Apple’s new custom emoji come with climate costs

Soon, you’ll be able to create you own emoji and generate custom images with new AI features on Apple devices. I hate to be the bearer of bad news (and I’m guilty of grossly overusing emoji myself) — but those images could have the biggest environmental impact of all of Apple’s new AI tools.

Generative AI is notoriously energy-hungry, which means it leads to a lot of planet heating greenhouse gas emissions. Image generation, in particular, is one of the worst offenders. The AI arms race has already started to grow some companies’ carbon footprints. Now, new fruit emoji varietals and sketches generated by AI are likely to add to the problem.

Apple announced its new suite of AI features during WWDC 2024 yesterday under the banner “Apple Intelligence.” There are all sorts of ways Apple is injecting AI into everyday screen life, from sorting notifications and helping users write emails to making Siri smarter. Apple also launched a partnership with OpenAI to bring ChatGPT to iOS, macOS, and iPadOS “later this year.” Its AI image features, Genmoji and Image Playground, are expected to come with iOS 18, iPadOS 18, and macOS Sequoia.

Apple didn’t immediately respond to questions from The Verge about exactly how much energy it expects these new features to use or how this all jives with its climate goals. But each of those moves comes with energy and climate costs.

Data centers used to train and run AI models have an even more voracious appetite for electricity than other data centers used to send and store the world’s digital messages and cat photos. Research from AI firm Hugging Face and Carnegie Mellon University found that large image generation models used far more energy and created more than greenhouse gas pollution than others used to sort information or generate text. (The study didn’t include generative AI video models.) Creating a single image could use as much electricity as charging a smartphone to 50 percent capacity, it found. For comparison, a text generation model might use around 9 percent of a full smartphone charge for 1,000 inferences.

Imagine charging your cellphone after every couple of custom images you create in a text conversation, and you can see how the energy costs could add up quickly.