Whenever a new groundbreaking technology comes to the masses, we tend to look at only the positive aspects and how it is changing the world in a good way. The same story goes for Artificial Intelligence (AI). In the last five years, we have seen Generative AI tools like ChatGPT, Gemini, and DeepSeek totally transforming the way people write and create content. Then there are the advanced Machine Learning (ML) algorithms that are transforming how we view content. But we rarely discuss the environmental impact of AI.
Keeping that in mind, let this be a guide to understand the environmental impact of AI and how it is going to affect us in the long run.
What is AI Doing to the Environment?
Use of AI and its impact on the environment is a hot-button issue for sure. People in support will tell you how AI can help combat climate change, optimize energy usage, and make better environmental decisions. While this is all true to some extent, AI, especially large-scale AI, comes with a significant environmental cost. Unfortunately, that aspect is largely hidden from the popular outlets, and we don’t know how AI affects the environment.
Exploring the Environmental Impact of AI
Let’s take a look at some aspects in which AI is affecting the environment:
AI is Always Hungry
A very common statement regarding AI use is that it helps you save energy and makes you efficient. If you are saving energy, it does not mean the energy isn’t being used. It has to come from somewhere else. Training AI models, especially the massive ones like GPT-4, Claude, and Gemini, requires immense computational power.
If we talk about numbers, as per OpenAI, GPT-3 consumed about 1,287 MWh (megawatt hours) of electricity during training (Source).
Another stat will help you put this into context. A study by the University of Massachusetts Amherst estimated that training a single AI model can emit as much carbon dioxide as five cars over their entire lifetimes (Source).
Now, you have to understand that this is just the training part. Then there is the inference part where the model actually answers the queries. That is another energy-hungry process, and it will get more hungry as millions of users interact with AI systems daily.
More Context
A simple AI-powered search on ChatGPT consumes around 10 times more energy than a traditional keyword Google search (Source).
The Carbon Emissions
It’s not just about the energy appetite of AI. There is the aspect of carbon emissions as well, and it matters even more.
If the electricity powering data centers comes from fossil fuels (coal, natural gas, etc.), the carbon emissions are proportionately higher.
If we talk about numbers, training a Large Language Model(LLM) emits about 284 tons of CO2. To put it into context, each passenger on a trip from New York to London is liable for emitting around one ton of CO2. So, training one LLM is equal to 284 passengers going from New York to London on an airplane.
Again, that is just the training phase. If we talk about regular use, then the carbon emissions go higher. So every time you:
- Ask ChatGPT a question,
- Ask ChatGPT to generate Ghibli-style art,
- Generate an AI image via Midjourney or DALL-E,
- Use AI for code generation…
… you are indirectly contributing to carbon emissions.
In short, AI may be digital, but its environmental footprint is very real.
Usage of Water by AI
Whenever we talk about the environmental impact of AI, we often overlook water usage when running AI farms. Water is heavily used to cool the powerful servers in massive data centers.
Without cooling, the hardware would overheat and malfunction. According to a report, training GPT-3 may have consumed 700,000 liters of clean, drinkable water (Source).
To put it into context, that’s enough water for about 300 people to survive for a year! Now, it may not sound much like it’s water for just 300 people for a year, right? I mean that is a medium neighborhood. But the thing we have to understand is that it is just for the training of such models.
The server farms always require water to keep the systems cool. As more and more AI models come, the usage of water will increase in such server farms.
The Paradox
The environmental impact of AI is quite wide and has multiple aspects to it. On one side, AI is truly transforming the way we work towards climate change. Some examples of how AI is helping to fight climate change include:
- Optimizing energy grids.
- Predicting extreme weather patterns.
- Monitoring deforestation and ocean health via satellite imaging.
- Designing new eco-friendly materials through AI-driven simulations.
But on the other hand, we have these issues associated with AI use:
- Increasing energy demands.
- Boosting carbon emissions.
- Straining water resources.
