Generative AI data centres are bad for the planet

Generative AI requires large data centres, that consume incredible amounts of energy and water. Not only are the running costs for these data centres high, the cost to the environment is staggering too. So, what is it costing the planet when you ask gen-AI a question?

Using generative AI is unsustainable and the tech companies know it

Rumours fly about the environmental cost of using Large Language Models (LLMs). All we really have are estimates, because the tech companies do not properly share their consumption and emission rates in their sustainability reports. They purposely hide them, but more on that later.

Researchers say that asking ChatGPT one question has the same environmental impact as five simple web searches. At least, that was the case before Google started adding a gen-AI answer at the top of its results pages. So if you want to do a simple search with a lower environmental impact, you are better of not using Google. 

The tech companies behind LLMs do not want you to think about the impact of them training and retraining their models, or the impact of image generation. But recognising that there is a problem is the first step to finding a solution. Time to look at the damage that generating content does.

Gen AI need special data centres that use a lot of energy

Genereative AI models do not exist in ‘the cloud’, they need physical data centres to exist. And not just any data centre, no they need special data centres. Noman Bashir, a Computing and Climate Impact Fellow at the MIT Climate and Sustainability Consortium, told MIT News that generative AI data centres require “seven or eight times more energy than a typical [data centre]”. 

MIT News also points out that in one year’s time (2022-2023) the power requirements for data centres in North America have nearly doubled. Bashir continued to MIT News: “The demand for new data centres cannot be met in a sustainable way. The pace at which companies are building new data centres means the bulk of the electricity to power them must come from fossil fuel-based power plants.”

Google and its competitors are in a data centre arms race.

While not all data centres are for generative AI, the increase in construction of data centres is equal only to the increase in popularity and availability of generative AI. Aaron McGarry, a senior director at Google, even called it a “data centre arms race”, according to the Roanoke Rambler. So while data centres have been around for almost 80 years, they are a problem now.

In 2022, MIT expected data centres to consume 1050 terawatt hours of energy in 2026. The enormous energy consumption of generative AI data centres is leading to a lot of pressure on electrical grids around the world. Grids that are often already full to the point where newly constructed homes cannot be connected to them.

In the Netherlands, for example, in March 2024 it could already take up to a year for a new home to be connected to the grid. The situation has not improved since. And the housing crisis that the country faces is only getting worse, partially due to the energy grid being so congested. Meanwhile, data centres use the same amount of energy as 2 million households in the country of 8.4 million households (18 million people).

The water usage by AI data centres exceeds that of whole countries

The companies behind generative AI want to hide how much water their data centres use. They believe it is proprietary information, as they argued in front of a Roanoke, Virginia, judge after Henri Gendreau took them and the Western Virginia Water Authority to court under the Freedom of Information Act.

According to Bashir, two litres of chilled, clean water is needed for every kilowatt hour of energy a data centre uses to cool down the hardware. A study estimated that in 2025, AI data centres used 312.5-764.6 billion litres of water. If we take the average of the two, than, according to Worldometer, AI data centres use the same amount of water as the country of Slovakia did in 2016.

The amount of drinking water that is used by data centres is not only a problem due to it being a limited resource, the water can also not be reused safely. Reusing the water used by data centres can have detrimental effects on the environment as was shown in Morrow County, Oregon by Rolling Stone and the FERN.

Two-thirds of data centres are being built in areas with existing water shortages.

According to experts, Amazon’s data centres in Morrow County worsened the process of rising nitrate levels at the local water source. The water source—that 45,000 people depend on—now tests at 40 to 70 parts of nitrate per million. The national safety limit in the United States is 10 ppm, levels higher than that have been linked to miscarriages, kidney failure, cancers and other medical issues, as the people of Morrow County are now well aware of.

Furthermore, according to Bloomberg, two-thirds of new data centres built in or since 2022 in the United States are in areas that already face water shortages. A report by Ceres found that the data centre arms race “could increase water stress in already strained basins by up to 17% annually” and even more in peak seasons. Which makes it even more problematic that the water used by AI data centres cannot be reused safely.

CO2 emissions by AI data centres

Sadly, the water and energy consumption of AI data centres are not the only issues that make generative AI unsustainable. They also emit large quantities of CO2. A December 2025 study by Alex de Vries-Gao, estimated that “the carbon footprint of AI systems alone could be between 32.6 and 79.7 million tons of CO2 emissions in 2025”.

According to The Guardian, that is the same as all of New York City. They also reported that the largest data centre that is being planned for the UK will produce the same amount of CO2 as 24,000 homes. But much like with the water usage, there is no real emissions data available because the environmental disclosures by tech companies just do not contain this information.

Why you should care

The lack of sustainability of generative AI tools can feel like a distant problem, but it is causing real world issues everywhere. And it is not the tech companies that will pay the price for all these issues, it is society as a whole. Energy prices will rise, as will the cost of drinking water. And so will the price of your next laptop and smartphone, according to the International Business Times we can expect a price increase of 15-20%.

AI data centres not only need energy and water to run, they also need high-bandwidth memory. Memory manufacturers are now prioritising those over regular DRAM which are needed for new laptops and smartphones. But the demand for regular DRAM is not falling, in fact as new laptops are nearly always equipped with an AI agent, they require even more memory than before.

While Sam Altman, CEO of OpenAI, may want you to stop saying ‘please’ and ‘thank you’ to ChatGPT because it is costing the company millions, as they do have to pay for the water and energy they use, just not the issues that come from their usage. I am asking you to stop using generative AI completely, because it is destroying our planet, the only home we have.