Transformer shortages gate AI data centre buildout as equipment queues stretch globally
Physical hardware procurement, not power purchase agreements, has become the binding constraint on AI-driven electricity demand growth.
Large transformers, switchgear, generators, busways, breakers, UPS systems and high-voltage components have all become high-risk procurement categories for AI data centre developers, DataM reported on Thursday (2026-07-16), identifying equipment supply rather than GPU availability as the deciding factor in which projects get built and when.8
Even projects that have cleared utility approval queues are stalling. Developers who secured grid connection commitments now find themselves competing for physical hardware in short supply across multiple categories simultaneously, according to the DataM analysis. The bottleneck has migrated from the planning table to the order book.8
The scale of committed demand makes the supply squeeze consequential. US utilities have already agreed to add 116 GW of large load to their networks, Wood Mackenzie tracking shows — a volume that is straining approval pipelines and construction schedules well before equipment delays compound the pressure.3
Spending on facilities by the five largest hyperscalers is forecast to jump 50% to over US$300 billion in 2025, Wood Mackenzie data show, and base-case projections from the firm's Lens Energy Transition Scenarios tool see power demand compounding at more than 5% annually in China, India and Southeast Asia, with those developing markets accounting for the bulk of global growth.3
In the Asia Pacific region the numbers run harder still. APAC data centre power consumption is projected to triple between 2024 and 2030, at a compound annual growth rate of 21.1%, driven by cloud and hybrid adoption, according to GlobalData. India, Russia, Japan, South Korea and Australia all carry sizeable project pipelines spanning the full lifecycle from planning to commissioning.7
But pipeline totals obscure a more concentrated risk. Upcoming projects are frequently measured in hundreds of megawatts and in some cases approach gigawatt-scale campuses, reporting from May (2026-05-26) showed, as hyperscalers prioritise scalability and delivery certainty in the race for power. One delayed transformer order can stall an entire campus.5
Europe faces the same squeeze from the grid operator side. ENTSO-E warned in May (2026-05-08) that transmission system operators may be forced to reduce renewable energy penetration if data centre demand growth is not properly managed — a signal that the constraint extends beyond building new generation into how existing grids are operated day to day.4
Montel senior analyst Huangluolun Zhou frames the European dynamic as rapid, localised demand spikes rather than uniform load growth across national grids. Data centre clusters form in specific regions and create hot spots that test local network capacity before national demand statistics register any movement.1
Wood Mackenzie's longer-range projections capture the geographic unevenness. The firm expects an overall compound annual growth rate of just over 2% for power demand between 2025 and 2050 globally, but country-level outcomes range from 30% cumulative growth in Serbia to 157% in Denmark — a spread that reflects how exposed individual power markets are to hyperscaler site selection decisions.3
Cooling hardware is adding a parallel procurement constraint. The data centre liquid cooling market is set to expand from USD 4.07 billion in 2026 to USD 27.65 billion by 2033, a compound annual growth rate of 31.5%, according to a Research and Markets forecast published in May (2026-05-10). Air-cooled legacy designs are hitting density limits as AI chips push rack power loads higher, making liquid cooling less optional.6
The fuel mix implications sit uncomfortably alongside decarbonisation commitments. BloombergNEF concluded in May (2026-05-19) that the data centre expansion required to support AI is expected to keep fossil fuels in use for longer, with AI-driven energy use a key new source of electricity demand going into the coming decade.2
For traders watching European power, the ENTSO-E curtailment warning carries a specific operational signal. Concentrated demand in particular grid nodes can move regional price spreads well before national demand statistics shift, and any system operator decision to curtail renewables in response to localised congestion would complicate the generation mix in ways that market-wide forecasts do not capture.4
Order books for large power transformers remain stretched globally, with utilities serving data centre campuses competing for delivery slots against those reinforcing grids for broader electrification. Whether hyperscaler capital spending forecasts hold once equipment lead times force schedule slippage into 2027 delivery windows is the concrete execution risk the supply chain has not yet priced.8,3