Compute & Infrastructure
Korea’s AI buildout is outpacing grid headroom. Seoul is weighing nuclear-backed campuses, time-of-use tariffs, and high-voltage upgrades that could reset North Asia’s compute map.
Korea’s AI buildout is slamming into a power bottleneck. Seoul and KEPCO are floating nuclear-backed campuses, 345–765 kV corridors, and new tariffs to keep GPU clusters online. If siting and baseload align, North Asia’s AI map—and who captures training and inference—could shift.
Key Takeaways
AI clusters from Naver, KT, and global clouds are scaling to tens of megawatts, coinciding with new semiconductor fabs and stressing 154 kV nodes in Gyeonggi and Chungcheong. Interconnection queues and water permits slow inland sites. KEPCO filings flag double‑digit industrial growth this decade, with data centers plus fabs driving over a quarter of incremental use. Without high‑voltage access and faster approvals, Korean AI tasks could spill to Tokyo, Singapore, or U.S. West Coast capacity, adding latency and cost for domestic services.
MSIT has floated priority siting near 345–765 kV corridors, time‑of‑use tariffs to push training off‑peak, and demand‑response for inference, plus substation co‑location, queue transparency, and carbon/water reporting. Local permits are tightening on cooling water and heat reuse. Private consortia are studying SMRs paired with hyperscale campuses to deliver 24/7 baseload, leveraging Korea’s reactor supply chain. Financing, licensing, and public acceptance remain hurdles, and grid upgrades still take years. Success could seed copycats from Taiwan to the UAE; failure would weaken the case for clean firm power for AI elsewhere.
Want to go deeper?
Subscribe to Asia AI Front for AI signals from Asia and Russia before they reach the English-language mainstream.
Watch KEPCO’s capex plan and MSIT siting rules for signals on nuclear‑backed or substation‑adjacent campuses. Track coastal land and water permits and any SMR MOUs naming a data center and grid node.
Sources & References
