We’ll send a password for the resource area and regular UK-focused AI updates. Unsubscribe anytime.

Why this matters to you, not just “the planet”

It’s easy to treat AI and the environment as someone else’s problem. But there are two reasons it’s worth your attention as a small business owner.

First, environmental cost and financial cost are connected. AI that runs constantly for tasks that don’t really need it is AI that also costs you money. If you pay per query or per token, waste shows up directly on your card statement.

Second, your staff and customers may start asking. A business that markets itself on sustainability — a landscaping firm, a tree surgery service, an eco-friendly valeting business — might get an awkward question from a client about its AI use. Some employees may feel uneasy using tools they see as environmentally costly. Having an honest answer, such as “here’s what we use AI for and why”, is worth more than avoiding the subject.

The real environmental cost: energy, water and infrastructure

AI feels almost costless when you type a question and get an answer. But somewhere, computers process the request.

The International Energy Agency estimates that global data-centre electricity consumption was around 415 terawatt-hours in 2024, roughly 1.5% of global electricity use. Its latest analysis says data-centre electricity consumption grew by 17% in 2025, while electricity use by AI-focused data centres grew by around 50%. The IEA expects global data-centre electricity consumption to roughly double from around 485 TWh in 2025 to around 950 TWh in 2030.

AI is a major reason for that growth.

But there’s another side. AI is becoming more efficient. The IEA says energy use per individual AI task has fallen dramatically as hardware and software improve. A single simple AI text query isn’t the main problem. The concern is the total scale of billions of AI tasks every day, plus the much heavier demands of training large models and running complex AI systems.

The UK picture is particularly relevant. Data centres currently use less than 10 terawatt-hours of the UK’s 319 terawatt-hour total electricity consumption — about 3% — but demand is expected to rise to as much as 71 terawatt-hours between 2025 and 2050. In London, data centres already account for nearly a fifth of the capital’s electricity use.

Water is also part of the story. Many data centres use evaporative cooling systems. A UK government-commissioned report found that the UK faces a projected daily water deficit of nearly 5 billion litres by 2050, and current national water resource plans don’t adequately account for growing demand from AI data centres. Most current UK data centres rely on evaporative cooling systems that consume drinking-quality water and can lose up to 80% of it through evaporation.

A recent UN report warned that water used by AI data centres globally is expected to equal the everyday water needs of 1.3 billion people by 2030.

This doesn’t mean every AI question you ask is draining a reservoir. But it does mean the industry-wide picture is strained, and you can reasonably factor that into how much AI use is proportionate for your business.

What AI costs your business

The environmental impact is global. The financial impact lands directly on your bank account.

Suppose you run an independent online clothing retailer. You subscribe to an AI service for £20 a month to help write product descriptions. That seems reasonable. But then staff still have to check every description, correct wording, verify measurements and make sure products aren’t described inaccurately. If you sell hundreds of products, that checking time becomes significant.

The real cost is:

subscription + staff time + checking + mistakes + management time

Looking only at the monthly price gives the wrong answer.

As of mid-2026, typical UK pricing for common SME-friendly AI tools looks roughly like this:

• ChatGPT Plus — around £18 to £20 per person per month
• Claude Pro — around £16 per person per month
• Microsoft 365 Copilot — around £16 to £23 per person per month on top of an eligible Microsoft 365 licence
• ChatGPT Business — roughly $20 to $25 per seat per month, with a two-seat minimum

Multiply any of those by your team and by twelve months. A five-person design or marketing agency paying £20 per seat is spending £1,200 a year before any additional usage costs. That’s a real business expense.

Then there’s pay-per-use or API pricing. Every time an automation runs, it costs money again. Run it a thousand times a month and you pay a thousand times.

Ask whether you need AI at all

AI is being added to almost everything. That doesn’t mean you need to use every feature.

Consider a hair salon with six employees. The owner might be offered an AI chatbot for the website. But if most customers book through an existing booking system or phone the salon, what problem is the chatbot solving? The business could end up paying for a solution to a problem it doesn’t have.

The better question isn’t:

“What can AI do for my salon?”

It’s:

“Where do we currently have a problem that AI might solve better than what we’re doing now?”

This single change in thinking can save money before you spend a penny.

Don’t collect AI subscriptions

One of the easiest ways to waste money is collecting AI tools. You start with one. Then someone recommends another. Marketing wants an image tool. The office manager wants a meeting assistant. Someone else wants a writing tool. Before long, you have five or six subscriptions.

