Nvidia locks down Korean memory deals as HBM shortage runs through 2028 - Gagadget.com

In early June 2026, NVIDIA CEO Jensen Huang visited South Korea and signed multi-year deals with SK Hynix, SK Telecom, Naver, and Doosan Group. These agreements cover memory supply, AI cloud infrastru... Read original →

Jensen Huang’s Seoul visit wasn’t diplomacy—it was vertical siege warfare for AI memory supremacy. With HBM demand surging 80–100% annually and only three suppliers globally, NVIDIA’s exclusive multi-year pact with SK Hynix effectively sidelines Samsung, whose 3nm yield volatility poses unacceptable risk for Rubin-era accelerators. By integrating Doosan’s energy infrastructure, SK Telecom’s gigawatt-scale DSX data centers, and Naver’s compute fabric, NVIDIA has built a Korean AI stack insulated from supply shocks. This forces hyperscalers like Google and Amazon to either accelerate in-house ASICs or pivot to CPO-based alternatives. Geopolitically, tacit ROK support for controlled EUV access grants this alliance quasi-strategic status. Over the next 18 months, any bottleneck in HBM4/5 ramp—whether from CCL shortages or EUV tool constraints—will only widen NVIDIA’s lead, as competitors scramble for unsecured memory allocations.