South Korean chipmakers face financing dilemma behind AI memory demand

The structural shift: memory makers are being pulled into a quasi-creditor role for AI infrastructure — a position no DRAM supplier has occupied since 2008, and the risk vector is inverted. Demand is real and structural; the binding constraint is now buyer solvency. Technical cascade: The HBM3E-to-HBM4 transition demands 2-3x capex per bit. Both Samsung's 1c and SK Hynix's 1b nodes face yield ceilings. If hyperscaler financing fractures, already-committed fabs become stranded assets — worse than 2022, because capacity is locked in. Compliance exposure: Extending credit terms or taking equity stakes in customers blurs the supplier-investor boundary, inviting dual antitrust scrutiny in Seoul and Washington. Competitive dynamics: Micron can position as the "untangled" supplier. NVIDIA may push direct equity into memory fabs, fundamentally restructuring the oligopoly. 12-24 month outlook: HBM pricing will fully decouple from commodity DRAM. The financing gap will likely close via sovereign wealth vehicles (KIC, GIC) or data-center-revenue-linked ABS structures. The true tail risk isn't default — it's memory becoming a balance-sheet line item that compresses AI firms' R&D budgets, thereby strangling the very demand cycle that justifies the capex.