Some folks envision the digital realm as a space that has freed us from physical constraints, a sphere detached from the complexities of pipelines, permits, and power lines, existing in an abstract utopia often called the cloud. This notion is quickly challenged when projects for data center construction halt because there’s a shortage of transformers.
To illustrate the staggering energy requirements needed for advancing construction, let’s refer to figures from the Federal Research Institute and Utilities Research Institute. By this year, U.S. data centers are expected to use about 176 terawatt-hours of electricity, accounting for roughly 4.4% of the nation’s total energy consumption. It’s anticipated that by 2028, that figure will jump to between 325 and 580 terawatt-hours. By 2030, data centers could be responsible for 9% to 17% of the country’s electricity usage. While Virginia currently leads the nation in data center capacity, there’s a possibility that by decade’s end, 39% to 57% of the electrical output will be dedicated solely to these machines.
Political ramifications are becoming increasingly evident.
On the climate front, a report published in February 2026 estimated that at least 16 gigawatts of new data center capacity is set to go live in the U.S. by 2026. Out of this, about 5 gigawatts are actively under construction, while the other 11 gigawatts have been announced but remain stagnant. Due to these delays, it’s doubtful that we will see 30% to 50% of the expected annual capacity in operation by year’s end. It’s not lack of ambition or funding that impedes construction; it’s rather that the electrical systems necessary for functioning are themselves under strain.
Advancing technology, lagging power
The tech industry often labels its setbacks as temporary challenges, things to iron out over time. But issues with power supply are a different beast. Modern data centers prioritize electric infrastructure above all else; they’re not just real estate or statements of computing prowess. IT machinery can consume as much as 95% of total energy demands. Even the cooling systems exert significant electrical loads. For top-tier AI facilities, there’s a trend where operators bypass conventional backup systems for servers, instead relying on software solutions. This is mainly a workaround due to the heavy demands of electrical infrastructure. However, being “construction complete” doesn’t equate to readiness for operation. Finalizing a build means ensuring that utility connections, internal electrical frameworks, backup mechanisms, and cooling solutions are all synced and tested. This aspect of construction is often limited.
Physical constraints tend to reside in the lesser-discussed areas of the supply chain. There’s currently a shortage of distribution transformers, the essential equipment that turns high-voltage electricity into usable formats for buildings. Back in 2019, wait times for these components stretched from 3 to 6 months, and things have only worsened. Large transformers are customized, hard to replace, costly to store, and rely on materials like grain-oriented electrical steel, aluminum, and copper, all of which have faced post-pandemic challenges. Delays commonly arise when hyperscale data centers necessitate larger power grids along with utility services and specialized substation capabilities.
Why the unpreparedness?
A historical analogy could be drawn to the railroad boom. Similar structural elements are evident; data centers today can be seen as the new railroads. Railroads emerged as a significant wave of private investment that quickly morphed into a political and economic concern. They compelled shifts in regulations and cost distribution that builders hadn’t foreseen. Data centers appear to be treading a similar path. U.S. electricity demand is evolving at a pace reminiscent of the post-war industrial surge, yet there are significant differences—the current demand for electricity is heavily concentrated in specific areas and operates on venture capital timelines. In contrast, power grids function according to utility timelines, which vary considerably.
The political fallout is emerging rapidly. Just in December 2025, the Federal Energy Regulatory Commission tasked PJM Interconnection, a regional transmission operator, with creating clearer rules for AI-driven data centers. The North American Electrical Reliability Association noted that 13 out of 23 assessed regions are likely to encounter resource adequacy issues over the next decade. In March 2026, the Energy Information Administration revealed that it would conduct a pilot study examining energy consumption in data centers, focusing on power use, cooling needs, server performance, and site features. For two decades, data centers operated largely in the background; now, federal regulators are searching for greater understanding.
Grid intersections
Maine became the first state to approve a moratorium on large data centers in April 2026, halting approvals for facilities larger than 20 megawatts while lawmakers evaluate the implications for the power grid, air, water, and costs. Governor Janet Mills, the only Democratic governor currently in office, has the power to veto further pushes for the time being. There is also a lawsuit in Mississippi, where a major AI firm has been accused of running a gas turbine near Memphis without the proper permits, highlighting the tension between rapid development and environmental regulations. In March 2026, the incoming Trump administration announced a commitment from major hyperscalers to invest in new generation facilities and cover these costs themselves instead of passing those on to consumers, signaling a shift in focus from sector-specific regulations to the broader politics of fairness.
The term “cloud” has always carried a certain rhetoric. It implies that geographic, electrical, and operational lead times are detached from physical realities. Yet the realities of transformer shortages, connection delays, and emergency turbines shatter that illusion. The internet re-emerges as a battleground over physical infrastructure: pipes, wires, substations, permits, emissions, and how costs are distributed. It arrives at particular locations, connects to specific grids, and expects specific communities to bear the consequences designed by the abstract language of cloud computing.
Delays in construction serve as a stark reminder. They compel governments to confront the challenging questions of what should be built, by whom, at what expense, and according to whose timelines.
