Column: Samsung's memory boom reshapes its investment strategy

Samsung's 107.4 trillion KRW quarterly operating profit is a milestone that masks a structural vulnerability. When memory contributes over 70% of operating earnings, the company has effectively become a leveraged proxy for the AI capex cycle rather than a diversified semiconductor platform. The 2017-18 supercycle taught this lesson; the market has chosen to forget. The real story is HBM, not commodity DRAM. Samsung trails SK Hynix by roughly one generation on HBM3E, and the HBM4 mass-production window in H2 2026 is its last credible catch-up opportunity. The investment pivot almost certainly means compressing 256Gb DDR5 capacity and redirecting capex toward HBM4 and hybrid-bonding advanced packaging. But ASML EUV and BESI equipment lead times constrain capacity elasticity — Samsung cannot meaningfully shift HBM market share before Q1 2027. Competitively, SK Hynix has locked in an architectural-partner role with NVIDIA at 60%+ HBM share. Micron has carved a second pole via AMD. Meanwhile, CXMT's DDR5 ramp will erode Samsung's commodity DRAM pricing power within 18 months — the most fragile layer of this profit structure. The 12-24 month question is not whether the memory boom sustains, but whether Samsung converts from chaser to peer in HBM4. Failure means a 2027 P&L worse than 2019, because the competitive threat is now structural, not cyclical.