Samsung Foundry gets boost from HBM4 as 2nm push continues

HBM4's production ramp is dissolving the boundary between memory and foundry — a structural shift most sell-side desks are still underestimating. The base die migration from DRAM to logic process nodes means Samsung's 4nm order surge is not independent foundry demand; it is a byproduct of HBM4's architectural evolution. Technically, HBM4's advanced packaging and TSV requirements force Samsung to simultaneously improve 4nm yield and packaging co-optimization. The 2nm push is a direct counter to TSMC's N2 — if Samsung fails to lock in NVIDIA's next-gen accelerator base die orders before 2026, its foundry business reverts to backup-supplier status. On compliance, ASML EUV export restrictions remain a hard constraint on 2nm mass production; HBM4's Cu-Cu hybrid bonding demands extreme material purity, making Japanese silicon wafer supply a single-point-of-failure risk. In competitive dynamics, SK Hynix's HBM3E head start is being eroded by the generational reset. TSMC leverages CoWoS capacity as a lock-in mechanism. Samsung's real card: offering a bundled memory-plus-foundry solution to NVIDIA and AMD as a one-stop alternative — a structural advantage TSMC cannot replicate. 12-24 month call: HBM4 becomes the de facto AI accelerator standard, giving Samsung's foundry its first credible shot at TSMC. But if 2nm yield lags N2's production timeline, that window closes by mid-2026.