A wave of artificial-intelligence data centre construction across East Asia is running into a hard physical limit: electricity. Operators in Japan, South Korea and across Greater China are racing to add capacity to handle AI training and inference workloads, but power availability and grid connection timelines are emerging as the binding constraint on how fast that capacity can come online.
Demand for the high-density facilities that house AI accelerators has climbed sharply over the past year. These sites draw far more power per rack than conventional cloud data centres, and the concentration of that load is straining transmission infrastructure in regions that were not built to deliver it.
In response, operators are securing long-term power agreements, moving projects to locations with spare grid capacity, and pressing equipment suppliers for more efficient cooling. The shift is reshaping where data centres get built, pushing development away from saturated metropolitan clusters toward areas with available electricity and land.
Power deals and new locations
Major operators have begun signing supply agreements directly with utilities and independent generators to lock in capacity years ahead of construction. According to industry filings, several projects in Japan and South Korea have been timed around grid-connection slots rather than construction schedules, a reversal of the usual sequence where power followed the building.
Greater China's operators face a parallel challenge, balancing rapid AI capacity expansion against national efficiency rules that cap the power-usage effectiveness of new facilities. That has pushed developers toward cooler northern provinces and toward liquid-cooling systems that cut the energy spent on temperature control.
Cooling and efficiency under pressure
The density of AI hardware has made cooling a central engineering problem. Air cooling struggles with the heat output of racks packed with accelerators, and operators across the region are adopting direct liquid cooling to manage it. The technology lowers the overhead energy a facility spends on keeping equipment within operating temperatures, improving the efficiency metric regulators and customers increasingly scrutinise.
Suppliers of power and cooling equipment have reported strong order books tied to the buildout, and the competition for grid access has lengthened lead times for new connections in the most active markets. Operators that secured capacity early are now better positioned than rivals still negotiating for it.
The constraint is unlikely to ease quickly. Grid upgrades and new generation take years to deliver, while AI demand continues to climb, leaving power availability as the factor most likely to determine which projects across East Asia reach completion on schedule.