AI is a power-hungry tool.
It can also be a problem-solver for energy efficiency.
There's a lot of talk about the environmental cost of AI.
On one side, people are that AI is burning the planet.
On the other, some seem to think it's just another tech cycle - nothing to panic about.
The impact is real. But how we build and use AI matters.
The numbers
AI models, especially large ones, use a ludicrous amounts of energy compared with regular computing.
By 2027, AI could consume 85-134 terawatt-hours a year - similar to Argentina or the Netherlands.
Most projections assume AI systems run at full capacity, all the time. That's rarely the case. Businesses have strong financial incentives to be efficient.
Also, we were doing a pretty good job burning the world even before AI came along. Fossil fuels, inefficient systems, short-term thinking... AI didn't start the fire, although it certainly seems to be accelerating it right now.
The question now is if and how it can help us solve the problem too.
What you can influence
Whether you or I use AI personally won't change much - it's here to stay. Big tech has already committed billions upon billions. That trajectory won't change overnight.
You and I can influence how AI is applied - in your team, your projects, or your company.
Here's how I approach it in practice:
Look at the impact
- Start with cost and benefit
- Assess each use case individually
- Keep up with efficiency research and tools
Build with restraint
- Use models sized for the job
- Fine-tune existing models when possible
- Prefer cloud providers using renewables
Prioritise net-positive outcomes
- Choose projects that reduce waste or resource use
- Help teams measure energy impact
- Adjust or retire systems when they overperform without purpose
Cutting emissions with AI
AI can help cut emissions - when used deliberately.
For example, one of my medical lab clients used AI to automate their testing workflow. Processing time dropped by 92%, while capacity jumped 12x. The energy savings from more efficient operations far outweighed the energy used by the AI system.
We're seeing similar results in logistics, infrastructure, and manufacturing - wherever there's waste to remove or systems to optimise.
But only if we're intentional about it.
Counting the environmental cost
When you start an AI project, do you factor in the environmental cost? Or is it still an afterthought?
I've been bringing this issue into client conversations from day one.
What tools can help?
Platforms like Hugging Face, OpenAI, and Google Cloud are starting to publish energy usage metrics and sustainability dashboards. Research communities are also publishing benchmarks - from model size to estimated carbon output.
It's not perfect, but it's improving quickly.
Final thought
Sometimes the best decision might be not to use AI for a particular task.
AI will consume significant energy. But with thoughtful implementation, we can ensure it delivers value that justifies those costs - and in some cases, helps reduce our overall environmental impact.
We're making tradeoffs. Every AI implementation has an environmental price tag. The question is whether the benefits - efficiency gains, waste reduction, or new capabilities - are worth it.
I believe we can build AI that supports both people and the planet, but only if we approach it with eyes wide open to the real challenges involved. By putting humans - and humanity itself - at the centre of our decisions about AI - considering their needs, their work, and their future - we naturally create more thoughtful, efficient systems that respect both people and resources.
How are you factoring sustainability into your AI work?
I'd love to hear what you're doing - or what's been hard to figure out.
PS: If you're building AI solutions and want to make them cleaner and smarter, feel free to reach out. I'm happy to share what's worked.
I went back to the numbers on this in more detail, including the claims that turned out to be badly wrong, in AI and energy. Sorting real risks from loud ones is the same habit as reducing AI risk.