Memory Bottleneck Boosts Samsung and SK hynix

A CLSA research note on Korea's semiconductor sector, reported by the Korea Times, positions Samsung Electronics and SK hynix as primary beneficiaries of persistent memory scarcity in the AI supply chain. Memory chips' share of total semiconductor industry revenue has risen from 28% to 52% over the past decade, driven by AI infrastructure expansion.

High-bandwidth memory (HBM) - the vertically stacked DRAM used in AI accelerators from NVIDIA, AMD, and others - is produced at scale by only Samsung, SK hynix, and Micron. Goldman Sachs raised its 2026 DRAM supply shortfall estimate to 4.9% of total demand, what it characterized as the most severe shortage in 15 years. SK hynix has shipped 12-high HBM4E samples to customers; NVIDIA's next-generation Vera Rubin Ultra GPU is expected to require 12 HBM4E stacks per chip, up from 8 HBM4 stacks in the current generation, meaning each product cycle demands proportionally more stacked memory capacity.

Samsung's memory division chief has warned shortages could persist through at least 2027. Major AI customers are signing multi-year binding contracts and paying deposits to secure allocations. New US domestic capacity tied to CHIPS Act investments is not expected at mass-production scale until 2028. SK hynix chairman Chey Tae-won has indicated AI-driven memory demand pressure may persist to 2030.

Constrained HBM supply raises pricing power for manufacturers while increasing infrastructure costs for model trainers and data center operators. Standard DRAM and consumer RAM markets are also tightening as fabs redirect wafer capacity toward HBM - Tom's Hardware reports roughly 93% of combined production from the three major producers is now directed toward AI-focused memory products.