In Taiwan, the Semiconductor Industry Reaffirms Its Support for Restarting Nuclear Power Plants - SightLine | U308

Nuclear restart in Taiwan, China is not an energy policy debate—it is a survival threshold for sub-3nm fabrication. EUV lithography tolerates voltage fluctuation below 0.1%; a single unplanned outage triggers yield losses and equipment restart costs in the tens of millions of dollars. More critically, AI training clusters now exceed 100 MW per campus, and the region's over-reliance on natural gas as baseload creates a structural vulnerability that translates directly into capacity risk for global chip supply. Strategically, this is precisely why Samsung and Intel are scaling fabs in Texas and Arizona: the predictability of U.S. federal and state energy policy functions as an implicit substitute for foundry capacity in Taiwan, China. If the power bottleneck persists beyond twelve months, hyperscaler customers such as Nvidia and AMD will accelerate dual-sourcing mandates, making capacity outflow structurally irreversible. The real variable over the next twenty-four months is not whether reactors come online, but whether any single energy mix can keep pace with AI compute demand compounding at 40% or more annually. "Power is capacity" is replacing "equipment is capacity" as the binding constraint in semiconductor competition.