Samsung breaks ground on Onyang HBM packaging fab as Korea's Lee says 40 trillion won of equipment will follow - MLQ.ai

Samsung's Onyang HBM packaging fab groundbreaking, backed by 40 trillion won in follow-on equipment spend, is not a capacity play—it is Seoul's declaration that HBM is now strategic infrastructure, not a commodity memory product. The technical chain reaction is the critical insight: packaging, not wafer fabrication, is HBM's true bottleneck. TSV drilling, hybrid bonding, and micro-bump processes dictate yield and bandwidth ceilings. By building a dedicated packaging facility rather than outsourcing to CoWoS-style external lines, Samsung is verticalizing the entire stack from die to final assembly. Expect immediate order surges for ASMPT and BESI bonding tools, and a reshuffling of NVIDIA's HBM4 supply base. Competitively, SK Hynix holds roughly 70% of NVIDIA's HBM3E allocation. Samsung's move is capital-for-time—securing capacity ahead of the HBM4 inflection window (2026-27). Hynix will almost certainly accelerate its Cheongju packaging expansion; Micron will differentiate via CoWoS-L. The 12-24 month tail: packaging will replicate the 2020 advanced-logic "equipment-first" playbook. Government subsidies effectively de-risk capex, converting fab capacity into quasi-public infrastructure. The core thesis: once stacking process becomes AI compute's primary constraint, whoever owns the packaging owns the pricing power of the next generation of compute.