AI chip boom tripled South Korea's economic growth in the third quarter - Euronews.com

South Korea's Q3 GDP tripling is not a semiconductor export story — it is the monetization of an HBM oligopoly. SK Hynix and Samsung collectively command roughly 90% of the global HBM market, and every NVIDIA B200/GB200 stack requires 8 to 12 HBM3E packages. The AI compute bottleneck has shifted from wafer fab to memory packaging, and Seoul sits precisely at that chokepoint. Technology cascade: the HBM4 transition (H2 2025 mass production) will push TSV stack counts beyond 20 layers, creating a second-order squeeze on CoWoS and SoIC advanced packaging. Upstream photoresist and precursor orders are already locked through 2026. Compliance risk: US export controls are accelerating China's HBM domestic substitution. If CXMT's HBM3 program reaches volume production by 2026, Seoul's 90% share faces structural dilution — market re-architecture, not a cyclical dip. Market dynamics: Micron is doubling HBM3E capacity in 2025 to enter NVIDIA's second-tier supply; Samsung is betting on HBM4 full-stack integration. Yet the global bottleneck remains TSMC's CoWoS scheduling — Korean shipment cadence is ultimately gated by Taiwan, China's packaging throughput. 12 to 24 month outlook: from Q4 2025, HBM enters the classic demand-confirmation, capacity-release, price-compression inflection. The triple-GDP print is not repeatable, but HBM's role as AI infrastructure's new oil is irreversible. The real variable is not Seoul — it is the pace of US-China tech decoupling.