DRAM Shortage Could Deepen in 2027 as HBM Demand Eats Into Memory Supply
The global memory market may be heading for one of its tightest supply periods yet, with commodity DRAM availability potentially dropping sharply in 2027. According to Apacer CEO C.K. Chang, DRAM manufacturers could release only around 30% of their 2026 volume next year, creating a possible 70% year-over-year supply squeeze.
The warning comes as high-bandwidth memory, better known as HBM, continues to take priority across major memory fabs. HBM is in extremely high demand because it is a key component in AI accelerators, advanced servers, and high-performance computing systems. As manufacturers allocate more wafer capacity to HBM, traditional DRAM used in PCs, smartphones, consumer electronics, and standard servers may face tighter availability.
The concern is not limited to Apacer. SK hynix CEO Kwak Noh-jung has also indicated that the memory industry could face an exceptionally difficult supply environment in 2027, describing next year as potentially the worst in the industry’s history from a supply perspective.
The core issue is that memory makers are expanding production, but much of that additional capacity is being absorbed by HBM. Industry expectations suggest memory wafer supply may increase by roughly 12% annually across 2027 and 2028. However, about half of that new supply could go toward HBM production, leaving far less room for conventional DRAM growth.
Samsung is expected to ship around 12 billion GB of HBM in 2026, increasing that figure to about 20 billion GB in 2027. SK hynix is projected to supply roughly 18 billion GB of HBM in 2026 and around 24 billion GB in 2027. These numbers highlight just how aggressively the industry is shifting toward AI-focused memory products.
That shift may create a major imbalance. Non-HBM DRAM bit growth is expected to be limited to about 15% annually in 2027 and 2028, while demand could grow closer to 22%. If that gap materializes, buyers across the PC, smartphone, server, and consumer electronics markets may face higher prices, longer lead times, and more competition for available memory supply.
New production facilities are on the way, but they may not provide immediate relief. Samsung’s P5 Fab 1 is expected to begin operations by July 2027, while P5 Fab 2 and another Yongin facility are not expected to contribute until 2029. SK hynix’s Yongin Y1 fab is scheduled to come online in February 2027, with Y2 expected in the second half of 2028.
Even with these expansions, the explosive rise of AI infrastructure is changing the memory industry’s priorities. SK hynix has noted that memory demand is expected to keep growing not only for HBM used in AI computing, but also for server DRAM that supports agentic AI systems, high-performance workloads, high-capacity NAND, and the broader expansion of AI services and data usage.
For everyday consumers, the impact could be felt through more expensive laptops, desktops, smartphones, gaming devices, and other electronics that rely on DRAM. If supply becomes constrained while demand keeps rising, manufacturers may pass higher component costs on to buyers.
One possible long-term solution could come from CXMT’s development of 3D DRAM technology. Unlike traditional DRAM scaling, which focuses on shrinking horizontal features, 3D DRAM stacks memory cells vertically to improve bit density and capacity. This approach could help the industry increase memory output without relying solely on increasingly complex node shrinks and advanced lithography tools.
For now, however, the memory market appears to be entering a critical period. HBM demand from AI is reshaping supply chains, and traditional DRAM may become harder to secure in 2027. If the forecasts prove accurate, the next major memory shortage may not be driven by weak production, but by the industry’s rapid pivot toward AI-first hardware.
