Transformers, turbines, and switchgear face multi year manufacturing queues, according to a RAND analysis warning that more than half of North America could face power shortfalls by 2030.
The race to build out AI infrastructure is colliding with an industrial constraint that no amount of capital can quickly fix: the physical hardware that moves electricity from new power plants to new data centers is itself stuck in a multi-year manufacturing queue.
That is the central warning in a new RAND Corporation analysis, "Supply Chain, Energy, and AI Nexus", published June 25, 2026. The report reframes the AI energy debate from a question of generation capacity and policy to a supply-chain question: the equipment that connects new generation to new load, including large power transformers, switchgear, and the gas turbines and balance-of-plant components needed to bring new power online, is itself constrained.
The headline finding is blunt. RAND concludes that more than half of North America faces a substantial risk of energy shortfalls within the next five to ten years, driven by the combined pressure of data center load growth, electrification-driven decarbonization, and broader industrial growth (RAND report PDF). Generation-capacity initiatives are in motion across the continent, but the report flags that the pace of those builds and the friction inside the equipment supply chain that supports them may not be enough to meet forecast AI demand by 2030.
The mechanism is less glamorous than the chip story that dominates AI coverage, but it may be more binding. Large power transformers are heavy, specialized units fabricated in a small number of factories worldwide, with order books that already stretch out years. Gas turbines for new generation face similar backlogs. Permitting and grid-interconnection queues add further years before electrons actually flow. Each of these constraints is, on its own, a slow-moving problem. Together, they describe an industrial system that cannot be spun up on the same timeline as a hyperscaler capex program.
The supply-chain theme is already visible elsewhere in the AI buildout. IBM's same-day unveiling of a sub-1 nanometer chip demonstration and Micron's FY2026 supply-chain capex surge to roughly $20 billion both underscore how much of the AI race now runs through industrial capacity rather than just algorithms. The RAND report places the energy side of that same race inside the same constraint: the question is not whether AI demand will arrive, but whether the equipment, the wires, and the grid can be delivered fast enough to serve it.
What to watch next: whether utility procurement plans shift toward longer, forward-contracted equipment orders; whether the United States adds domestic transformer and turbine manufacturing capacity at the scale RAND implies is needed; and whether data-center developers begin to publicly adjust site-selection timelines around grid-interconnect dates rather than chip-availability dates. The chip bottleneck is solvable with money. The equipment bottleneck is solvable only with factories and time, and the clock on the 2030 horizon is already running.