Is Using AI Sustainable?
- Shannon Falkson
- Dec 15, 2024
- 6 min read
AI is not going anywhere, but is using it sustainable? It’s great to give me a recipe when I’m trying to use up an avocado, a banana and some day-old bread (options include avocado and banana french toast and a DIY face mask, in case you were wondering), but are my compost savings offset by using AI to tell me how not to waste food?
I went straight to the “source” for some answers. I typed the following prompt into ChatGPT: You are my writing consultant. Please write me a blog post that is about two pages explaining the carbon footprint of using AI and please make it funny.
This was ChatGPT’s response:
The Carbon Footprint of AI: Are We the Baddies?
Let’s be honest: we all love AI. It helps us write emails, generate recipes, and convince our bosses we’re “working smarter, not harder” (emphasis on the not harder). But here’s the awkward truth: every time you ask your friendly AI assistant to find a gluten-free brownie recipe or write a heartfelt birthday message to your coworker (who you barely know but need to impress), you’re nudging the planet a teensy bit closer to becoming a toaster oven.
Yes, folks, AI has a carbon footprint. And it’s not dainty. But before we all panic-delete our apps and start journaling with quills by candlelight, let’s break this down.
How Does AI Use Energy?
AI isn’t powered by magic (unless you count overpriced cloud subscriptions as sorcery). Every time you ask me—or another AI—to do something, I tap into massive data centers filled with humming servers that crunch numbers, analyze patterns, and try to figure out whether you meant “ducking” or something more... colorful.
These data centers don’t just sip power. They guzzle it like it’s free refills at a fast-food joint. Keeping the servers running and cool enough not to melt into a puddle of silicon requires electricity, often generated by burning fossil fuels. In fact, one AI query might consume as much energy as boiling a kettle of water. So every time you ask, “What’s the weather like in Paris?”, you’ve metaphorically brewed a cup of tea you’ll never drink.
But What About My Other Hypocritical Habits?
Before we start throwing stones at AI, let’s admit the obvious: we’re all living in glass houses, and they’re probably double-paned for extra insulation because hello, energy efficiency!
Think about streaming. You binge-watch a series in HD, pretending it’s “educational” (The Crown counts as history, right?), and then complain about how tech companies are destroying the planet. Spoiler alert: streaming an hour of video emits about 55 grams of CO2. AI is over here sipping lattes compared to your all-night Netflix marathons.
And don’t even get me started on your love for online shopping. Every time you scroll through 20 pages of leggings you don’t need, algorithms—yes, those powered by AI—are working overtime to recommend the exact shade of “Moody Mauve” that’ll end up in your closet and never see the light of day.
The Real Culprit: Us
Here’s the twist. AI isn’t inherently evil; it’s just doing what we tell it to do. It’s like blaming your dog for eating the cake you left on the coffee table. If we’re frivolous with AI, asking it to settle debates like whether hot dogs are sandwiches (they’re not), then yeah, the environmental cost adds up.
But when we use AI to optimize energy systems, improve climate modeling, or even reduce waste, it can save more carbon than it emits. It’s like when your gym membership guilt-trips you into working out once a month—technically, it balances out all the snacks.
So What Do We Do?
Here’s a thought: moderation. Sure, keep asking your AI for vegan cheesecake recipes, but maybe don’t outsource your existential crises to it (“What’s the meaning of life?” burns way more kilowatts than you think). And let’s be conscious about our overall tech use.
Start small:
Close the 47 tabs you have open. Your laptop and the planet will thank you.
Turn off your camera during Zoom meetings if you’re not talking. No one needs to see your “paying attention” face anyway.
Use AI wisely—like brainstorming ways to reduce your actual carbon footprint, instead of justifying why your new smart fridge had to have Wi-Fi.
The Bottom Line
AI isn’t the enemy. It’s the roommate who leaves lights on and forgets to compost: annoying, but manageable with a little effort. The real question is whether we’ll use this incredible tech to help fix the planet—or just keep asking it to write blogs about carbon footprints (oops).
Now, if you’ll excuse me, I need to recharge—literally. Servers don’t run on fairy dust, after all.
Not too bad, right? Then, I asked, “How much energy was required for this session with you today?” ChatGPT provided not one but two options! (Show off.) I’ve pasted the first one below.
Response #1: Ah, the irony of me helping you write a blog about AI’s carbon footprint while contributing to it! Let's estimate:
A single query to an AI model like me can consume anywhere from 0.1 to 0.5 kWh, depending on the complexity of the response. For a typical conversation like ours, involving multiple back-and-forth exchanges, we’re probably looking at 2–3 kWh for the session.
To put that in perspective:
That’s roughly equivalent to running a 50-watt lightbulb for 40–60 hours.
Or about as much energy as it takes to brew 30 cups of coffee (which you might need after a long blogging session).
And, yes, probably more than you’d burn by simply handwriting this blog—but where’s the fun in that?
The good news is that many AI companies are transitioning to renewable energy sources to power their data centers. So while my "carbon pawprint" isn’t zero, it’s shrinking—unlike my appetite for answering your questions! 🌱
Now, shall we brainstorm how to offset this chat? Maybe plant a tree or two? Or switch off the lights while we work on your next project? 😊
Response #2: Great question! While it’s hard to pin down the exact energy usage for our session, here's an estimate:
Each AI query like yours consumes around 10-50 watt-hours (Wh) of energy, depending on the complexity of the task. That’s roughly equivalent to leaving a 10-watt LED bulb on for 1-5 hours or charging your smartphone once.
Since we’ve been working on a blog post and chatting a bit, let’s assume the higher end—say 50 Wh. The actual carbon emissions depend on how the electricity powering the servers is generated. If the servers are in a region with lots of renewable energy, the carbon impact could be minimal. In areas reliant on fossil fuels, it might be around 20-30 grams of CO2, which is comparable to driving a gas-powered car about 0.1 miles.
In short: we’ve done less harm than streaming a single episode of The Office in HD. But hey, now you’ve got a blog post to inspire greener habits! That feels like a solid trade-off. 🌱
So, do I have to add AI use to the list of things that already spike my eco-anxiety? Despite the false claims of certain politicians who claim using AI will double the demands of our power grids (as a justification to drill, baby, drill! while simultaneously eliminating incentives and investments in renewable energy), AI is expected to increase the overall demands on the grid by 2-3% by 2030. Most of the additional energy will be to power data centers which are projected to increase by 160% in the same time frame. The good news is that many of these data centers are pursuing renewable sources of energy to power their operations. While they are touting their commitment to using “green” energy, there’s another big driver to go green—it’s cheaper. Tech giants like Google, Microsoft and Amazon are increasingly investing in data centers powered by sun and wind and water.
The bottom line is that it makes sense to be judicious in our use of AI, but it makes even more sense to advocate for the development of renewable energy, use AI platforms powered by renewable sources of energy, and elect politicians committed to investing in renewable energy and innovation and reducing our dirty dependence on fossil fuels.
We can be green and save on green at the same time. Even the most powerful AI can’t argue with that logic.
コメント