Industry Analysis
Jensen Huang’s public endorsement of SK Hynix post-GTC 2026 in Taipei, China, is a calibrated move to lock in HBM3E supply for Blackwell while pressuring SK to accelerate HBM4 readiness. SK currently dominates HBM3E output with superior TSV yield and CoWoS integration—critical for NVIDIA’s AI training dominance. Samsung, despite leading HBM4 mass production, lacks design wins as NVIDIA hasn’t yet qualified HBM4 for its next-gen platforms. This forces Samsung to fast-track validation on Grace-Blackwell Ultra architectures while SK must preemptively scale HBM4 to avoid dependency risks. Geopolitically, NVIDIA is deliberately deepening ties with non-U.S. memory suppliers to mitigate supply chain exposure amid U.S.-China tech decoupling. Over the next 18 months, the HBM market will bifurcate: HBM3E anchors mainstream training workloads, while HBM4 battles for inference and edge AI adoption—making SK and Samsung’s ramp timelines the decisive bottleneck for global AI chip delivery.
This page displays AI-generated summaries and metadata for research purposes. Original content belongs to the respective publishers.