'No permanent friends': Nvidia’s two-track play may reduce Korean chipmakers to subcontractors - Korea JoongAng Daily

Nvidia's two-track playbook is a 2007 Apple-TSMC rerun: design hegemony compresses the manufacturing layer into interchangeable capacity nodes. Once HBM4 architecture definition and packaging interface standards consolidate under Nvidia's ecosystem, Samsung and SK Hynix shift from technology partners to capacity price-lists. Technically, Korea's 3D-stacking lead is being diluted by architecture-level lock-in—superior process means nothing if the interface isn't yours and the customer isn't yours. On compliance, tightening export controls amplify the risk: higher single-buyer concentration thins Korea's bargaining chips in Washington policy negotiations. In the competitive arena, AMD will almost certainly accelerate MI400 co-development with Samsung to hedge against Nvidia's exclusivity grip; Broadcom may redirect custom-ASIC HBM demand toward SK Hynix, building a de-Nvidia second curve. The 12-24 month tail effect: if Nvidia unilaterally defines the HBM4 standard, Korean memory equities will re-anchor from 'AI memory scarcity premium' to 'foundry margin,' with a 15-20% PE compression as the base case. Korea's only escape vector is upstream penetration into EDA and architecture, but the window closes within two product cycles.