📌 Transformer Shortage Forces Cancellation of 7 GW Computing Projects, Highlighting Value of Existing Power Allocations
The rapid expansion of AI computing infrastructure is running up against rigid constraints in physical grid hardware. Across the United States, 7 GW of AI data center projects have already been forced to cancel their 2026 construction plans due to the inability to secure transformers for grid interconnection. Tracing back upstream in the supply chain, constrained supplies of copper metal—with annual production growth of only about 2%—have further lengthened delivery lead times for grid equipment, prompting downstream developers to scramble for this critical bottleneck resource at any cost.
Delays in incremental grid connection approvals and equipment lead times are directly altering the pricing logic of computing assets. Since new greenfield projects struggle to connect to the grid in the short term, existing power capacity that has already secured approvals and offers interconnection certainty has become an extremely scarce, exclusive resource in the industry. Capital expenditures are inevitably shifting from self-built new projects to competing for existing operating entities that control power loads.
Among peer targets possessing such resources, existing approved but uncontracted power capacity totals approximately 5.2 GW, yet its distribution is highly bifurcated. Among them, $IREN holds up to 2,000 MW (approx. 2 GW) of uncontracted power capacity, alone accounting for roughly 40% of this group’s share, reaching twice the scale of $CIFR (1,000 MW), and significantly ahead of $GLXY (830 MW), $CLSK (585 MW), $MARA (480 MW), and $CORZ (274 MW). Until bottlenecks in transformers and upstream metal supplies ease, whether available power can actually be locked in has become the core threshold determining whether computing projects can be realized.
▌ Sources