SK chief defends Samsung and SK's US$3.5 trillion investment plans

$3.5 trillion is not a capex line—it is a geopolitical statement. Technical cascade: This capital will concentrate on HBM4/5, hybrid-bonding advanced packaging, and 2nm GAA. The paradigm shift: memory is no longer a commodity layer; it is the compute bottleneck itself. Samsung's HBM3E yield gap versus SK Hynix's NVIDIA qualification lead reveals this is fundamentally a defensive catch-up. Upstream, ASML EUV order books, AMAT CVD tooling, and Japanese photoresist supply chains will see backlogs stretching past 2027. Compliance exposure: US BIS export controls on sub-14nm logic and HBM have already bifurcated the market. Korea's 70%+ dependency on US/EU equipment means any decoupling escalation hits both players simultaneously. The CHIPS Act's no-forced-tech-transfer clause offers minimal real protection. Competitive dynamics: The true adversary is not the other Korean giant but TSMC's CoWoS-L expansion and NVIDIA's custom silicon trajectory. Intel 18A, if it reaches volume in 2025, further fragments advanced logic supply. The signal to Washington: we are not leaving the table. 12–24 month outlook: HBM4 becomes the new pricing anchor; CXL-attached compute storage hits commercial inflection. The long tail: AI inference workloads shift memory demand from capacity to bandwidth, fundamentally redefining the DRAM/NAND value split.