Before I begin this edition of the Wednesday newsletter, I must address some comments from last week’s newsletter.
In last week’s newsletter, as you may recall, I talked about how I thought AI was getting a little out of control on social media. If you haven’t read it, you might want to go back and check your email and read that.
So, before we begin, let me tell you who I am. I am Michael Crose. I am a volunteer here at the Tampa Bay Technology Center. Some time ago, I started writing these Wednesday newsletters. I do a lot of research on these topics to get them done correctly. But do know this: a lot of what I write is my opinion. It doesn’t necessarily represent the views of the Tampa Bay Technology Center. I don’t write for the club officially, and I never try to be controversial. I try to give you good information about technology.
We have done more to teach AI than just about anyone else in the Tampa Bay area. We’ve had one of the foremost experts on AI at our club no less than 14 times, conducting seminars about it. And I use AI all the time. As a matter of fact, I am using AI to research what I’m going to be talking about in this edition.
The Environmental Cost of AI
So, what I want to talk about today is the environmental cost of AI. Yes, there is an environmental cost to AI. Do you think that all of AI runs on a Commodore 64 hooked up to a double floppy disk drive? No, it doesn’t. It requires a lot of servers. And when I say a lot of servers, it takes thousands and thousands of very large computer servers that are housed in very large buildings.
All AI models, especially the large language models and image generators, require data centers or server farms. These server farms are often the size of football fields—and often much bigger. They consume vast amounts of natural and industrial resources. They require tons of electricity. They emit a lot of heat. They use enormous amounts of water to cool these server farms and keep them running. So, it’s a big deal.
The Carbon Footprint
The carbon footprint of AI model training is enormous. A single AI model can emit as much CO2 as five cars over their entire lifetime. This includes the energy used during training as well as upstream emissions from power generation.
Hardware and Electronic Waste
The demand for faster, more powerful hardware means companies are constantly upgrading their equipment. They replace servers quickly and discard them on a constant basis, creating a lot of electronic waste.
Here at the Tampa Bay Technology Center, we have repair people who bring in computers that are just gone. Right now, the last time I was at our location, I saw a pretty good-sized pile of dead computers that needed to be recycled. We don’t throw those into dumpsters—we take them to an electronic waste recycling center. It’s the right thing to do, though I feel guilty about even the 10 computers we have to recycle. But what are these big server farms doing with their massive amounts of electronic waste?
Environmental Impact Beyond Energy
All of this doesn’t stop with electricity and water usage. You have to mine rare earth minerals for the chips, batteries, and server components. This mining causes deforestation, water pollution, and habitat destruction.
And this is only going to get bigger. There’s a global push to expand AI capabilities, and major tech companies’ AI divisions consume unsustainable amounts of water and electricity. Many of these claims about environmental responsibility rely on carbon offsets rather than actual reduction of resource consumption.
A Call for Awareness
So the carbon footprint is significant. The next time you go to your favorite AI—which, by the way, I do just about every day (a lot of people are not using Google now; they’re using AI to look things up)—you might want to think a little bit about what this is costing us and what it’s going to cost us in the near future.
Again, I don’t have an answer for this problem. I wish I did. But it is something for all of us to think about.