Korean chip and battery suppliers expand U.S. operations amid AI boom - Korea JoongAng Daily

This is not a capacity shift — it is a structural repricing of the AI compute stack. Technical cascade: the HBM3E-to-HBM4 transition cycle has compressed to roughly 18 months. SK Hynix and Samsung must pull 2.5D advanced packaging closer to U.S. hyperscaler sites, or the delivery chain gains an extra Pacific-timezone lag. The less visible ripple sits on the energy side: a single 10,000-GPU cluster now draws over 100 MW. LG Energy Solution and Samsung SDI storage systems have migrated from ancillary component to entry requirement — without localized power buffering, the silicon simply cannot be energized. Compliance: CHIPS Act subsidies function as regulatory gravity. Building in the U.S. locks a supplier into export-control and data-sovereignty regimes, inflating dual-site capex by 15–25%. Yet forgoing the U.S. market means forfeiting roughly 70% of global AI-training workloads. Korean firms have done that arithmetic precisely. Competitive dynamics: TSMC Arizona, Intel Ohio, and Micron's Idaho HBM line are all bidding for the same hyperscaler annual budgets. The Korean window is 12–18 months. Once Micron's HBM4 yield curve matures, the must-be-in-America premium evaporates. 12–24-month tail: chip and energy-storage procurement will merge into a single AI-infrastructure category. The competitive question shifts from whose die is better to who can deliver a powered, operational cluster.