Since 2018, carbon emissions from data centers in the United States have tripled. For the 12 months ending August 2024, data centers were responsible for 105 million tonnes of CO2accounting for 2.18% of national emissions (for comparison, domestic commercial airlines are responsible for approximately 131 million metric tons). About 4.59% of all energy used in the United States goes to data centers, a figure that has doubled since 2018.
It is difficult to quantify the extent to which AI in particular, booming since the launch of ChatGPT in November 2022, is responsible for this surge. That’s because data centers process many different types of data: in addition to training or testing AI models, they do everything from hosting websites to storing your photos in the cloud. However, researchers say, AI’s share is certainly growing rapidly, as almost every segment of the economy attempts to adopt the technology.
“It’s a pretty significant increase,” says Eric Gimon, a senior researcher at the Energy Innovation think tank, who was not involved in the research. “There is a lot of breathless analysis about how quickly this exponential growth could happen. But it’s still early for the company to determine efficiency gains or different chip types.
It should be noted that the sources of all this energy are particularly “dirty”. Because many data centers are located in coal-producing regions, such as Virginia, the “carbon intensity” of the energy they use is 48% higher than the national average. The document, published on arXiv and has not yet been peer-reviewed, found that 95% of data centers in the United States are built in locations with dirtier electricity sources than the national average.
There are causes other than simply being located in a coal region, explains Falco Bargagli-Stoffi, one of the authors of the article. “Dirtier energy is available throughout the day,” he explains, and many data centers need it to maintain optimal operation 24/7. “Renewable energy, such as wind or solar power, may not be as available. » Political or fiscal incentives, as well as local resistance, can also affect where data centers are built.
A key shift in AI right now means emissions from this field are likely to skyrocket soon. AI models are rapidly evolving from fairly simple text generators like ChatGPT to very complex image, video and music generators. Until now, many of these “multimodal” models have been stuck in the research phase, but this is changing.
OpenAI released its video generation model Sora to the public on December 9, and its website has been so flooded with visitors wanting to test it that it still doesn’t work properly. Competing models, like Google’s Veo and Meta’s Movie Gen, still haven’t been made public, but if these companies follow OpenAI’s lead as they have in the past, they could soon be. Suno and Udio’s music generation models are growing (despite lawsuits), and Nvidia released its own audio generator last month. Google is working on its Star project, which will be a video-AI companion capable of conversing with you about your environment in real time.