So it is really difficult to assess the net impact of AI on the environment. Good or bad is a very complex thing to say in this context, as there are many variables involved.
One could say that the use of AI is good if the energy source is renewable. Then there is the topic of how efficiently models are trained and deployed. As the technology is advancing, we have seen positive results. For instance, DeepSeek was trained on a significantly smaller data set and much efficiently.
Also, AI in its true form has been around for a limited time. Its progress might seem exponential, but its time period is relatively small. So convincing evidence to assess its true environmental impact is not enough at the time. With more time and research, we should see more solid evidence that would clear the environmental impact of AI in a better way.
What are the Companies Doing?
It is quite easy to point at the user and say, “Oh,! You are using AI, you must hate the environment.” But we should also look at the accountability of the companies that are providing these AI tools to the people.
Big tech obviously knows about the criticisms related to the environmental impact of AI, and they are taking some steps. First, Google says that it has been carbon neutral since 2007.
Now it aims to run on 24/7 carbon-free energy in all data centers by 2030. Microsoft pledged to be carbon negative by 2030. OpenAI knows about the environmental impact but largely relies on Microsoft Azure’s sustainable infrastructure. Companies like Meta are investing in wind and solar projects to power data centers.
Be Skeptic
One thing to learn over time is that the Big Tech companies aren’t very open and transparent when it comes to how their technology works, and the same applies to AI. Many critics say that carbon offsets (buying credits) aren’t the same as cutting actual emissions. Unfortunately, many companies simply buy credits. Also, the transparency around AI-specific energy usage remains limited.
AI is Bad for the Environment, but Is it the Full Picture?
A few days ago, when Ghibli-style images were flooding the internet, many people raised different issues regarding the trend. However, one that was relevant to today’s guide was the outrage against ChatGPT’s energy consumption and how it is bad for the environment. The outrage was a bit odd, and here is how.
A person saying that ChatGPT is bad for the environment and posting that using your smartphone, which is powered by a cobalt battery, while binge watching Adolescence on Netflix in 4K as you order from Amazon is just odd. There is no doubt that training and running AI uses a lot of energy, but let’s put it into context to get a broader picture.
If you watch a single season of a popular TV show, roughly 10 hours in 4K, then that can consume 5000 watt hours. Now, one can easily do around 40K ChatGPT queries with this much energy.
In fact, global streaming emissions are around 300 million tonnes of carbon dioxide per year(Source). For context, that is equal to Spain’s total annual emissions. Then there is Bitcoin mining that consumes more energy than Argentina in a whole year.
Playing Call of Duty for one hour can use around 450 watts per hour of energy, and that is easily comparable to a ChatGPT farm.
Now, everything that I have put here is an activity that you do for pure indulgence, except for Bitcoin mining. On the other hand, AI tools like ChatGPT are truly transforming the way people access and consume information.
The Real Culprit
Again, we come to Big Tech, especially social media giants. For instance, Meta and Google use more AI computing power for targeted ads than ChatGPT does for answering queries. They have these algorithms that track every click, scroll, pause, and then use enormous GPU clusters that go through electricity just like this. So why this selective outrage against ChatGPT?
Wrapping Up
Well, that was a heavy one. At the end of the day, the environmental impact of AI is a complex, layered conversation. As humans, we have this tendency to view everything in black and white, but that is rarely the case.
The issue of the environmental impact of AI deserves nuance rather than knee-jerk outrage. Yes, AI models consume a massive amount of energy, water, and other resources. And yes, there is an undeniable carbon footprint attached to every chatbot conversation, image generation, and smart search.
But AI also holds extraordinary potential to address some of the very environmental challenges it contributes to, from optimizing clean energy grids to innovating sustainable solutions across industries.
Moving forward, the real question isn’t whether AI is inherently bad for the environment, but how we users, developers, and corporations alike can build, use, and regulate it more responsibly. We hope you like this guide and it helps you understand a different yet educated perspective.
