Tech companies hide the water and power demands of AI computing

Every time someone uses ChatGPT to write an essay, create an image, or give advice on planning their day, the environment pays a cost.

It’s estimated that a query to a chatbot that uses artificial intelligence requires at least 10 times more power than a standard search on Google.

Alex de Vries, founder of Digiconomist, a website that aims to uncover the unintended consequences of digital trends, calculates that if all Google searches used generative AI in the same way, they could consume as much electricity as a country the size of Ireland.

But people who use ChatGPT and other artificial intelligence applications have no way of knowing how much electricity their questions are using to be processed in tech companies’ giant data centers.

De Vries said the skyrocketing energy needs of AI technology will undoubtedly mean the world will have to burn more climate-warming oil, gas and coal.

“Even if we could power AI with renewable energy, we have to realize that there is a finite supply, so we end up using more fossil fuels elsewhere,” he said. “The net result of that is increased carbon emissions.”

AI is also thirsty for water: Xiaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside, and his colleagues calculated that ChatGPT can down roughly a 16-ounce bottle in just 10 queries.

Increased energy and water consumption from AI has raised concerns in California and around the world, with experts detailing how AI could slow the transition to green energy and increase consumers’ electricity bills and the risk of power outages.

To prevent this outcome, de Vries, Ren and other experts are calling for tech companies to disclose to users how much electricity and water their queries use.

“I think the first step is to be more transparent,” Ren said. “AI developers tend to be secretive about their energy usage, their water consumption.”

Wren said websites where users type in a search term should tell them how much energy and water that search requires, similar to how Google currently tells people searching for flights the carbon footprint of the trip.

“If we had that knowledge, we could make a more informed decision,” he said.

Data centers, the giant warehouses of computer servers that power the internet, have long consumed large amounts of electricity. But the specialized computer chips needed for generative AI are designed to read vast amounts of data, so they consume far more power.

The new chips also generate so much more heat that they require more power and water to cool them.

Although the benefits and risks of AI have yet to be fully understood, companies are increasingly incorporating AI technology into their existing products.

For example, in May, Google announced it was adding a feature to its search engine called “AI Overview,” where when someone types a question into Google Search, the company’s AI will generate an answer from the search results and highlight it at the top.

Not all of the answers generated by Google’s AI are correct, like when it instructed users to add Elmer’s glue to their pizza sauce to stop the cheese from sliding off the crust.

But searchers who don’t want AI-generated answers or want to avoid extra power or water usage can’t turn the feature off.

“Currently, users do not have the option to opt out,” Ren said.

Google did not respond to questions from The Times.

OpenAI, the company that developed ChatGPT, responded with a prepared statement but did not answer specific questions such as how much electricity and water the chatbot consumed.

“AI can be energy intensive, so we are constantly working to improve efficiency,” OpenAI said. “We carefully consider how to best use our computing power and support our partners in their efforts to achieve their sustainability goals. We also believe AI can play a key role in accelerating scientific progress in finding climate solutions.”

Three years ago, Google pledged to achieve “net zero” greenhouse gas emissions by 2030, meaning its greenhouse gas emissions would equal the amount it removes.

The company isn’t making progress toward that goal: In a July report, the company revealed that its total carbon emissions will increase by 13% in 2023. Since 2019, emissions have increased by 48%.

“As we further integrate AI into our products, increased AI computing intensity could increase energy demands and make it more difficult to reduce emissions,” the company said in the report.

Google added that its emissions will continue to grow, but that it expects them to decrease in the future, without saying when that might happen.

The company also revealed that data centers will use 6.1 billion gallons of water in 2023, up 17% from the previous year.

“We are committed to developing AI responsibly by addressing its environmental impact,” the report said.

De Vries said he was disappointed that Google did not clarify in its report how much AI is increasing electricity demand, saying the company said in its report that such “distinctions between AI and other workloads” are “meaningless.”

He said it’s not possible to calculate how much more power Google Search has become with the addition of AI Overview because the company doesn’t report AI power usage separately.

“Despite having the ability to provide the necessary information, they are currently hiding it,” he said.