SK Hynix CEO Predicts AI Memory Shortage to Persist Until 2030
SK Hynix CEO Kwak Noh-jung announced that the AI-driven memory semiconductor shortage is expected to last until the end of 2030. He noted that the transition of memory products like High Bandwidth Memory (HBM) into customized goods will stabilize demand and prevent sharp declines during future industry downturns. This forecast signals a structural shift in the semiconductor industry from volatile commodity cycles to more predictable, contract-based customized production. It highlights the long-term supply constraints that AI hardware developers will face as they compete for limited HBM capacity. Unlike traditional standardized DRAM where manufacturers struggle to predict demand, HBM requires early-stage collaboration with customers, allowing SK Hynix to forecast market needs more accurately. The high demand for HBM also crowds out general-purpose DRAM capacity due to a high wafer conversion ratio.
## BACKGROUND
High Bandwidth Memory (HBM) is a high-performance 3D-stacked DRAM technology designed to meet the massive bandwidth requirements of AI and high-performance computing. Because HBM production requires more wafer capacity than standard DDR5 memory, its rapid adoption has significantly squeezed the supply of general-purpose memory.