Samsung, SK hynix Cut NAND Output, Prices Set to Rise
Recent reports indicate that major memory makers are becoming more cautious with NAND Flash output, a move that could keep the market tight through 2026. Samsung Electronics is expected to slightly reduce its NAND Flash wafer production from about 4.9 million wafers in 2025 to roughly 4.68 million wafers in 2026. SK hynix is likely to follow a similar path, with output projected to fall from around 1.9 million wafers in 2025 to about 1.7 million wafers in 2026.
With Samsung and SK hynix together holding more than 60% of the global NAND Flash market, these cutbacks suggest that overall supply will remain constrained next year. As a result, NAND Flash prices may face further upward pressure.
One key reason behind the cautious production strategy is profitability. Compared with DRAM, which currently delivers stronger margins, NAND Flash offers relatively lower returns. This has pushed capital investment priorities toward DRAM, while NAND expansion takes a back seat. Management teams at both companies reportedly see little incentive to aggressively raise NAND output, especially given the historically modest profitability of NAND products. Instead, they appear focused on protecting prices and maximizing returns during the current memory upcycle rather than chasing volume growth.
At the same time, demand from AI data centers is reshaping the NAND landscape. High-capacity solid-state drives are increasingly critical for AI workloads, prompting Samsung and SK hynix to transition their NAND technology mix. Production is gradually shifting from TLC (triple-level cell) to QLC (quad-level cell), which can store more bits per cell and significantly increase storage density. While QLC typically trades some endurance for capacity, it is well suited for data center use cases that prioritize large volumes of data storage with fewer write cycles, especially when paired with advanced controllers.
Industry data shows that moving from TLC to QLC allows more data to be stored on the same silicon area, lowering overall cost per bit. However, this transition is not instantaneous. Equipment upgrades, process stabilization, and early yield challenges are all part of the shift, making a temporary reduction in effective output almost unavoidable. This transition period is another factor contributing to lower NAND production in the near term.
AI hardware demand is also adding a new layer of pressure. According to Citi's projections, NVIDIA's next-generation AI accelerator platform, known as “Vera Rubin,” is expected to enter mass production in the second half of this year. Each Vera Rubin NVL144 server system is estimated to require around 1,152 TB of SSD capacity to support NVIDIA's Integrated Compute Memory System (ICMS). If shipments reach 30,000 systems in 2026 and 100,000 in 2027, ICMS-driven NAND demand alone could reach about 34.6 million TB in 2026 and 115.2 million TB in 2027. That would account for roughly 2.8% of global NAND demand in 2026 and as much as 9.3% in 2027.
Given this scale, simply increasing wafer starts is not seen as the most efficient response. From a cost and efficiency standpoint, improving bits per wafer through QLC adoption offers a more sustainable way to meet demand growth. Still, the transition itself brings short-term supply constraints.
Some industry observers note that it remains unclear whether the current NAND production cuts are fully intentional or partly a natural outcome of technology transitions. Either way, the positive impact of tighter supply on pricing is expected to peak this year.
There is also a competitive angle. Analysts suggest that Samsung and SK hynix are adjusting their product mix in response to growing competition from Chinese memory makers, which have been steadily increasing output of mainstream NAND products since last year. By reducing exposure to highly competitive mobile and PC segments and placing greater emphasis on higher-margin enterprise and server-grade NAND, the two companies aim to defend profitability rather than compete aggressively on price.
Overall, with NAND Flash demand increasingly driven by AI-related applications, reduced output from leading producers is likely to intensify supply tightness. This impact will not be limited to AI servers; smartphones, PCs, and other end markets could also feel the effects.
Market research firm TrendForce has previously forecast that conservative capacity strategies among major NAND producers could push NAND contract prices up by 33% to 38% quarter over quarter in the first quarter of this year. IDC has also noted that NAND capacity supply growth is expected to be around 17% this year, below the long-term average.
In a recent client note, Nomura Securities added that channel checks show continued price increases across multiple memory vendors, with enterprise-grade NAND seeing especially aggressive hikes. The firm highlighted that NAND used in enterprise SSDs under the SanDisk brand could see month-over-month price increases of more than 100% as early as March.
