AI emissions are set to soar even higher
Since 2018, carbon emissions in US data centers have tripled. In the 12 months that ended August 2024, data centres were responsible for 105 millions metric tonnes of CO2, which accounted for 2.18% national emissions. (For comparison, domestic commercial aircraft are responsible for approximately 131 million tons). Data centers use 4.59% of the US’ energy, a figure which has doubled in recent years. Data centers are used to process a variety of data types, including hosting websites and storing photos in the cloud. They also train or ping AI models. However, the researchers say, AI’s share is certainly growing rapidly as nearly every segment of the economy attempts to adopt the technology.
“It’s a pretty big surge,” says Eric Gimon, a senior fellow at the think tank Energy Innovation, who was not involved in the research. There’s been a lot written about the exponential growth of AI. But it’s still early days for the business in terms of figuring out efficiencies, or different kinds of chips.”
Notably, the sources for all this power are particularly “dirty.” Since so many data centers are located in coal-producing regions, like Virginia, the “carbon intensity” of the energy they use is 48% higher than the national average. The paper was published by arXiv, but has not been peer reviewed. It found that 95% data centers are located in areas with dirtier electricity sources than the average. The paper’s author, Falco Bargagli Stoffi, believes that there are other factors at play than just being in coal country. He says that dirty energy is readily available all day long, which many data centers need to operate at peak performance 24-7. Data centers may not be built in areas where renewable energy sources, such as wind and solar, are available. One key change in AI is likely to cause emissions from the field to soar. AI models are moving rapidly from simple text generators such as ChatGPT to highly complex image and video generators. Many of these “multimodal models” were stuck in research until now. But that is changing.
OpenAI’s video generation model Sora was released to the public on 9 December. Its website is so overloaded with people wanting to test it that it still does not work properly. OpenAI has released its video generation model Sora to the public on December 9. Competing models like Veo by Google and Movie Gen by Meta have not yet been made available, but they may be in the near future if they follow OpenAI’s lead. Suno’s and Udio’s music generation models are growing in popularity (despite legal battles) and Nvidia has released its audio generator last week. Google is developing its Astra project. It will be a video AI companion that can talk to you in real-time about your environment.
