US Data Center Power Shortfall Projections Revised

Analysts forecast a 38-gigawatt capacity gap by 2028 as next-generation server hardware drives massive spikes in electricity demand.

Updated on Sept. 21, 2026 in Data Centers

US Data Center Power Shortfall Projections Revised

Live Poll

Is the rapid expansion of AI data centers creating an unsustainable demand on the power grid?

Morgan Stanley analysts have revised projections for United States data center power shortfalls, warning of a 38-gigawatt gap by 2028. This expected deficit is fueled by a fivefold increase in IT power demand as new Nvidia architectures become standard.

Why it matters

The surge in computing density enabled by advanced chips accelerates the Jevons paradox, where efficiency gains lead to higher overall resource consumption. This structural shift is currently straining national energy infrastructure beyond existing supply capacity.

Nvidia's Vera Rubin NVL72 server rack consolidates 72 GPUs, pushing rack-level heat output to the equivalent of 10 to 25 ovens. While processor operation costs have decreased by 94 percent, the total power demand per rack continues to climb.

The players

Nvidia

A designer of graphics processing units and data center hardware that sets the performance standard for AI training infrastructure.

Morgan Stanley

A financial services firm providing market analysis on energy infrastructure and technology sector capital expenditures.

Elon Musk

An entrepreneur and executive currently investing in independent power generation assets through the acquisition of APR Energy.

The details

Data center operators increase computing capacity by packing more GPUs — specialized circuits designed to accelerate image processing and machine learning — into existing footprints to maintain unit economics. These systems utilize advanced cooling and power management hardware to handle the intense thermal load. As hardware throughput improves, the density of power demand per rack rises, consistently outpacing current grid delivery capabilities.

Q3 2022: Nvidia released the Hopper chip family.
2025-2028: Period of projected fivefold increase in IT power demand.
July 2026: Elon Musk acquired APR Energy.
H2 2027: Expected launch of Rubin Ultra chip architecture.
2028: Target year for current US power shortfall estimates.

The Tech Race

This trend follows the Jevons paradox, where the economic efficiency of newer chips triggers higher total consumption of grid electricity. The race to maintain leadership in AI compute is now bottlenecked by the physical limitations of the national power supply.

The projected 38-gigawatt shortfall suggests that industrial and commercial energy costs may face significant upward pressure as data centers compete with existing grid users. Businesses and residents should anticipate regional grid management programs aimed at balancing the extreme load spikes caused by these server deployments.

The takeaway

The trajectory of AI hardware design ensures that rack-level power density will continue to rise through at least 2028. Observers should monitor the release schedules for the Rubin Ultra and Fenyman architectures, as these will serve as milestones for the next major jumps in grid load.