Are Electricity Bills About to Skyrocket Due to AI? In the USA, Some Already Pay 76% More
The growth of artificial intelligence, as has been clear for some time, is transforming data centers into the main consumers of energy, especially in the United States. According to a new analysis by BloombergNEF, by 2035 these infrastructures will account for one fifth of all electricity produced in the country, a share four times higher than the current amount.
The estimate represents a significant update from previous forecasts. BloombergNEF has increased its projection of electricity demand from data centers for 2035 by 83% compared to what was indicated last December. Other organizations have also revised their calculations upward: EPRI has more than doubled its 2024 estimate, while S&P has raised its forecasts by over a third between October and April.
The report predicts that the overall capacity of U.S. data centers will reach almost 200 gigawatts in the next decade. Of this capacity, nearly half will be dedicated to training and inference of AI models. The United States is also expected to maintain a dominant position in the sector: by 2033, it is projected to host 64% of the global AI chip demand in terms of energy consumption, according to the research agency. BloombergNEF also emphasizes that these figures could prove conservative, considering the accelerated pace at which new data centers are being developed in the country.
The main obstacle to the expansion of AI data centers is not just the computing capacity but also, and above all, the resilience of energy infrastructures. According to BloombergNEF, most of the new facilities will connect to already heavily stressed networks.
The most critical situation concerns PJM Interconnection, one of the largest managers in the United States responsible for the purchase and distribution of energy capacity, a sort of electricity wholesaler. By 2035, 34% of PJM’s electricity will be allocated to data centers. In Texas, the ERCOT network will need to dedicate 22% of its generation capacity to these facilities.
PJM has already faced difficulties in managing connection requests from both large energy producers and major consumers. The organization had suspended connection applications for new generation sources for four years. This “expansionist” policy for data centers clearly has real costs that, as usual, are passed on to local communities that have experienced substantial adjustments in energy costs. According to the report, electricity prices have increased by 76% in the last year in areas affected by significant growth in data centers.
Despite the network congestion, data centers continue to target PJM: in the latest capacity auction conducted by the manager, 38% of the requests came from these facilities.
The growth will not be limited to the United States. BloombergNEF estimates that by 2033, data centers will generate 1,935 terawatt-hours of new electricity demand worldwide, an amount of energy nearly equivalent to the annual consumption of all of India.