Power quality determines semiconductor yields
Renewable energy alone in the southwestern region has supply limitations
Forecasting demand is an initial challenge for physical AI
“For semiconductor plants and AI data centers, we can calculate how much electricity is needed at a fixed location. But for physical AI, even how to calculate electricity demand is not yet clear. We must consider that each industry needs different power characteristics and supply methods.”
Kim Yang-paeng, a researcher at the Korea Institute for Industrial Economics and Trade, made these remarks at Electric Nation Forum 2026, jointly hosted by the Korea Electrical Contractors Association and the Electrical Contractors’ Financial Cooperative at The Grand Lotte Seoul in Jung District, Seoul, on October 28. Kim’s point was that power supply measures determining the success or failure of the three megaprojects, semiconductors, AI data centers and physical AI, must also differ by industry.
Kim first emphasized that the stability of electricity supply in the semiconductor industry is directly linked to production yields. “A single wafer goes through at least 500 to 600 processes before becoming a finished product, and complex system semiconductors go through as many as 1,000,” Kim explained. “Even a small disruption in an individual process caused by an unstable power supply can reduce yields.”
The problem is that electricity demand from semiconductor production facilities is surging. The power required by the Yongin semiconductor cluster has risen from the initially reported 10 GW to 14 GW. Adding approximately 6.3 GW of demand from four memory semiconductor fabs in the southwestern region, involving investment of 800 trillion won, brings the total above 20 GW. This excludes additional demand from facilities such as advanced packaging hubs in the Chungcheong region.
Kim also said caution was needed when considering the use of the southwestern region’s abundant renewable energy resources for semiconductor plants. “Renewable energy has periods of high electricity generation, but also periods when output falls,” Kim said. “To supply the electricity semiconductor plants need reliably around the clock, we must also consider an energy storage system (ESS) or other power sources.”
Securing large amounts of electricity was identified as a key challenge for AI data centers. The government’s proposed development scale expands from 8.4 GW in the first phase to a total of 18.4 GW in 2035. “AI data centers differ fundamentally in scale from existing data centers,” Kim explained. “Operating computing equipment requires a great deal of electricity, and cooling equipment used to remove the resulting heat also consumes substantial amounts.”
For physical AI, Kim’s assessment was that a new approach is needed even to estimate electricity demand. “With semiconductor plants or data centers, large facilities are built at specific locations, so we can calculate the electricity required,” Kim said. “With physical AI, it is difficult to predict how widely robots will be deployed, where they will consume electricity and how much they will use.” Kim then emphasized that methods must first be developed to measure and forecast electricity demand distributed across industrial sites and everyday spaces nationwide.
“The success or failure of South Korea’s industrial policy will increasingly depend on electricity policy,” Kim said. “The core of an electric nation is how effectively it generates, supplies, manages and operates the electricity it needs.”
Original Korean article: Electric Times
English brief by Korea Wind Intelligence, koreawindintel.com
