The conversation about artificial intelligence has been dominated by what it can do. The story of 2026 is about what it needs. Data center electricity consumption is projected to approach 1,050 terawatt-hours globally this year — enough to make data centers, if counted as a country, the fifth largest energy consumer on earth, between Japan and Russia, according to Brookings Institution analysis. U.S. data center energy use has grown from 76 terawatt-hours in 2018 to 176 terawatt-hours in 2023, and could reach between 325 and 580 terawatt-hours by 2028 — representing up to 12% of all national electricity consumption — according to research published by the Belfer Center for Science and International Affairs at Harvard. The U.S. grid was not built for this.
The defining constraint on AI growth has shifted from computational efficiency to the physical availability of grid-scale power. A single AI-related task can consume up to 1,000 times more electricity than a traditional web search. AI workloads now pull 50 to 100 kilowatts per rack — compared to 5 to 10 kilowatts for traditional cloud storage just a few years ago. The concentration of that demand in specific geographies is testing the limits of regional grids that were designed around a fundamentally different consumption profile. In July 2024, a voltage fluctuation in northern Virginia triggered the simultaneous disconnection of 60 data centers, prompting a 1,500-megawatt power surplus that forced emergency grid adjustments to prevent cascading outages. The incident was treated as an anomaly. It was a preview.
The grid buildout required to support the AI infrastructure buildout operates on a different timeline than the data centers themselves. A state-of-the-art data center can be constructed in 12 to 24 months. Expanding the electrical grid to support it can take a decade. Utilities face stranded-asset risks on both sides: if they build transmission infrastructure to serve projected data center demand and that demand does not materialize, they absorb the cost. If they do not build and demand arrives on schedule, entire regions face power constraints that no amount of capital expenditure can quickly resolve. The PJM Interconnection — the regional transmission organization serving the mid-Atlantic and Midwest — saw electricity costs rise from roughly $60 per kilowatt-hour in 2024 to more than $300 per kilowatt-hour in 2025, according to the Information Technology and Innovation Foundation. Gas turbine prices are up nearly 200% since 2019. Utility capital expenditure growth accelerated from 13% year-over-year in 2023 and 2024 to 23% in 2025.
Data centers have historically received discounted energy tariffs and tax incentives as state and local governments competed to attract the investment. That model transferred the infrastructure risk to utilities and, ultimately, ratepayers — who absorb higher electricity costs whether or not the data center jobs and tax base materialize as projected. The Belfer Center analysis found that if infrastructure is built to serve projected AI data center demand and that demand does not arrive, those assets become stranded costs carried by the communities that hosted the facilities.
The hyperscalers understand the constraint. Amazon, Microsoft, Alphabet, and Meta are collectively pursuing nuclear power agreements, on-site generation, and dedicated transmission arrangements to secure power outside the public grid entirely. Microsoft has signed agreements to restart the Three Mile Island nuclear plant. Amazon has purchased a nuclear-powered data center campus in Pennsylvania. The companies large enough to build private energy infrastructure are doing so. The communities hosting the data centers they do not own are absorbing the grid pressure of those they do.
AI’s physical buildout has made energy the primary bottleneck for commercial AI growth — not regulation, not compute, not talent. The grid is the constraint. The communities and ratepayers absorbing the cost of upgrading it are not the same entities capturing the returns from the infrastructure being built on top of it.
— SSC News Desk | Social Storytellers Collective
Get SSC analysis delivered to your inbox every day. Subscribe free on Beehiiv: socialstorytellerscollective.beehiiv.com