Google Clarified How Much Energy AI Requests Require

The growing popularity of AI systems for both personal and institutional use has prompted plenty of debates over the ethics involved, from copyright issues to AI’s effect on misinformation. There’s another issue on the table that’s also led to plenty of intense moral debates: specifically, the impact that requests to AI systems have on the environment. Whether someone is asking for information or a generated image or video, that involves processing at data centers — and that, in turn, requires energy.

The more information we have about the energy cost of AI, the better we can all assess the cost of this technology. All of which makes a recent announcement from Google that much more interesting. This document includes information on the energy usage of a text query sent to Google’s AI systems. The result, as Google’s Amin Vahdat and Jeff Dean explain, might be lower than you’d expect: “The per-prompt energy impact is equivalent to watching TV for less than nine seconds.”

Vahdat and Dean went on to explain that Google’s AI systems are getting more efficient. They point to one 12-month-long period during which time “the energy and total carbon footprint of the median Gemini Apps text prompt dropped by 33x and 44x.”

They went on to argue that Google’s method of calculating the energy use of AI queries takes into account more factors than other methods, including monitoring idle servers and factoring in data centers’ water use. Vahdat and Dean also note that their company is “heavily investing in reducing the power provisioning costs and water required per prompt.” Hopefully, being more transparent about these matters can help get them there.

Can AI Restore an Orson Welles Film? Should It?

Showrunner is looking to use AI to restore lost footage to “The Magnificent Ambersons”

This announcement is very significant; as Casey Crownhart of MIT Technology Review pointed out, it “[contains] more public information than has been available before” on the subject. Crownhart also points to some limitations of this data, including that it only focuses on text queries and focuses on the median demand, as opposed to more complex inquiries. But it’s a welcome start, and one hopes it will lead to more companies following suit.

Meet your guide

Tobias Carroll

Tobias Carroll lives and writes in New York City, and has been covering a wide variety of subjects — including (but not limited to) books, soccer and drinks — for many years. His writing has been published by the likes of the Los Angeles Times, Pitchfork, Literary Hub, Vulture, Punch, the New York Times and Men’s Journal. At InsideHook, he has…