Ahead of tomorrow’s White House meeting, where representatives from Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI are expected to pledge that their data-center buildouts will not drive higher power bills for households near those facilities, UBS analyst Arend Kapteyn has published a new note featuring a chart of the day that visualizes the rapid growth in data center power demand on the grid.
After staying mostly flat for years, data center electricity demand has been surging since 2017, rising from about 70 TWh to more than 200 TWh, driven by cloud computing, social media, and AI workloads.
We outlined this theme a few years back in the piece “The Next AI Trade.”
The IEA’s 2025 Energy and AI report projects that U.S. data center power demand will double between 2025 and 2030, with a high-case scenario of a 160% increase to 582 TWh.
That would lift data centers’ share of total U.S. electricity demand from about 4.5% today to around 10% by 2030.
Kapteyn explained:
The share of US data centre demand for electricity is set to rise from roughly 4½% to 10% of total US demand (with substantial risks to the upside given the speed of revisions in hyperscaler capex)
US electricity demand from data centres is set to rise sharply as AI scales, raising concern that energy constraints could slow AI capex. After all, energy infrastructure takes far longer to build than data centres (4–8 years for transmission versus 1-3 years for data centres), and grid bottlenecks are worsening as wait times for critical components have doubled over the past three years. After remaining broadly flat between 2007 and 2017 despite rapid digitalisation, data-centre electricity demand has surged since 2017 – roughly tripling from around 70 TWh to over 200 TWh – driven by cloud computing, social media use, and the rise of AI, with hyperscale AI training facilities being especially energy-intensive.
According to the IEA’s 2025 Energy and AI report, US data-centre electricity demand is expected to double between 2025 and 2030 (and rise by around 160% in a high-case scenario of 582 TWh usage), lifting its share of US electricity consumption from 4½% to around 10%. That 2030 level of energy use is roughly equivalent to the total energy consumption of the UK or France at that stage, and implies roughly a 5% increase in overall US electricity demand over five years, after two decades of stagnation. The IEA estimates that about 20% of planned data-centre projects already face grid-related risks, although actual outcomes will depend on the pace and economic impact of AI adoption. While the implied demand increase is large, it remains smaller than the roughly 20pp rise in electricity generation over the past decade from gas, wind and solar that offset coal’s decline – suggesting that sustained renewable growth could help mitigate the impact.
Next week’s White House meeting with tech reps comes as the Trump administration is committed to rebuilding the power grid and lowering energy costs for all Americans after failed “green” policies collided with the data center boom and sparked a power crisis across the Mid-Atlantic region.
The chart above, showing that a greater share of U.S. electricity demand will come from data centers by the end of the decade, only reinforces our bullish stance on nuclear (explained here).
Loading recommendations…