Analysis: Samsung's own memory unit pushes its foundry toward a 2027 capacity crunch

Samsung's foundry is locked in a structural self-cannibalization loop: its memory division's surging demand for 8nm-class logic—HBM controllers, CXL bridges, AI-era memory SoCs—is consuming the EUV capacity external customers depend on. This is not a cyclical squeeze. It is a permanent reallocation of who demands advanced logic, and the memory side will only grow. The hard constraint is EUV throughput. Insider-first allocation means Qualcomm, MediaTek, and others face a 2027 capacity window that is already narrowing. Design-in cycles run three to five years; miss the 2025-2026 commitment window and the roadmap is locked to a rival for the entire product generation. TSMC's 3nm and 2nm lines in Taiwan, China are the direct beneficiary, absorbing displaced demand with a clean allocation record. Intel's 18A captures the secondary tier. The deeper damage is reputational: once customers internalize that Samsung's insider priority is structural, not tactical, switching costs become psychological, not just contractual. Over the next 12-24 months, the decisive variable is capex timing. If Samsung does not add EUV tools by late 2026, the 2027 shortfall escalates from tight to a hard supply break. Foundry share could slide below 12 percent, and the memory-founding synergy that was supposed to be Samsung's moat becomes its ceiling.