The AI We Build Should Not Cost Us the Planet

We are building machines that can write, see, listen, predict, create and solve problems that once seemed impossible. Artificial intelligence is becoming part of almost every area of our lives. But there is one question I don’t think we ask enough:
What does it cost the environment to make AI this powerful?

AI feels completely digital because all we usually see is a screen. But behind that screen are huge data centres filled with servers. They need electricity, physical infrastructure and importantly cooling. As AI systems become bigger and more people start using them, the resources needed to run them matter too.

Cooling is one part of this that I find especially interesting. Some data centres use large amounts of water for cooling and that feels like a problem when water is already such a limited resource in many parts of the world.

It made me wonder: do we always have to keep improving the same cooling systems or could we find completely different ways of doing it?
I started thinking about this after learning about the charcoal waste produced by traditional ironing carts. It made me wonder whether discarded charcoal could be used for something instead of simply being thrown away.

Could charcoal waste, when combined with another material, somehow be turned into a cooling material that can do some of the things water does in a cooling system while using much less water?
I don’t know if it would work. There could be very good physical or engineering reasons why it wouldn’t. But I think questions like this are worth exploring. AI and machine learning could potentially help researchers look through different combinations of waste materials and predict which ones might have useful thermal properties before testing all of them in the real world.

In that situation, AI would not just be using resources. It could actually help us find ways to save them.
But this doesn’t mean we should reject AI.
That would be too simple.
AI can also be incredibly useful for the environment. It can help scientists analyse huge amounts of climate data, improve weather and disaster predictions, monitor forests and oceans, detect pollution, make electricity grids more efficient and help farmers use water and fertilisers more carefully.
So I don’t think the question is whether AI is good or bad for the environment.
The question is how we use it.
I think we also need to change the way we measure progress in AI.

Right now we usually talk about how accurate a model is, how fast it is, how many people use it or how much more capable it is than the previous model. Those things are important but they don’t tell us everything.

We should also be asking how much energy and computing power were needed to get those results.
If a simple search can answer a question why use a huge AI model for it?
If a smaller model can do the same job just as well why automatically choose the larger one?
This is where my idea of right-sized AI comes from.
To me right-sized AI means using the simplest and most efficient technology that can actually do the job properly. It doesn’t mean that we should stop building powerful AI. It means we shouldn’t use more computing power than we actually need just because we can.

Imagine two systems that solve the same problem. One needs many times more computing power but gives almost the same result. We would normally compare their price and speed.
I think we should also compare what they cost the environment.

Before using a large AI system we could ask some simple questions:
What problem are we actually trying to solve?
Could something simpler do it?
How much computing power does it need?
What benefit are we getting from using it?
Is that benefit worth the resources being used?
These questions could change how we think about technological progress. Maybe the goal shouldn’t always be to build the biggest or most powerful AI possible. Maybe it should be to build the right AI for the job.
At the same time I don’t think making AI more efficient is enough.

We should also actively use AI to solve environmental problems.
AI can analyse satellite images to help detect illegal deforestation. It can help predict electricity demand and make it easier to use renewable energy efficiently. It can help farmers decide when and where they actually need water. It can also find patterns in ecosystems that would be extremely difficult for people to identify by looking at the data manually.
In these cases AI isn’t just consuming resources. It is helping us use resources better.
That difference matters.

Almost every major technology has created new environmental problems while also giving us new ways to solve old ones. I think AI will be the same.
That’s why I don’t think we should ask whether AI is environmentally good or bad. We should ask whether we are using it responsibly.
I also think this should become part of how we teach AI.

Students learning artificial intelligence shouldn’t only learn how to train models and build applications. They should also learn to ask what their technology consumes, where those resources come from and what happens outside the computer.

The people learning AI today will eventually decide what AI looks like tomorrow. If we teach them to think about sustainability from the beginning it won’t have to be something we try to fix later.

We are gaining the ability to create technology incredibly quickly.
Our responsibility needs to keep up.
AI could help us understand our planet in ways we couldn’t before. It could help predict disasters, protect forests and oceans, manage water and energy and make many of our systems more efficient.

But we shouldn’t assume that every new technology is automatically progress.
Progress isn’t just about making machines more powerful. It’s about learning how to use that power well.

I don’t want a future where we have to choose between artificial intelligence and a healthy planet.
I want a future where we can have both.
AI exists on this planet. It isn’t separate from it.
Maybe the smartest AI won’t be the one that uses the most computing power or has the most impressive capabilities.
Maybe it will be the one that knows when it actually needs them.

WriteToWin Monsoon 2026