher)

A useful counterweight if you’re feeling AI guilt. It shifts attention away from individual tool‑use and towards accountability and systemic change:

“Pointing fingers at AI tools while empty private jets criss‑cross the globe… That’s a distraction.

Demand more: Holding companies, industries, and governments to account is one of the most powerful actions we can take.

…One simple switch could do more than a lifetime of prompt avoidance.”

Read: https://www.hrwhealthcare.com/the-environmental-impact-of-ai/

So where have we landed at Knapton Wright?

We’re increasingly comfortable using AI to augment our work — with clear boundaries and conscious intent. We’re using it more often to speed up research, sense‑check thinking, explore alternative angles, and reduce time spent on low‑value drafting. Used well, it can help teams learn faster, iterate better, and widen perspective.

At the same time, we think it’s important to hold two truths together:

AI has an environmental impact, even if any individual prompt feels insignificant.

Focusing solely on personal AI use risks missing bigger, higher‑impact opportunities for change.

As the MIT piece points out, even if a single use case is small, these impacts add up as AI becomes integrated into everyday tools - and as outputs move from text into more compute‑heavy formats.

So our takeaway is this: be aware, be intentional, and keep perspective. Compare AI’s impact to the wider footprint of your organisation - and use that view to prioritise the changes that actually move the needle.

A quickfire primer: what tends to be most power‑hungry?

Q: Does generating an image use more energy than a basic text prompt?

A: Surprisingly, not always. Depending on the prompt and model, a text interaction can require more computation than a simple image generation. (MIT Technology Review touches on this nuance here: https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/)

Q: So are AI‑generated videos less of a problem?

A: Unfortunately not. Video generation is typically far more compute‑intensive than text or still images, and it’s often the highest‑impact category in day‑to‑day content workflows.

Q: Aren’t modern data centres powered by renewable energy now?

A: Some are shifting quickly, but energy density is extremely high — and because renewables are intermittent, fossil fuels still play a significant role in many grids. Big tech is even proposing new nuclear capacity to meet growing demand (BBC overview: https://www.bbc.co.uk/news/articles/c748gn94k95o).

Q: If we can’t accurately measure the footprint of each AI query, what’s the practical approach?

A: Treat AI like any other digital tool: use it intentionally, avoid waste, and prioritise the biggest levers. Set light internal guidance (when to use it, when not to), choose tools/providers transparently where you can, and focus your sustainability effort on the areas that dwarf AI use (travel, procurement, hosting, supply chain and energy).

Q: What’s a sensible ‘lower‑impact’ way to use AI for marketing without pretending it’s free?

A: Use it for thinking and editing before generation. Start with smaller, specific prompts; reuse and refine existing drafts; avoid re‑prompting for tiny tweaks; and be cautious with high‑compute outputs (especially video). In short: fewer, better prompts — and only when it meaningfully improves the work.