The United States is heading toward one of the largest energy shortfalls in modern history, as demand from AI-driven data centers accelerates far faster than the nation’s grid can expand. New estimates show that between 2025 and 2028, U.S. data centers will require 69 gigawatts of electricity, an amount comparable to the consumption of a mid-sized industrial nation.
Only a fraction of that demand is currently covered. Roughly 10 gigawatts will come from data centers already under construction with dedicated power infrastructure, while another 15 gigawatts can be supported by today’s utility grid. That leaves an enormous 44-gigawatt deficit, almost identical to the combined output of 44 nuclear power plants, a gap the country is nowhere close to filling.
We are facing a major electricity shortage:
The US will need 69 gigawatts of power for data centers between 2025 and 2028.
Only 10 gigawatts of this demand will be satisfied by data centers already under construction with dedicated power infrastructure.
At the same time,… pic.twitter.com/1HlA1iX9Kb
— The Kobeissi Letter (@KobeissiLetter) November 15, 2025
The investment challenge is equally staggering. Nvidia estimates that 1 gigawatt of new data-center capacity can require $50–$60 billion in total CapEx. If grid operators attempted to close the 44-gigawatt gap entirely, the cost would approach $2.6 trillion, and that figure does not include the additional $2 trillion required to build the data centers themselves.
The crisis is forming at the intersection of AI, cloud computing, and the exponential rise in model training workloads. Each new generation of AI chips consumes more power, and hyperscalers, Amazon, Google, Microsoft, are racing to build server campuses at record speed. But the grid cannot expand at the same pace, and utilities are warning that they lack both the infrastructure and regulatory lead time to deliver the power AI companies are demanding.
Data-center developers are already scrambling for alternatives. Some are building private substations, others are negotiating long-term renewable PPAs, and several are exploring on-site generation options. Even with these efforts, the country is still approaching a structural bottleneck.
The message from energy analysts is clear: the AI revolution cannot scale without an unprecedented expansion of America’s power grid. And unless billions flow into generation, transmission, and local grid upgrades immediately, the U.S. risks slowing down one of the most important technological transformations of the century.
