A report this week from ET Datacenters highlights how persistent supply chain bottlenecks are significantly hampering the rapid expansion of AI-focused data centers globally. The piece points to constrained availability of high-voltage electrical transformers, liquid cooling infrastructure, and advanced semiconductor components — particularly the specialized networking chips and power delivery units that large-scale GPU clusters require — as the primary chokepoints slowing construction timelines.
The transformer shortage alone has drawn sustained attention from infrastructure analysts. Lead times for large power transformers, which substations require to step voltage up or down for data center interconnection, have stretched to between 18 months and 3 years in some markets, according to figures that have circulated across multiple industry reports throughout 2025 and into 2026. That figure predates AI demand; the surge in planned data center capacity has pushed that queue further. The ET Datacenters report situates the current crunch within a broader pattern of demand from cloud hyperscalers — companies like Microsoft, Google, and Amazon — racing to bring AI training and inference capacity online faster than the physical supply chain can support.
Cooling systems represent a second significant constraint. The shift toward denser, hotter GPU racks — particularly those supporting NVIDIA's Blackwell and successor architectures — has accelerated demand for direct liquid cooling (DLC) components and rear-door heat exchangers that many manufacturers were not producing at scale as recently as two years ago. Some facilities have reported construction delays of six months or more tied specifically to cooling procurement.
Geopolitically, the report notes that the bottlenecks are not uniformly distributed. Countries that have invested in domestic transformer manufacturing capacity or secured long-term supply agreements — including some European Union members and India — are better positioned than those relying heavily on import pipelines that run through constrained logistics corridors.
What this coverage rarely surfaces for a preparedness-minded audience is the grid dependency angle: AI data centers are being built to draw power at scales that require dedicated substation infrastructure, and those same transformers sitting in eighteen-month order queues are the identical components that electric utilities need to restore service after major grid disruptions. The civilian grid restoration supply chain and the AI infrastructure buildout supply chain are competing for the same constrained manufacturing capacity — primarily from a small number of producers in the United States, Germany, South Korea, and India. When utilities in storm-affected regions have historically needed replacement transformers on short timelines, they drew from a relatively thin strategic reserve; that reserve faces new pressure as hyperscalers place large forward orders. The transformer shortage and grid resilience intersection is one reason analysts who track critical infrastructure have been watching AI capital expenditure announcements with something other than pure enthusiasm.
