Nvidia Eyes New Inference Chip as Samsung, SK Battle for HBM Lead - Seoul Economic Daily

The semiconductor industry is witnessing intense competition as NVIDIA advances its new inference chip strategy amid growing AI computing demands. Meanwhile, Samsung and SK Hynix battle for leadership... Read original →

NVIDIA’s push into next-gen inference chips signals a structural shift toward edge AI deployment, forcing HBM memory to evolve in bandwidth and power efficiency simultaneously. Samsung and SK Hynix aren’t just racing on HBM4 specs—they’re vying for control over the CoWoS-like advanced packaging ecosystem. The first to mass-produce 12+ layer HBM stacks will likely lock in supply agreements for NVIDIA’s Blackwell Ultra successors. Tightening U.S. export controls on advanced equipment now threaten TSV yield ramp timelines for Korean makers, inflating capex. Within 18 months, HBM availability—not logic die capacity—will become the critical bottleneck in AI chip delivery. If SK Hynix leverages its tighter TSMC integration to lead, Samsung may be forced to seek packaging partners outside Taiwan, China, extending lead times. This memory battle will ultimately define the resilience—and geopolitical fragility—of the global AI hardware supply chain.