The backlog is evident in the industrial yards outside of the major U.S. electrical transformer manufacturers, such as ABB in St. Louis or Eaton in Pittsburgh, where the large power units are assembled, tested, and staged for shipment. Financial reports seldom show this. While utilities, data center developers, and renewable energy project managers make phone calls that increasingly fail to yield the answers they require, transformers wait weeks or months for delivery.
Before the present spike in demand, the lead time for a large power transformer was measured in months; in several market segments, it currently ranges from 115 to 210 weeks. That might last up to four years. In order to have the equipment deployed before the middle of the decade, an electric company must order it today in order to update a substation to accommodate additional load from an AI data center.
| Category | Details |
|---|---|
| Topic | Global Shortage of Copper, Transformers & Grid Infrastructure |
| Transformer Lead Times | 2–4 years (115–210 weeks) as of mid-2025 |
| Transformer Price Increase | 60–80% since 2020 |
| Copper Use Per Hyperscale AI Facility | Up to 50,000 tons |
| Data Center Build Time | 18–24 months |
| Grid Infrastructure Build Time | 5–10 years |
| COMEX Copper Stockpile | Highest levels in 30+ years (U.S. domestic buildup) |
| Affected Global Hubs | Northern Virginia, Dublin, London, Frankfurt, Singapore |
| Competing Demand Sources | AI data centers, EVs, solar, housing, grid modernization |
| Projected Duration | Structural bottleneck for next 20 years |
| Reference Website | energy.gov |
It is more difficult to resolve the issue with industrial investment alone because of the copper dilemma that lies beneath the transformer shortage. Transformers, cables, wiring, motors, charging equipment, and the physical layer of the energy economy all rely on copper as a conductor to function. Thousands of tons of copper are often used in a major commercial data center.
Up to 50,000 tons may be needed for each of the massive campuses housing the GPU clusters running massive language models and training workloads in the hyperscale AI facilities being constructed by Microsoft, Google, Amazon, and Meta. That is a remarkable amount for a single construction project, and dozens of these facilities are being constructed all over the world. Once used for a wide range of industrial applications, copper is now being drawn concurrently toward the same focused end uses at a rate that mine output, which follows development timeframes of ten to twenty years, cannot keep up with.
In response, the United States is stockpiling; while domestic policy initiatives attempt to create a buffer against supply chain disruption, COMEX-approved copper storage has reached its greatest levels in over thirty years. Although it doesn’t solve the underlying manufacturing capacity shortfall, that buffer offers some temporary insulation.
From discovery to initial production, new copper mines require years of environmental study, permitting, infrastructure building, and capital expenditure at sizes that don’t react very quickly to short-term price signals. The supply side of the copper market is structurally slow, which is precisely the wrong trait for a crucial mineral whose demand has started to rise fast and from several angles at once.
The bottleneck is particularly noticeable when there is a discrepancy between the speed at which the demand has materialized and the rate at which the necessary infrastructure can be constructed. The IT corporations are very good at designing, permitting, and building hyperscale data centers in eighteen to twenty-four months. Depending on the jurisdiction and the particular work needed, the substation renovations, transmission line extensions, and distribution network modifications needed to supply that data center with power operate on completely different timelines: five to ten years from original planning to energization.
The constraint is in the space between those two figures, twenty-four months and 10 years. The world’s highest concentration of data centers is located in Northern Virginia, which has been handling interconnection lines for years. As the geography of available grid capacity collides with the geography of AI infrastructure deployment, Dublin, London, Frankfurt, and Singapore are all feeling different forms of the same strain.
The struggle for the few components occurs concurrently across all energy transition use cases, which makes the bottleneck especially obstinate. Transformers are necessary for solar farms. Copper cable and electrical service modifications are required for EV charging networks. Grid connections are necessary for new housing constructions. Distribution equipment and switchgear are required for industrial electrification projects.
At the same time, all of these users are vying for the same materials, manufacturers, and construction and engineering personnel. When a utility must decide whether to connect a new residential development or a new data center campus to available capacity, it is making a decision that the regulatory framework was not intended to make clear. The answer varies depending on the state, the utility, and the current political priorities.
The cost and availability of solar panels, wind turbines, and batteries—generation and storage technologies that have drastically decreased in price and drawn significant investment—have received a lot of attention in the energy transition discourse, but the infrastructure that links those technologies to the economy has gotten relatively less attention.
The cost of the panels is decreasing. The transformer at the substation, the copper that connects everything, and the cable from the panel to the house are becoming more and more problematic. None of those items can be produced at the rate that the demand side of this equation is moving, and none of them are becoming more affordable quickly enough.
