Since 2018, carbon emissions from US data centers have tripled. In the 12 months to August 2024, data centers emitted 105 million tonnes of CO2, accounting for 2.18% of domestic emissions (for comparison, domestic commercial airlines emitted approximately 131 million tonnes of CO2 1,000,000 tons). Approximately 4.59% of all energy used in the United States goes to data centers, a number that has doubled since 2018.
In particular, it is difficult to quantify the extent to which AI, which has grown rapidly since ChatGPT launched in November 2022, is responsible for this surge. Because data centers handle all kinds of data, from hosting websites to storing photos in the cloud, in addition to training and pinging AI models. However, researchers say that the share of AI is certainly growing rapidly, as almost every sector of the economy looks to adopt AI technology.
“This is a pretty big spike,” says Eric Guimon, a senior researcher at Energy Innovation, a think tank that was not involved in the study. “There’s a lot of breathtaking analysis about how fast this exponential growth will happen. But in terms of efficiency and understanding the different types of chips, this business is still in its infancy. It is located in
It’s worth noting that the source of all this power is particularly “dirty.” Because so many data centers are located in coal-producing regions such as Virginia, the “carbon intensity” of the energy used there is 48% higher than the national average. The paper, published on arXiv and not yet peer-reviewed, found that 95% of U.S. data centers are built in locations with dirtier power sources than the national average.
The study’s author, Falco Bargari Stofi, says there are other reasons than simply being located in coal country. “There is more dirty energy available all day long,” he says, and many data centers require it to maintain peak operations 24/7. “Renewable energy such as wind and solar may not be as available.” Political or tax incentives, as well as local opposition, can also influence where data centers are built. there is.
One significant change in current AI means that the sector’s emissions could soon skyrocket. AI models are rapidly moving from very simple text generators like ChatGPT to very complex image, video, and music generators. Until now, many of these “multimodal” models have remained at the research stage, but that is changing.
OpenAI made its video generation model Sora publicly available on December 9, but the company’s website has been flooded with traffic from people wanting to test it and is not yet functioning properly. Competing models like Google’s Veo and Meta’s Movie Gen aren’t publicly available yet, but if these companies follow OpenAI’s lead as they have done in the past, they may be available soon. Suno and Udio’s music generation models are growing (despite lawsuits), and Nvidia released its own audio generator last month. Google is working on the Astra project, which will be a video AI companion that can have real-time conversations about its surroundings.
