US AI Data Center Expansion Stalls Amid Critical Electrical Equipment Shortages
Nearly half of the data centers expected to come online in the United States by 2026 are likely to face significant delays or outright cancellations. This alarming trend emerges as a shortage of key electrical infrastructure begins to constrain the country's rapid expansion of artificial intelligence (AI). Despite aggressive investments from tech giants like Google-parent Alphabet, Amazon, Meta, and Microsoft, which have collectively committed over $650 billion this year, the bottleneck lies not in funding or computing technology, but in the availability of essential components such as transformers, switchgear, and batteries required to power these facilities.
Widening Gap Between Ambition and Execution
According to a Bloomberg report, analysts estimate that while projects consuming up to 12 gigawatts of power are planned, only about a third are currently under construction. This underscores a widening gap between ambition and execution. The basic problem hindering the development of data centers is the scarcity of electrical equipment needed to supply power to the centers and extend the network grid. This equipment includes transformers for voltage control, switchgear for power distribution, and batteries to balance power consumption.
Benjamin Boucher, senior analyst at Wood Mackenzie, explains, "There's not enough domestic capacity to go around, so people are pretty much forced to go to the export market." The shortage is being driven by overlapping demand, with electrification efforts such as electric vehicles and heat pumps competing with AI infrastructure for the same equipment. As a result, US manufacturers are struggling to keep pace, forcing developers to depend heavily on imports, particularly from China and other countries.
Surge in Imports and Extended Delivery Timelines
Data from the US International Trade Commission shows that imports of major electrical equipment have surged in recent years, coinciding with the rapid expansion of AI data center projects. In some cases, delivery timelines for high-capacity transformers have stretched from 24–30 months before 2020 to as long as five years today, reflecting both severe supply constraints and rising demand.
This dependence on imports has led to several problems in both strategic and operational terms. Andrew Likens, head of energy and infrastructure at Crusoe, noted, "Even a small delay in getting any of the components you need to construct your well can stall your entire project. It's a really wild puzzle right now." US President Donald Trump highlighted the significance of competitiveness as early as December 2025, stating, "There will be only one winner, and that will most likely be the US or China." However, any trade strategy aimed at reducing import dependency could create further complications.
Non-Traditional Approaches and Failed Reshoring Efforts
These circumstances have pushed companies to adopt non-traditional approaches. Some developers order equipment well before final approval, while others refurbish old transformers from defunct energy facilities. For instance, Crusoe Company decided to rely on domestic procurement and assembly to prevent disruptions to its plans. However, the United States' attempt to strengthen its manufacturing capacities has been largely unsuccessful. Reshoring measures adopted over the past decade have failed to meet the increasing demand, leaving the country reliant on imported equipment.
According to Wood Mackenzie, in 2025, thousands of high-capacity transformers will have been imported to the US from China alone. Despite large investments by companies like GE Vernova and Siemens Energy to enhance production capacities, these efforts are not expected to address the existing problem in the short run. Additionally, lithium-ion batteries, another critical component, continue to be imported mostly from China at a rate of over 40%.
Broader Implications for the AI Race
Experts warn that unresolved supply constraints could undermine the broader AI build-out. Likens added, "We've seen firsthand the value it can create if you are not hamstrung by electrical infrastructure lead times. They can make or break a project." This chasm between the ambition of artificial intelligence and the system's ability to fulfil it represents a structural challenge confronting America. While the US remains dominant in advanced circuits and programming software, its inability to procure enough electronics could jeopardize its position in the global AI race, highlighting a critical vulnerability in the nation's technological infrastructure.
