A recent survey by Elon University has shown that, since the rapid rise of Artificial Intelligence (commonly known as AI) around 2023, 52% of adults in the United States use AI. Of that 52%, 34% use AI at least once a day, and 10% have said they use it “almost constantly.” While some use it for work, others simply see it as a replacement for Googling something.

All this increased use of AI adds up—just one generative AI search results in 4-5 times the energy consumption of a normal search engine. The use of AI billions of times per day leads to a massive amount of energy and storage consumed in data centers. Stored all around the country, data centers are temperature-controlled buildings that house data from servers, network equipment, data storage drives and user-generated content like browsing history, including any use of AI. 

AI’s energy use is often divided equally between data processing, model training and inference, which is the process of predicting the outcome of new data by using a trained model. But the computing hardware inside the data centers itself also requires energy.

A graphics processing unit (GPU), which is a type of powerful processor often used to store data for AI models because of its ability to handle lots of data, has a carbon footprint constituted by emissions from product and material transport. Getting the raw materials to actually create GPUs also requires toxic chemicals for processing and involves harmful mining techniques, including soil and water pollution. 

While many people understand the impact toxic chemicals have on the environment, they are likely unaware that creating, testing and manufacturing AI models requires an immense amount of energy. In other words, it’s not just the use of AI that is creating environmental consequences, it’s also the AI models’ development.

According to a 2021 research paper by scientists at Google and the University of California,  Berkeley, the estimated amount of electricity it takes to train an AI model is 1,287 megawatts of electricity, which is the same amount of energy it takes to power 120 homes for a year. 

Because new AI models are released every few weeks, the energy used to train the previous versions is wasted, and even more energy is used to run the new models. These are more “high-tech” and use more data, which requires increased space in data centers that are being built more rapidly and at a higer rate than ever before.

Amazon alone has over 100 worldwide data centers, with around 50,000 servers per data center that support cloud computing services. Scientists estimate that data center use in North America went from 2,688 megawatts in 2022 to 5,341 megawatts in 2023, mostly because of growing AI use in the country. While AI is not the sole reason for the growing energy demands of data centers, it plays an impactful role.

Aaron Olsen, the Health Occupations teacher at Ida B. Wells High School, has personally seen the data centers. He describes them as “acres of these buildings…these big warehouses filled with all that space to store information.” There are 137 data centers listed in Oregon, and with increases in AI use, the U.S. energy demand is up by 20%, meaning that those numbers are expected to jump in the next few years. 

Part of that energy demand includes water use—data centers use significant amounts of chilled water that absorbs the heat from the equipment and cools it off. Rough estimations predict that data centers require 2 liters of water for every kilowatt hour of energy consumed by the centers. 

Abby Griffin, an English teacher at IBW, comments on the water use of AI. “They use a bunch of water to cool down the servers that have to be running…to respond to prompts to do the AI thing,” she says. “So water usage, right? The water bills of the community go up.”

All this energy consumption is a result of the growing use of AI in billions of people’s day-to-day lives. Everyday users of AI don’t think much about the role they play in AI energy consumption, simply because AI is so easily accessible. “I just think it’s such an easy tool, and it’s easy access,” says Olsen. “I mean, it’s just too simple.” 

Lack of information about AI’s energy consumption doesn’t help, either. Most people think of AI as a disembodied intelligence existing solely in the cloud, and having no physical ties to the real world. But the truth is that generative AI has a real-world impact that threatens our environment and our society.

“I mean, if the [fact that] you are literally getting dumber by using AI isn’t enough, I do hope that the environmental impact can get there,” Griffin concludes. “We’re using finite resources and polluting our land so that people can do as little thinking as possible.”