Please consider supporting us by disabling your content blocker.
loader

AI’s Energy Demands: A Clear Challenge

It’s no secret that the current AI boom is using up immense amounts of energy. Now we have a better idea of just how significant this usage has become.

A new paper from a team at the Harvard T.H. Chan School of Public Health has examined 2,132 data centers operating in the United States, which represent 78% of all facilities across the country. These data centers—essentially buildings filled with rows of servers—are critical for training AI models and processing requests from AI applications like ChatGPT. Due to their operational demands, these centers require substantial energy, not only for powering servers but also for essential cooling measures.

Tripling Carbon Emissions Since 2018

Since 2018, carbon emissions from US data centers have tripled. For the 12 months ending August 2024, these facilities accounted for 105 million metric tons of CO2, representing 2.18% of total national emissions. For perspective, domestic commercial airlines contributed approximately 131 million metric tons during the same period. Furthermore, an astounding 4.59% of all US energy consumption is attributed to data centers—a figure that has doubled since 2018.

However, quantifying how much of this surge is directly tied to AI is challenging. Data centers are involved in various tasks beyond AI processing, including hosting websites and storing data. Still, the researchers highlight that the influence of AI is rapidly escalating as more sectors adopt this transformative technology.

The Impact of Energy Source on Emissions

Notably, the energy sources powering these data centers tend to be “dirty.” Many are located in coal-producing areas, like Virginia, where the “carbon intensity” of their power is 48% higher than the national average. The study found that 95% of US data centers are situated in regions with higher pollution levels compared to the national average.

According to Falco Bargagli-Stoffi, one of the authors of the report, various factors contribute to this environmental burden. “Dirtier energy is available throughout the day,” he states, while renewable sources may not always meet the energy demands required by data centers that operate around the clock. Additionally, political incentives and local opposition can affect where these centers are built.

AI’s Expanding Footprint

The paper outlines a significant transition within AI technologies: a shift from text-focused models toward more complex image, video, and music generation models. Until now, many of these advanced systems had remained largely confined to research. However, recent developments—such as OpenAI’s release of its video generator Sora—indicate that public access to these technologies is imminent, further amplifying energy consumption.

Research Efforts to Measure Data Center Emissions

Researchers designed a comprehensive data collection platform to accurately assess data center energy consumption across the country. Francesca Dominici, director of the Harvard Data Science Initiative, notes, “As we scale up to images and video, the data sizes increase exponentially. Coupled with widespread adoption, emissions will soon soar.” This data will inform regulatory efforts aimed at curbing the growing carbon footprint of these facilities.

A Call for Sustainable Practices

With rising pressure from environmentally conscious communities, the forecast suggests an urgent need for Big Tech to respond. “There’s going to be increased pressure, between the environmental and sustainability-conscious community and Big Tech,” predicts Dominici, expressing skepticism about imminent regulations in the next few years.