AI data centers in the US will reportedly use a fifth of all American power by 2035.
Data centers will consume approximately 20 percent of all electricity generated in the United States by 2035. This is predicted by research firm BloombergNEF in a new report. This growth is primarily driven by the increasing demand for AI infrastructure. According to the researchers, the total capacity of data centers will rise to nearly 200 gigawatts over the next ten years. About half of that will be used for training and running AI models.
AI drives electricity consumption sharply upward
BloombergNEF expects the United States to remain the center of gravity for global AI infrastructure. By 2033, the country would account for 64 percent of all AI chips, measured by their power consumption. These expectations have been revised upward. Compared to an earlier estimate from December, the expected electricity demand for 2035 is now 83 percent higher. Other research organizations have also raised their expectations for data center power consumption.
According to BloombergNEF, the growth of AI data centers will primarily take place in regions where the power grid is already heavily burdened today. For example, by 2035, about 34 percent of the electricity within the American PJM Interconnection network would go to data centers. In ERCOT, the power grid for most of Texas, that would rise to 22 percent. PJM has long struggled with a congested connection process for new power plants and large-scale consumers. Rising demand has caused electricity prices in the network to increase by 76 percent over the past year.
Global demand continues to rise
While the United States continues to host the majority of AI computing power, demand is also growing rapidly worldwide. If this pace of AI development continues, data centers worldwide will consume an additional 1,935 terawatt-hours of electricity by 2033. That is nearly equivalent to the entire annual electricity consumption of India.