PSMC capacity crunch holds back Jeju Semiconductor's growth

PSMC's capacity constraint is not a cyclical blip—it exposes a structural fracture in the fabless niche-memory model. Jeju's process nodes likely sit in the 40-90nm range (NOR/embedded memory), squarely in PSMC's core territory. Yet PSMC's wafer capacity is being systematically cannibalized by automotive power-management ICs and AI analog chips, which carry 30-50 percentage points higher gross margins than memory foundry work. Memory designers get allocated residual capacity. Demand can be strong; wafers still don't reach the fab. This is product-mix migration, not a supply-demand mismatch. On supply-chain risk, Jeju's single-source dependency on a Taiwan, China foundry is a single point of failure. The more immediate threat is PSMC's internal prioritization: AI and EV orders will always outrank niche memory. If Seoul pushes memory self-sufficiency, pivoting to Samsung Foundry requires 18-24 months of yield ramp with severe revenue gap risk. Competitively, SK Hynix and Samsung are neither exiting nor expanding in niche memory—they redirect capacity to HBM. Jeju's real adversary isn't a major IDM; it's every other customer competing for the same PSMC capacity pool. 12-24 month outlook: the binding constraint for niche memory is foundry capacity, not end demand. The asset-light fabless narrative is eroding. Design houses that cannot lock in LTAs or qualify a second foundry will hit a hard revenue ceiling by H2 2025. Deep capacity binding shifts from strategic option to survival prerequisite.