SK hynix CEO Warns 2027 Could Be Memory Industry’s Worst Supply Year
SK hynix Chief Executive Kwak Noh Jung has warned that 2027 could become the most severe year in the memory industry’s history from a supply perspective, with demand potentially remaining above available production capacity beyond 2030. The warning arrives as artificial intelligence infrastructure continues consuming increasing quantities of High Bandwidth Memory, server DRAM and advanced packaging capacity.
““We forecast that next year will be the worst year in the industry’s history from the supply perspective.””
SK Hynix CEO Kwak Noh-jung: “We forecast that next year will be the worst year in the industry's history from the supply perspective. We still forecast that customer demand will remain higher than our supply capacity even beyond 2030.”$SKHY $MU $AMD $NVDA
— Beth Kindig (@Beth_Kindig) July 21, 2026
peaking with Reuters, Kwak explained that customer demand continues to increase while SK hynix faces physical limitations on how quickly it can expand manufacturing. The company expects demand to remain above its production capacity beyond 2030, despite plans to significantly increase wafer output over the coming years.
The supply problem is not simply caused by insufficient factory expansion. A recent Commercial Times report estimates that global memory wafer capacity could grow by approximately 12% annually during 2027 and 2028. However, around half of that additional capacity is expected to be allocated to HBM production, limiting growth for conventional DRAM used in servers, gaming systems, notebooks and smartphones.
Samsung is projected to increase HBM shipments from approximately 12 billion Gb in 2026 to 20 billion Gb in 2027. SK hynix is expected to expand from 18 billion Gb to 24 billion Gb during the same period. This rapidly increasing allocation could restrict conventional DRAM bit growth to approximately 15%, while market demand is forecast to rise by around 22%.
HBM requires considerably more wafer capacity than conventional DRAM because of its larger dies, advanced stacking and more demanding manufacturing process. As NVIDIA, AMD, Google, Microsoft, Meta and other major companies expand their AI infrastructure, memory manufacturers are prioritizing products that deliver higher margins and are frequently reserved through long term supply agreements.
New factories should eventually provide relief, but meaningful capacity requires several years to construct, equip and qualify. Samsung’s Pyeongtaek P5 Fab 1 is reportedly expected to begin operating in July 2027, while P5 Fab 2 and another facility in Yongin could begin production during 2029. SK hynix is expected to bring its Yongin Y1 facility online during February 2027, followed by Y2 during the second half of 2028.
Chinese manufacturer CXMT could become an important variable as it expands DDR5 production and develops more advanced memory technologies. Major PC manufacturers are securing CXMT DRAM supply through 2027, potentially providing an alternative source for mainstream systems. However, export restrictions, customer qualification and uncertainty surrounding its international distribution could limit its ability to ease the global shortage immediately.
The warning follows similarly pessimistic forecasts from other memory industry executives. ADATA Chairman Simon Chen recently suggested that the broader memory shortage could continue for another 10 years, while Micron expects supply conditions for both DRAM and NAND to remain tight beyond 2027.
For consumers, continued pressure could keep DDR5 memory, graphics cards, notebooks, smartphones and gaming hardware significantly more expensive. System manufacturers must compete with hyperscale AI customers for access to limited memory capacity, while higher component prices are increasingly transferred into finished products.
The most important detail is that wafer capacity can grow while conventional DRAM availability still deteriorates. HBM is absorbing a growing share of new production, meaning additional factory output does not automatically translate into more memory for gaming PCs, workstations or consumer electronics.
The planned Samsung and SK hynix factories should improve supply, but their production schedules indicate that relief will arrive gradually rather than immediately. CXMT could introduce additional competitive pressure, although its ability to supply international customers at scale remains uncertain.
If demand grows by 22% while conventional DRAM output increases by only 15%, 2027 could bring another period of aggressive pricing and restricted availability. PC builders and system manufacturers may need to secure supply earlier, reduce memory configurations or accept considerably higher component costs.
Do you think increased production from Samsung, SK hynix and CXMT will ease the memory shortage, or will AI infrastructure continue consuming every available capacity increase?
