Samsung Electronics posts record quarterly profit in preliminary results

Samsung's 10-trillion-won milestone is not a cyclical rebound—it is a structural repricing of memory from commodity to quasi-custom component. When DRAM prices crashed below cash cost in 2023, the industry bled; now HBM3E contract premiums are locked to AI capex cycles, making demand far more rigid than the 2018 super-cycle ever was. The critical inflection: HBM4's logic base die transforms Samsung's 4nm foundry from a cost center into a moat. SK Hynix, lacking advanced logic capacity, risks losing architectural definition rights in the 2026-2027 transition. The binding constraint has shifted from wafer output to TSV yield and CoWoS packaging throughput. Competitively, Micron trails NVIDIA qualification by roughly two quarters; its hybrid-bonding HBM4 route, if yield lags, could see share erode from 15% to under 8% by 2027. But Samsung's existential threat is not a peer—it is Broadcom and Marvell embedding on-die SRAM into custom ASICs, structurally capping HBM demand growth. Risk vector: if export controls extend to hybrid-bonding equipment, P4/P5 HBM4 expansion becomes dependent on Taiwan, China packaging partners, eroding 3-5% of gross margin through supply-chain redundancy. Verdict: 2026 Q2 marks the pricing-power peak. As three suppliers ramp simultaneously, contract premiums compress from 40%+ to 15-20%. The "10-trillion" figure is a cycle-top confirmation, not a new normal. Consumer devices face a 2026 H2 scissors gap between rising component costs and softening end-demand, pressuring phone gross margins by 2-3 points.