Individually they don’t look expensive. Together they become a significant annual bill. Staff also have to learn several systems, information gets copied between them, and nobody is quite sure which tool the business is actually supposed to be using.

Before buying another AI service, check what you already have. Your Microsoft 365, Google Workspace, accounting software, CRM or other business systems may already contain useful AI features. You may not need another subscription at all.

When AI can reduce your environmental impact

AI has an environmental cost. But the thing AI helps you do might have a larger environmental cost of its own.

Consider a regional courier company. If AI improves route planning and reduces unnecessary mileage, the business could use less fuel and produce fewer emissions. The AI processing has an environmental cost, but so does driving hundreds of unnecessary miles. The relevant question is which overall process uses fewer resources.

The same principle applies to a commercial printer organising jobs to reduce paper waste, a joinery workshop using AI to double-check cutting lists before ordering timber, or an accountancy practice processing routine client data more efficiently.

The IEA has identified potential for AI applications to reduce emissions in transport, buildings and industry. But these benefits aren’t automatic. They depend on how AI is actually used.

The big idea: write once, run many

Here’s the single most useful idea for controlling both the cost and the environmental impact of AI.

There’s a difference between:

1. asking AI to do a task, and
2. asking AI to write you a tool that does the task.

Say you run a small bookkeeping practice. Every week you reformat a spreadsheet of client transactions into a specific layout. You could ask an AI chat tool to do the reformatting each week. That works, but you pay for a fresh AI response every time, and each response uses server time, electricity and indirectly water.

The alternative: ask AI to write a short piece of code, often in a programming language called Python, that does the reformatting automatically. Python is free to use, including commercially, under its licence. You run that code yourself, on your own computer, for free, as many times as you like. You only used AI once, to write the tool. After that, it costs nothing extra and doesn’t touch a data centre.

This is the “write once, run many” approach.

But don’t blindly run AI-generated code

AI can write code that contains mistakes or security problems. Don’t copy a Python script from an AI tool and immediately run it against important business systems. Test it first, understand what it does, keep backups, and get technical help if the task involves sensitive information.

If you find yourself asking AI to perform exactly the same predictable task again and again, ask:

“Can I build this once?”

The initial effort might be higher. The long-term cost could be much lower.

Use the smallest amount of AI that solves the problem

You don’t always need the most powerful model.

If you need a short customer email, you may not need an advanced reasoning system. If you’re summarising a straightforward document, you may not need an elaborate AI workflow. If you’re creating a simple spreadsheet formula, you may not need an AI agent with access to your entire business.

Think about the task first, then choose the simplest solution that works. That might be:

• no AI
• a normal search
• a spreadsheet
• a template
• an existing software feature
• a simple automation
• a small piece of code
• a free AI service
• a paid AI service

There’s nothing wrong with choosing the least sophisticated option. For many small businesses, that’s exactly what you should do.

Better prompts reduce waste too

Clearer instructions usually mean less wasted time and less back-and-forth.

Imagine you run a local bakery. You could type:

“Write me a post about our Christmas products.”

Then ask it to make it more local, shorter, friendlier, less salesy, more suitable for Facebook and in British English. You’ve now used several requests.

Instead, tell the AI what you want at the beginning:

“Write a 100-word Facebook post for a family-run bakery in Yorkshire promoting our Christmas cakes. Use British English. Keep the tone warm and local, don’t use sales jargon, and finish by inviting customers to order in the shop.”

You may still edit it, but you’re more likely to get a useful first draft.

Don’t generate things simply because you can

This is particularly relevant to AI images and video.

You might generate 20 versions of an image because there’s no photographer involved, then choose one. The other 19 still used computing resources. Don’t become obsessed with counting every AI image, but ask:

“Am I actually going to use this?”

If the answer’s no, don’t generate it. The same applies to text. You don’t need 50 social media posts just because AI can create them.

Measure the financial benefit instead of guessing

Pick one task. Record how long it takes without AI. Then use AI for the same task for a week.

Record:

• how long the AI process takes
• how much time you spend checking it
• how many mistakes need correcting
• what the AI tool costs
• how much useful work you actually get
• whether customers or employees notice any improvement

Suppose you run a small recruitment agency. A consultant spends 30 minutes preparing the first draft of a candidate profile. AI reduces that to ten minutes, but the consultant spends another ten minutes checking and correcting it. You’ve saved ten minutes. That’s useful.

