The environmental cost of Ai

By Jake from South Africa - Winner of the No To Apathy Article Competition

I love AI.

I don't think I am alone in this - everyone seems to use it in some way - even here in South Africa. It is a great equalizer for the most part but costs can be prohibitive and some turn to free Chinese models.

As someone from South Africa, technology has a strange part in our lives. We call traffic lights 'robots'. We are mostly lucky with the level of technology and infrastructure here compared to a lot of neighboring countries. But we also know what load-shedding feels like. That might be a very South African term - it means periods of time when the power is switched off to reduce pressure on the grid. We know what drought is. We know that water is something precious and crucial and its absence can destroy communities.

When I use AI I have a small voice in the back of head - what is the environmental cost of this question I am asking and if everyone is doing and companies are doing it - what is the environmental cost of that.

I use it all the time. It helps me and also can be fun to interact with. I really think AI is an incredible technology that will change our lives - but will it also impact the planet for the worse?

That little voice in the back of my head asking:

What does this actually cost the planet?

What does one question to AI cost?

Obviously I turned to AI to ask this question - and obviously it was not straightforward.

There isn't one environmental metric to figure out the cost for an AI query. It depends on the model, the type of request, the data centre, the cooling system and many other factors like the source of the power.

AI promptly told me that Google published a detailed measurement of its Gemini AI in 2025. It estimated that a median text prompt used around 0.24 watt-hours of electricity, 0.03 grams of CO₂ equivalent and 0.26 milliliters of water.

If I use some simple maths (I did still ask AI - because which dinosaur still uses a calculator), 20 questions would use roughly 4.8 watt-hours of electricity, 0.6 grams of CO₂ equivalent and 5.2 milliliters of water.

Five milliliters. That is like 5 drops of water - okay so I can relax. No harm done - back to using AI.

But wait - then I start to think maybe I should ask for some other sources so I did and:

A 2023 study from researchers at the University of California estimated that a conversation of around 20–50 questions with an earlier version of ChatGPT could consume around half a liter of water, depending on where and when the computation took place.

Woa - that is a bit different. Did I just waste litres of water trying to write this article for No to Apathy to try and get 150 Euros? I mean its a lot in Rands (our currency) but still - scary stuff.

Checking a bit deeper it does seem to show that AI systems have become rapidly more efficient and newer models are much better. There does still seem to be some unaddressed concerns but perhaps the environmental cost of asking AI a question is shifting as quickly as AI itself - and that is a relief.

I am not the only one though - what happens when we consider millions of questions per minute all over the world?

This is where AI promptly showed me that maybe I should be worried.

The International Energy Agency estimates that data centers consumed around 460 terawatt-hours of electricity globally in 2024. Its base-case projection is that this could rise to more than 1,000 TWh by 2030, with AI being a major driver of growth.

That is a number so big I can't even get my head around it as an environmental issue. I checked and South Africa - a big country by all accounts uses around 216.8 TWh per year.

And we are not just talking electricity - data centers have infrastructure, firmware, cooling systems etc - all with an impact. The electricity itself - if it comes from fossil fuels is a massive environmental impact.

This matters in the context of South Africa.

South Africa's electricity system has primarily relied on coal, and the country is also classified as water-stressed. The environmental consequences of additional electricity and water demand are real issues - is AI still that great when you have issues having electricity yourself?

And then there is water

This is what I started out thinking about and what I think is crucially important - think about how many people don't have adequate water access. And we can all agree that safe water supplies are dwindling. Data centers can use water directly for cooling, while electricity generation can also have an water usage footprint.

Using a few hundred liters of water in an area with abundant supplies like in the UK is not the same as using that amount of water in a drought-prone or water-stressed region. Then there is the issue: wastewater. The water that comes out after having been used by cooling systems is polluting to people and the environment.

The US Environmental Protection Agency has reported that cooling systems can create “blowdown" wastewater with high levels of concentrated minerals.

Could AI actually become a green technology?

I think the answer could be yes. And this is where my optimism about AI comes back. Because if AI can make advancements in medicine and technology possible - then it can also be a beacon of light for clean nuclear power, water desalination and much more.

AI could also help electricity grids balance supply and demand. And trust me in South Africa that is a big thing to look forward to. But not just here - I saw in the UK how much water is lost due to shitty infrastructure that has not been maintained - what if AI could help detect leaks and propel us to a much lower waste society. It could optimise buildings, transport networks, agriculture and manufacturing. So it has the potential to be our saving grace.

So in the short term, AI could consume resources and have a negative impact while also in the long term helping us use resources more efficiently and develop technologies that negate its short term impact.

That doesn't make it "green" though and companies should not get to say: "Yes, our data center uses crazy amounts of electricity, but our AI might eventually solve climate change."

That isn't enough. We need to do everything we can to make sure the environmental impact is as low as possible so that we actually get to see the long term potential of AI.

Hope?

Maybe things are already moving in the right direction. Microsoft says its newer AI-optimised data center use a closed-loop cooling system that consumes zero water for cooling, avoiding more than 125 million liters of water use per facility each year compared with conventional cooling approaches.

This should be the benchmark - together with renewable electricity to power those data centers. There are some things to look out for: More efficient models, smarter chips/better technology, renewable electricity, water recycling, closed-loop cooling.

Data centers built in locations where their environmental footprint is lowest - not just where companies decide is the best place for the balance sheet. And, more transparency about the resources AI really uses - so that when I ask AI next time about itself and how much impact I having on the Environment it can give me a clear factual answer.

So should we stop using AI?

Honestly? No. I don't think that is the answer - we need it. We have become dependent on it so quickly. I'm not going to pretend I will stop using something that has become so incredibly useful to me. But maybe we should be using AI thoughtfully and not ignoring or pretending it has no environmental cost. We need to be aware.

Maybe we don't need to ask ourselves whether every individual question is "worth" five milliliters of water or a tiny speck of carbon. Instead, perhaps we should ask a bigger question:

So efficiency needs to move faster than demand/adoption

Right? Because if AI becomes 10 times more efficient that is great, but if we use it 1000 times more, then we are shit out of luck. I want AI to help us build a more sustainable future - we need it! But that does not just happen magically - we need to demand requirements and safeguards and hope that our governments think the same way.

I will keep using AI.

I will probably use it more tomorrow than I did today. Maybe next year it is running my life and I don't need to open my PC (fingers crossed!). But I also want to know what that convenience costs. If we hope that AI really is going to change the world, we really need to make sure that getting there is not destroying the world while we 'save' it.

You might say that it is not up to us to solve this - but up to our governments. Since when can we trust a government to do anything right?

So maybe we can thoughtfully use AI and make the conscious decision to say: Hey it saved me 5 hours this week - I am going to go pick up rubbish for an hour on the beach. Just like AI queries - if we scale this thinking by millions - we could make an awesome start in balancing the scales.

Disclosure: I used AI to help research and organise the factual information in this article. The opinions and questions expressed here are my own.