Analysis forecasts US data centers could consume 20% of national electricity by 2035, up from current ~3%, driven by AI training and inference workloads.

Data centers are on track to consume 12% of the nation's total electricity supply by 2030 and one-fifth of US supply by 2035, according to the latest projections from BloombergNEF. This marks a significant increase from today, when data centers account for approximately 5.9% of power consumption.

The research firm estimates that US data centers will require 194 gigawatts by 2035—up 83% over its previous forecast from December of last year. To put this in perspective, one gigawatt is roughly the output of a single traditional nuclear reactor.

BloombergNEF attributes the surge in energy demand primarily to a wave of AI facilities scheduled to be built over the coming decade. In states where AI data centers are more densely concentrated, power needs will be even greater. PJM Interconnection LLC, a grid operator active in more than a dozen states including Virginia, expects more than a third of its capacity to go to data centers.

Lloyd Arnold, an analyst with BloombergNEF and one of the report's authors, noted that the grid is already struggling to keep up with current demand, which is affecting how assets connect to it. Beyond the new AI data centers coming online, grids remain responsible for powering cities, electric vehicles, and all other infrastructure.

These estimates may be conservative and cover only the US. Worldwide, data center energy demand will be significantly higher, with data centers collectively potentially requiring as much as 1,935 terawatt-hours—roughly equivalent to what India currently consumes annually.

Yet estimates remain subject to change. There remains a small possibility that an unexpected development in the AI industry—a true market correction—could prevent many planned data center facilities from being built. Given the current trajectory and momentum around AI, this seems unlikely, though history shows such reversals are possible. The Dot Com bubble of 2000 serves as a cautionary reminder.