Samsung Electronics and SK hynix are accelerating high-bandwidth memory (HBM) capacity expansion as artificial intelligence data-center investment keeps demand ahead of supply. Samsung plans to roughly double production of HBM4, the sixth-generation HBM, and its successor HBM4E next year, lifting monthly wafer input from 180,000 sheets to 250,000, a 40% increase. It also aims for HBM4-series products to account for 80% of HBM shipments. Improved HBM4 yields, from below 60% early in mass production to around 80%, have supported higher profitability and shipments. Samsung’s HBM revenue share rose from 21% in the first quarter to 33% in the second, while SK hynix’s declined from 58% to 50%, according to Counterpoint Research. SK hynix is responding by expanding M15X wafer input in Cheongju from about 10,000 to 80,000 sheets per month and advancing its Yongin cluster plans. Oversupply concerns remain limited because HBM demand growth is forecast at 56%, above projected supply growth of 50%, while Nvidia and AMD are increasing HBM content in next-generation AI accelerators. HBM4 prices are viewed as likely to be about 65% higher next year than this year. The shortage is also tightening conventional DRAM supplies, with TrendForce forecasting contract-price growth of up to 18% quarter-on-quarter in the third quarter. Meanwhile, China’s CXMT has begun mass production on its fifth-generation DRAM platform and reached a 10% global revenue share in the second quarter. Analysts say Korean companies must expand capacity and preserve technological leadership, although currency effects and uncertainty over AI investment remain risks.