AI's Energy Use Is Real. Here's the Full Picture.

If you've been wondering whether AI has an energy problem, you're right to ask. The headlines have been hard to ignore. Data centres expanding at pace, power grids under pressure, and AI named as a primary driver. For leaders with net-zero commitments, this sits uncomfortably alongside competitive pressure to adopt AI faster.

The concern is legitimate. And it deserves more than a dismissal.

The scale of it

The numbers are real. Global data centre energy consumption is predicted to more than double by 2030 - surpassing Japan's total energy use - with AI as a key driver. In 2023, data centres already consumed more than 20% of Ireland's electricity.

What's less often reported is the variation within that footprint. The Register's analysis found that a single query to one of the largest AI models requires around 6,700 joules of energy - equivalent to about eight seconds of microwave use. Run the same type of query on a model fifteen times smaller, and that figure drops to roughly 57 joules. Model selection alone changes energy consumption by a factor of more than 100.

That gap is important. It means the conversation isn't just "AI uses energy" - it's "which AI, used how, by whom, for what."

What is the environmental impact of AI?

AI's energy use is a genuine and growing issue - but the trajectory matters as much as the current position. Models are becoming more efficient year on year. Renewable energy is expanding. And AI is already being used to optimise the very energy systems it draws from: balancing electricity grids, forecasting demand, improving the efficiency of renewable generation. The net picture is more nuanced than most headlines suggest.

The direction of travel

Here's what I think people aren't fully seeing. The concern about AI's energy use is often treated as static - as if today's footprint is tomorrow's footprint, just bigger. That's not the direction of travel.

The Cambridge Judge Frugal AI Hub launched in November 2025 with a specific mission: to develop measurement frameworks and practical tools that help organisations align their AI use with sustainability goals. Their 2025 White Paper found that frugal AI approaches can achieve up to 80% less energy use and 70-90% reductions in total cost of ownership - without compromising what the AI actually delivers.

That's not a marginal improvement. That's a structural shift in what responsible AI use looks like.

The question for leaders isn't whether to use AI. It's whether to use it thoughtfully.

What frugal AI means in practice

Frugal AI isn't a new technology. It's a discipline - a set of habits around how you select, deploy, and manage AI in your organisation. And it's more accessible than it sounds.

The starting point is model selection. Most business tasks don't require a frontier model. Drafting internal communications, summarising documents, answering standard queries - these are well within the capability of much smaller, cheaper, more energy-efficient models. Using a large model for a small task is the equivalent of driving a lorry to collect a pint of milk.

Data practices matter too. AI trained or run on cleaner, smaller datasets uses fewer resources. 85% of UK IT leaders recognise that better data management reduces carbon footprints, according to research by Information Age and NetApp - yet single-use data storage remains common. Auditing what you're storing, labelling it properly, and deleting what you don't need is not glamorous work. But it has a measurable impact.

Finally, measure what you use. Cambridge's three-level framework covers total cost of ownership, return on investment, and alignment with sustainability goals. It gives organisations a way to track and report their AI footprint alongside financial returns - treating AI as a sustainability asset rather than a liability.

You have more choices than the headlines suggest

The fear is understandable. Energy use is a real issue, and business leaders are right to take it seriously. But "AI has an energy problem" and "AI is getting more efficient, and you can choose how you use it" are both true at the same time.

Frugal AI is the practical expression of that second truth. Right model. Right task. Right data. Measure what you use.

If you want to go deeper on the environmental and ethical dimensions of AI leadership, this conversation on AI Night School's YouTube channel is worth your time. And if you're building your own AI capability with these questions in mind, our programmes for CEOs and COOs cover this territory directly.