Now suppose the AI draft contains frequent errors and takes 35 minutes to check. You’ve made the process worse. That’s useful information too.

What about the cost of mistakes?

AI output isn’t automatically correct.

A cheap AI system that produces unreliable work can be much more expensive than a more expensive system that produces consistently useful results.

A minor wording problem in a property description might not matter. But if AI invents a feature the property doesn’t have, the estate agent has a problem. An incorrect figure in a client communication from an accountant could create a much bigger issue.

Don’t measure AI only by time saved. Also ask:

“What does an error cost me?”

The higher the potential cost of a mistake, the more human checking you need.

Practical ways to cut your AI energy and water footprint

You can’t control how a data centre is cooled. But you can control how much you ask of it.

1. Batch similar requests together rather than sending lots of small, separate ones. One well-written prompt covering three related questions is more efficient than three separate chats.
2. Choose the right-sized tool for the job. Not every task needs the most powerful and most energy-hungry AI model.
3. Turn one-off answers into reusable tools wherever you find yourself asking AI the same type of question repeatedly.
4. Set a team habit of checking before automating. Before anyone sets up an AI process to run automatically, ask how often it really needs to run.
5. Review subscriptions every few months. If no one’s used a paid AI seat in the last month, cancel it.

A veterinary clinic with three staff might find that only the practice manager uses AI daily, one vet uses it occasionally, and one has barely opened it. That’s not a moral failing. It’s just worth reviewing, the same way you’d review any other unused software licence.

What to ask your AI supplier

If you’re choosing between AI tools or renewing a contract, ask:

• What’s included in the price, and what happens if we go over our usage allowance?
• Is there a cheaper or lighter version of this tool that would suit most of our day-to-day needs?
• Do you publish any information about the energy or water used by your data centres?
• Where is our data processed, and does that affect cost or performance?
• Do you report emissions, and what are you doing to improve energy efficiency?

You may not get complete answers. That’s fine. The point is to ask. The more customers ask technology companies about environmental impact, the greater the pressure for better reporting.

But don’t let environmental credentials become another reason to buy a more expensive product without evidence that it solves your problem.

The human side: staff, customers and trust

Some of your staff may already be uneasy about AI’s environmental cost, especially younger employees who are more attuned to sustainability issues.

A car detailing business that prides itself on eco-friendly products might find staff genuinely bothered by a mismatch between that message and unlimited AI use behind the scenes. Being dismissive — “it’s just electricity, don’t worry about it” — tends to erode trust. Being open about it tends to build it.

The same goes for customers. If your business talks publicly about sustainability, it’s worth being honest that AI tools have a footprint too, rather than presenting AI adoption as automatically green or efficient.

Nobody expects you to have solved a global infrastructure problem. They do notice when a business claims to be sustainable while ignoring an obvious contradiction.

A simple AI cost test

Before introducing a new AI tool or continuing with an existing one, ask:

1. What problem am I solving?

If you can’t describe it clearly, stop.

2. Why does it need AI?

Could a spreadsheet, template or normal software feature do the same thing?

3. What will it really cost?

Include subscription fees, usage charges, staff time, training, checking, implementation and additional software.

4. What happens when AI gets it wrong?

Think about the financial and reputational cost of mistakes.

5. How much value will it create?

Are you saving time, reducing waste, increasing sales, reducing errors or improving customer service?

6. Can I test it first?

Don’t commit to an expensive long-term arrangement before you know it works.

7. Can I build a reusable solution?

If the task is predictable and repetitive, perhaps you don’t need AI every time.

The most responsible AI request might be the one you don’t make

This doesn’t mean you should stop using AI. It means you should use judgement.

If AI saves you three hours of work every week, that’s potentially valuable. If AI helps a courier reduce unnecessary mileage, a printer reduce waste, a workshop reduce timber waste or an accountant process routine information more efficiently, that could be valuable.

But if you’re generating endless social media posts nobody reads, creating hundreds of images you don’t use, or paying for five AI subscriptions that duplicate each other’s features, you probably have a different problem.

And don’t let the environmental argument become another source of guilt. You don’t need to calculate the carbon impact of every email. The sensible position is neither blind enthusiasm nor panic. It’s judgement.