File – The data center owned by Amazon Web Services, next to the Susquehanna Nuclear Power Plant in Berwick, Pennsylvania on January 14, 2025, is under construction next to the Susquehanna Nuclear Power Plant (AP Photo/Ted Shaffrey, file)Marissa Loewen first began using artificial intelligence as a project management tool in 2014. She said she has autism and ADHD, which would be very helpful in organizing her ideas.
“We are aware that it has an impact on the environment, so we are trying to use it conscientiously,” she said.
Her personal AI use is no longer unique. Now, this has become a feature for smartphones, search engines, word processors and email services. Every time someone uses AI, it uses the energy that is often produced by fossil fuels. It releases greenhouse gases into the atmosphere, contributing to climate change.
And it's becoming harder to live without it.
Climate Cost AI is mainly equipped with a data center that stores field queries, data and deploys information. When AI becomes ubiquitous, data center power demand increases, leading to issues with grid reliability for people living nearby.
“We are trying to build data centers at a pace that cannot integrate more renewable energy resources into the grid, so most of the new data centers are equipped with fossil fuels,” said Noman Bashir, Computing and Climate Impact Fellow at MIT's Climate and Sustainability Consortium.
Data centers also rely on fresh water to keep them cool, as they generate heat. According to an article from the Institute of Environment and Energy, large centers can consume up to 5 million gallons (18.9 million liters) per day. This is roughly the same as the daily water needs of a town of up to 50,000 people.
It's hard to imagine because the impact is invisible for most users, said AI and climate led AI company Sasha Lucciioni to hug Face.
“One of my research found that generating high-resolution images uses as much energy as charging half of a phone. And “That's not correct, because using MidJourney (generating AI program) won't bring down the phone battery,” she said.
John Ippolito, a professor at New Media at the University of Maine, said tech companies are constantly working to make chips and data centers more efficient, but that doesn't mean that AI will reduce the environmental impact. That's because of a problem called Jevons Paradox.
“The more cheap resources you get, the more you tend to use them anyway,” he said. He said commute times did not decrease when the car changed horses. We traveled far away.
Quantifying AI Footprints How much these programs contribute to global warming depends on many factors, such as the outside of the data center that handles queries, the clean grids, and the complexity of AI tasks.
The sources of AI's contribution to climate change are incomplete and contradictory, making it difficult to obtain accurate numbers.
But I tried Ippolito anyway.
He built an app that compares the environmental footprints of various digital tasks based on the limited data he could find. Simple AI prompts like “Tell me about the French capital” estimate that they use 23 times the energy of the same question entered into Google without the AI overview feature.
“Instead of using existing materials, I write them from scratch, and that requires more calculations,” Lucciioni said.
And it's just for a simple prompt. Complex prompts like “Please tell me how many gummy bears can fit in the Pacific Ocean” use 210 times more energy than a Google search without AI. According to the Ippolito app, the 3-second video uses 15,000 times more energy. This is equivalent to turning on the incandescent bulb and leaving it alone for more than a year.
It has a big impact, but that doesn't mean that the tech footprint was carbon-free before AI entered the scene.
For example, watching Netflix for an hour uses more energy than complex AI text prompts. For 1 hour zoom with 10 people, use 10 times that.
“It's not only bringing people to the impact of AI, it also brings all of these digital activities that we take for granted,” he said.
Limiting Tech's Climate Impact Ippolito said it would limit AI use whenever possible. He proposes using human-captured images instead of AI-generated images. He tells the AI to stop generating as soon as he gets the answer so that he doesn't waste any extra energy. He requests a concise answer and starts a Google search by typing “-AI”, so it doesn't provide an AI overview of the queries that he doesn't need it.
Lowen takes the same approach. Instead of asking a series of repetitive questions, she said she would try to organize her thoughts into one AI query. She also built her own AI that doesn't rely on large data centers. This saves energy in the same way you watch movies you own on DVD.
“By having something local to your computer at home, you can also control your electricity and consumption usage, which gives you a little more control over your data,” she said.
Lucciioni uses Ecosia. It is a search engine that uses efficient algorithms, uses profits to plant trees, minimizing the impact of each search. The AI function can also be turned off.
Also, ChatGPT has a temporary chat feature, which removes queries that you send to the data center in a few weeks, instead of occupying the storage space in the data center.
However, AI is a small part of the energy use of data centers. Ippolito estimates that around 85% of data collection is collected from sites such as Tiktok, Instagram and Cryptocurrency.
His answer there: Uses the screen time limit on your phone to limit scrolling on social media. Less time means that personal data is collected, less energy and water is used, and less carbon emissions enter the atmosphere.
“If we can cut the data center from the equation, I think it's a victory,” Ippolito said.
