How Much Energy Does GenAI Really Use?

Part 1: Examining Generative AI

Photo by Marcin Jozwiak

I am against Generative AI. Let’s make that explicitly clear before we get started. I’ve used Canva’s GenAI once to produce an image when it first released. I had yet to know or understand its ethical or environmental implications (and it didn’t produce what I wanted anyway).

Now that that is out of the way, this will be a series of articles on Generative Artificial Intelligence. I’m going to examine environmental and ethical factors as well as it’s impact on critical thinking skills and creativity. This first article is all about GenAI’s impact on the environment.

GenAI’s Carbon Footprint

If we’re already using digital technology — the internet, phones, tablets, computers, video conferences — daily, what’s the harm in using GenAI? We’re already destroying the planet by using up its natural resources, who cares if it happens even faster (I’m being sarcastic if that wasn’t clear).

I’ve heard this argument before (though I’d say it’s more of an excuse). The issue is GenAI isn’t like our other technology. Moreover, it’s growing at a rate no one was prepared for.

I want to take a look at our average internet use and how it creates carbon emissions to paint the full picture.

How does the internet use natural resources and energy?

The internet exists through computer servers hosted in data centers. These centers take up land. The average onsite data center has between 2,000 and 5,000 servers, with a total square footage between 20,000 to 100,000 square feet (2024). That’s trees being cut down and animal habitats being destroyed.

Then, of course, there is the electricity to consider. Electricity is most often made using coal or natural gas, though it can be made using wind, solar, and other greener technologies. These data centers also need to be kept cool to prevent the servers from overheating. This is done through the use of water and air conditioning (2023). Data centers are effectively flattening nature, using up water, and polluting the air with carbon emissions. There isn’t any natural, finite resource that isn’t affected.

Furthermore, Lorena Regattieri, a senior Trustworthy AI fellow at Mozilla Foundation, reminds us of the cost of creating the computers and servers themselves: “The extraction of minerals like lithium, cobalt, and rare earth elements for technology components often occurs in regions with environmental and social concerns.” She adds that “These issues disproportionately impact marginalized communities, particularly in the Global South, where mining operations can harm local ecosystems and communities.” (2023).

That’s the Global South…but what about locally?

Purdue University conducted a study that concluded a one-hour Zoom call produces between 150 to 1,000 grams of carbon dioxide. This can be reduced by 96% by going audio-only, but now is a good time to recall the exponential rise in Zoom calls during the COVID-19 pandemic.

It’s also worth noting that in 2018, China, the United States, and India accounted for nearly 70% of the rise in energy demand around the world, with the United States seeing the largest increase in oil and gas demand worldwide. I wonder what that looks like now, five years later?

How much energy is GenAI consuming?

I don’t have the numbers for this year, but I do have some for 2021–2023:

• In 2021, scientists from Google and the University of California at Berkeley estimated the training process alone consumed 1,287 megawatt hours of electricity — enough to power about 120 average U.S. homes for a year — generating about 552 tons of carbon dioxide.

• The power requirements of data centers in North America increased from 2,688 megawatts at the end of 2022 to 5,341 megawatts at the end of 2023, partly driven by the demands of generative AI.

Additionally, a 2023 Scientific American article wrote that, even with the limited data on the carbon footprint of a single GenAI query, some industry professionals estimate it to be four to five times higher than that of a search engine query. Another 2023 article, this one from the Columbia Climate School, stated that the world’s data centers account for 2.5 to 3.7 percent of greenhouse gas emissions, which is greater than even the aviation industry.

Text vs Images

Now, because I’m autistic and when I get into researching something, I go all in, I also looked into the individual energy consumption of GenAI text versus images.

An MIT Technology Review article cited a study that found that generating an image using AI uses as much energy as fully charging a smartphone. On the other hand, they found that generating text, even 1,000 times, is the equivalent of charging a smartphone to only 16%.

What I found most interesting was that smaller AI models designed to generate specific tasks used less energy than a larger, more general-use AI model. The interesting part is that OpenAI’s ChatGPT and other powerful AI models are designed to handle every type of query for “convenience.” Why have multiple models to serve different purposes when you can have one model that does everything you need it to?

Once again, I see history repeating itself. Companies continue to choose short-term profits over long-term sustainability — and they’re risking the health of the planet to do so.

Wrapping Up

I work remotely and mostly from home. I use my phone, tablet, and computer every day. This is partly how I make a living, by writing, so I won’t say I’m perfect and that I’m not contributing to the planet’s carbon footprint because that would be a lie. I will say, however, that I can choose not to increase my part in that footprint. I can choose not to make things even worse than they already are.

I do this by choosing not to use Generative AI: not as a text generator, or an image generator, or a search engine. That is my informed choice. I have the facts now, I have the research and data to back up my stance on AI, and if you’ve made it to the end of this article, so do you. So, I ask you this: What do you choose? Are you for or against GenAI? A combination of both? Leave a comment below and let’s start a healthy and civil dialogue.