AI power-hungry monster! Bloomberg: By 2035, electricity use by U.S. data centers will rise 4 times, accounting for 20% of the nation’s total power generation

The AI craze is putting power grids around the world under unprecedented strain! Citing BloombergNEF’s latest report, TechCrunch says that thanks to a surge in demand for AI computing power, by 2035, the electricity use of U.S. data centers is expected to skyrocket to 4 times the current level—accounting for as much as 20% of the country’s total electricity generation.
（Background recap: Moonshot AI behind Dark Side of the Moon (Moonshot AI) allegedly initiated final fundraising before its IPO! Valuation is close to $50 billion; likely to list in Hong Kong this year at the fastest）
（Background: AI makes employees faster—why haven’t enterprises become stronger? Atlassian reveals three major blind spots behind organizations’ inability to profit）

Table of Contents

Toggle

• AI computing power drives expansion, and institutions significantly revise forecasts
• Power grids are nearing the limit, and regional electricity prices have surged by 76%
• Global impact: will add the equivalent of “one India” in massive electricity use

With the rapid development of artificial intelligence (AI) technology, the data centers that support massive computing power are quietly becoming beasts that devour global electricity. On July 21, 2026 (Taipei time), according to a report by TechCrunch senior reporter Tim De Chant, BloombergNEF, a research and analysis firm, issued a strong warning about the future energy landscape in its latest report. The report predicts that by 2035, U.S. data centers will consume as much as one-fifth (about 20%) of the country’s total electricity generation, with electricity demand more than 4 times the current scale (currently only about 5% to 6%).

AI computing power drives expansion, and institutions significantly revise forecasts

The main driver behind this electricity-devouring frenzy is the enormous demand for AI training and inference. BloombergNEF estimates that over the next decade, the capacity of U.S. data centers will expand to nearly 200 GW, with nearly half of that capacity dedicated specifically to supporting AI operations. Because most AI infrastructure is still concentrated in the United States, it is expected that by 2033 the U.S. will account for 64% of global AI chip electricity demand.

The astonishing growth rate has led major research and consulting firms to break their previous forecasting models. BloombergNEF has raised its forecast for 2035 electricity demand by as much as 83% compared with the data from last December. Similarly, the non-profit power industry organization EPRI more than doubled its 2024 estimate, while S&P’s forecast increased by more than one-third within just a few months. These revisions all reflect the extremely rapid development pace of data centers across the United States.

Power grids are nearing the limit, and regional electricity prices have surged by 76%

With such rapid expansion, the tightening regional power grids are facing severe challenges. The report notes that within the next decade, most new data centers will connect to power grids with constrained infrastructure capacity. For example, in the PJM Interconnection grid spanning Virginia to Illinois, as much as 34% of electricity in the future would need to be supplied to data centers. Meanwhile, Texas’s main grid, ERCOT, also needs to allocate 22% of generation capacity to meet this demand.

The cost of supply-demand imbalance is already starting to show. The PJM grid is currently under massive pressure due to a large number of connection applications; it even paused new power generation interconnections for up to four years, which directly drove the 76% surge in electricity prices in the region over the past year. Even with high costs, data center operators are still competing fiercely for power. In the most recent capacity auction, they took up 38% of the allocated costs—so much so that American Electric Power has even said it would exit the system.

Global impact: will add massive electricity use equivalent to “one India”

This impact isn’t limited to the U.S.; the AI energy battle is global. BloombergNEF warns that if AI adoption rate continues along its current high-growth trajectory, by 2033 global data centers will add up to 1,935 TWh of additional electricity demand. The figure is staggering because it is nearly equal to India’s entire national electricity consumption for a full year. As the computing power race intensifies, accelerating grid upgrades and finding stable, clean alternative energy has become an urgent issue that big tech and governments around the world cannot avoid.