Morgan Stanley: Power Shortages are Becoming a Core Bottleneck for AI Infrastructure, Computing Power Expansion Enters the "Power Constraint Era"

June 15 – Morgan Stanley’s latest research points out that power shortages have escalated from a secondary supporting issue to a core bottleneck holding back AI infrastructure development. Delivery lead times for power transformers have surged from 12–16 weeks pre-pandemic to 128–144 weeks today. The U.S. backlog of new energy grid connections has exceeded twice its existing installed capacity, and a projected shortage of 300,000 electricians over the next decade, plus 43% of data centers being located in high water-stress areas, are all combining to slow computing power expansion. Power system growth is far slower than data center construction, as transmission networks and key equipment supply chains have much longer cycle times. The average delivery time for power transformers now hits 128 weeks, with generator step-up transformers taking around 144 weeks—up from 12–16 weeks before the pandemic. This means even if AI data centers complete funding, site selection, and equipment procurement, they could still face delays in grid power connections and fail to launch on schedule. In the grid connection phase, the backlog of U.S. new energy projects waiting to connect to the grid is more than double the country’s total installed power capacity, creating a structural problem where "finished power generation doesn’t translate to usable power." Electricity must be connected to the grid before it can power data centers, shifting site selection logic from "suitable for building data centers" to "areas where power can be accessed quickly and reliably." Meanwhile, the lines between AI infrastructure and energy system financing are blurring. Some projects are adopting off-grid or semi-grid solutions, including direct power supply paths like gas turbines, energy storage, and fuel cells. AI companies are also moving away from relying solely on utility expansions—they’re now investing directly in power assets to secure their own power supply, pushing capital markets to integrate pricing for AI and energy assets. Beyond power, labor and water are also constraints. The U.S. is projected to face a shortage of roughly 300,000 electricians over the next decade, with over 20% of the workforce already aged 55 or older. About 43% of data centers are in high water-stress areas, making cooling water and alternative solutions major limiting factors for new construction. Several states are also proposing or implementing stricter rules and approvals for data center builds, adding more uncertainty to projects. Overall, overlapping constraints from power, grid access, equipment, labor, water resources, and policy approvals could cause computing power expansion to lag behind demand growth. The report believes this supply-demand mismatch will deepen "computing power scarcity," giving participants with stable, deliverable computing power stronger pricing power. The market is gradually shifting from a "computing power expansion competition" to a "race to control available computing power."