AI Energy Use in Everyday Terms

Comparing chatbot energy consumption to the things people actually use and understand

I’ve discussed AI energy use from several perspectives in previous posts (here, here, and here), but this time I wanted something simpler and easier to grasp. What if we compared the energy used by a chatbot over an entire year to everyday things we all do?

Here’s the quick takeaway:

One year of regular chatbot use consumes less energy than:

🚗 Driving a car for 10 kilometers
🚿 Taking five hot showers (five minutes each)
🛁 Filling two hot baths

Or, to put it even more visually:

(Icons from freepik via flaticon)

Now, let’s dig into the details!

I’ve previously written about chatbot energy use, and while exact numbers can be tricky to find, the best estimate I came across is that a typical, text-only interaction with a GPT-4-like chatbot (the kind most of us regularly use) consumes about 0.2 watt-hours (Wh) of energy. For simplicity, let’s assume 100 interactions per day - this could mean something like 10 chats, each involving about 10 back-and-forth messages. Doing this every single day would add up to around 7.2 kilowatt-hours (kWh) over one year.

Is that a lot? Well, let’s take a closer look by comparing it to everyday things we do. Imagine going for a car ride. Let’s assume a regular gasoline-powered car that uses about 7 liters per 100 kilometers. For a 10-kilometer ride, that’s around 0.7 liters of gasoline. Each liter of gasoline contains roughly 9.7 kWh of energy, meaning our short trip uses about 6.8 kWh. Adding in real-world factors like acceleration or stop-and-go traffic bumps that number up slightly, so let’s round it to around 7.6 kWh. Transportation is definitely energy-intensive! Of course, if you’re using an electric car, the energy needed would drop - probably about four times less. That means you’d get about 40 kilometers of driving for the same energy. But since most cars still run on gasoline, let’s stick with that for our comparison.

What about hot showers? A five-minute hot shower uses about six to seven liters of water per minute, totaling roughly 30–35 liters per shower. Let’s go with 32.5 liters as an average. To heat this water, we typically raise its temperature by about 30°C. Converting that into energy use, we get around 1.13 kWh (32.5 L × 30°C × 4.187 J/mL/°C) But that’s in an ideal world. In real life, water heaters aren’t perfect; some heat always escapes. When we consider these losses, a five-minute hot shower actually uses around 1.6 kWh of energy. That means five showers add up to about 8 kWh in total.

If you’re more of a hot-bath kind of person, I have some bad news. Heating a bath uses much more energy compared to showers. Even though the energy required per liter is the same - assuming a temperature increase of about 30°C - a bath typically uses a lot more water. For example, even a small bathtub filled with around 80–100 liters, considering typical inefficiencies and heat loss, easily consumes around 5.2 kWh of energy per bath. So just two baths already put you over 10 kWh!

The numbers for the car, shower, and bath examples have some uncertainty, but it’s not huge. You could argue a bit either way, maybe plus or minus 20%, but the overall picture remains pretty similar. With AI energy use, things get a little trickier. As I mentioned before, my previous estimate of 0.2 watt-hours per chatbot request is based on what I think is the best available data right now. But I’m open to hearing other views or better numbers.

Of course, if most of your interactions involve very large AI models that reason for several minutes, specialized AI agents, or image and video generation, energy usage will jump significantly. But I wouldn’t classify these as “typical” chatbot interactions. I’d be happy to do a detailed estimation on these scenarios another time.

For now, this is a rough comparison that helps us keep AI energy use in perspective - especially next time someone says chatbot use is causing an energy disaster. Consider walking instead of taking that next short car ride, or choose showers over baths. Even shortening your showers slightly a few times per year can easily offset your entire chatbot-related energy footprint.

CODA

This is a newsletter with two subscription types. I highly recommend to switch to the paid version. While all content will remain free, all financial support directly funds EPFL AI Center activities.

To stay in touch, here are other ways to find me:

• Social: I’m mainly on LinkedIn but also have presences on Mastodon, Bluesky, and X.

• Podcasting: I’m hosting an AI podcast at the EPFL AI Center called “Inside AI” (Apple Podcasts, Spotify), where I have the privilege to talk to people who are much smarter than me.