Samsung and SK hynix are moving to switch their NAND flash memory fabs in China to products with more than 200 layers. On October 21, 2026, Reuters reported, citing four people familiar with the plans, that Samsung's Xi'an fab plans to transition to 286-layer NAND and SK hynix's Dalian fab to NAND with 240 or more layers. The backdrop is growing demand for AI storage and intensifying competition from China's Yangtze Memory Technologies Co. (YMTC).
Samsung says it is upgrading its equipment in close consultation with the U.S. government, and SK hynix has acknowledged installing equipment at its Dalian fab. However, Reuters could not confirm how far export licenses have actually been granted for the U.S.-made equipment required. Whether the planned technology upgrades can translate into actual mass production will also depend on U.S. export controls and how they are applied.
Samsung's Xi'an fab to move to 286 layers, SK hynix's Dalian fab to 240+ layers
According to Reuters, Samsung's Xi'an fab currently produces NAND with about 200 layers. Under an internal plan it calls "Monami," the company intends to switch to 286-layer ninth-generation V-NAND (V9). One of the sources said Samsung aims to raise monthly V9 output to 5,000 wafers in December 2026 and 40,000 wafers by December 2027. A wafer is the disc-shaped substrate used to make semiconductors, and these targets indicate how far the Xi'an fab will move toward V9.
Samsung told Reuters that the upgrade is intended to meet customer demand worldwide, including in the United States, while complying with U.S. export controls. It also stated clearly that it has no plans to expand the fab's overall production capacity or build new manufacturing facilities. A higher monthly V9 output target therefore does not mean the Xi'an fab's total wafer processing capacity will rise by the same proportion. Converting existing production lines to a newer generation of products is different from increasing a fab's capacity itself.
At SK hynix's Dalian fab, the plan is reportedly to move from the 192-layer NAND currently produced to products with 240 or more layers. One source said production of up to 240 layers is permitted, and that the company may seek U.S. government permission for research and development of products exceeding 300 layers. Note, however, that what was mentioned for the 300-plus-layer products is a possible application for R&D permission, not a decision on mass production. The information that production of up to 240 layers is permitted has also not been separately confirmed by the U.S. Commerce Department's Bureau of Industry and Security (BIS). Both companies have acknowledged equipment upgrades and installations, but the specific layer counts and production volumes are planned figures disclosed by anonymous sources, so actual production starts and target attainment will need to be verified as work progresses.
Higher NAND layer counts raise memory density but make manufacturing harder
NAND flash memory is semiconductor memory that retains data even when power is off, and it is widely used in smartphones, PCs, and SSDs for AI data centers. Today's mainstream 3D NAND stacks memory cells, which store data, vertically. Adding layers allows more data to be recorded in the same area of a wafer, so it can raise storage capacity per product without increasing a fab's wafer processing capacity. The significance of converting an existing fab to a newer product generation cannot be measured by changes in the number of wafers processed alone.
Samsung's ninth-generation V-NAND, announced on April 23, 2024, is a concrete example of density gains from higher stacking. The company said at the time that it had begun mass production of 1Tb TLC (triple-level cell, storing 3 bits per cell) products and that bit density improved by about 50% compared with the eighth generation. This involved shrinking memory cells, thinning stacking materials, and a technology that reduces unnecessary channel holes to shrink the planar area the cells occupy.
However, this figure of about 50% is Samsung's comparison between product generations and does not mean supply from the Xi'an fab will increase by 50%. Nor can Samsung's announcement alone determine whether all products made at Xi'an before the upgrade correspond to the eighth generation. Estimating the actual supply increase requires considering the product mix, the share of good chips obtained from each wafer (yield), and the throughput of manufacturing processes. The density improvement rate cannot be applied directly as the production increase rate for the entire fab.
As the number of NAND layers rises, manufacturing processes also become more difficult. According to Samsung, "channel hole etching," which forms the paths for electrons that run through the stacked memory cells, requires increasingly advanced processing as the layer count grows. Simply adding layers does not make stable mass production possible.
SK hynix's 321-layer NAND, announced on November 21, 2024, also uses technologies to overcome such manufacturing challenges. At the start of mass production, the company introduced a technology that electrically connects holes formed in three separate steps in a later process, a technology that automatically corrects hole misalignment, and materials that suppress wafer warpage. These have not been confirmed to be adopted as-is in the 240-plus-layer products planned for Dalian. Still, they show that higher stacking requires advancing the entire manufacturing process, including the machining and connection of minute holes, not just stacking memory cells.
SK hynix mass-produces 321 layers in Korea while competition with YMTC intensifies in China
When evaluating technology upgrades at the China fabs, it is necessary to distinguish between the manufacturing technology each company holds as a whole and the technology actually used at its fabs in China.
In its earnings announcement on July 29, 2026, SK hynix said 321-layer NAND had become the largest share of its NAND production. It also said it plans to have about 50% of its NAND production capacity in Korea support 321-layer products by the end of 2026. This roughly 50% figure applies to capacity in Korea, not to the global ratio including China.
Against this progress in Korea, the Dalian fab's current product, as reported by Reuters, is 192-layer. Even within the same company, the pace of technology upgrades differs greatly by site. Even if a company has the technology to mass-produce 321-layer NAND, a fab in China, where U.S. export controls restrict the introduction of manufacturing equipment, will not necessarily be allowed to move to the same generation. The current switch to more than 200 layers can also be seen as a move to narrow such technology gaps between sites.
Meanwhile, local manufacturer YMTC is strengthening its presence in the Chinese market. According to Omdia research cited by Reuters, YMTC's share of China's NAND market reached 38.5% in the second quarter of 2026, exceeding Samsung's 33.5%. This figure applies only to the Chinese market, however, and does not indicate rankings in the global market.
Citing another source, Reuters reported that YMTC's main products are 160-layer and 232-layer, and that it is also producing a 294-layer product in small volumes. NAND performance and manufacturing cost are not determined by layer count alone, but YMTC is also advancing to higher stacking, and if Korean makers' China fabs keep producing older-generation products, their competitiveness in the local market could be affected. In its July earnings announcement, SK hynix also said customer demand exceeds its supply capacity. The technology upgrades at the China fabs are aimed not only at maintaining competitiveness but also at meeting demand for high-capacity, high-performance storage for AI.
U.S. export controls constrain equipment upgrades
What will shape both companies' plans is the U.S. export control regime for semiconductor manufacturing equipment. On August 29, 2025, BIS announced that it would terminate the "Validated End-User (VEU)" authorizations that had allowed certain foreign-owned semiconductor fabs in China to receive covered equipment without individual export licenses. BIS indicated it was willing to grant licenses to keep existing fabs operating, but stated explicitly that it would not approve licenses aimed at expanding production capacity or upgrading technology.
This time, BIS did not directly answer Reuters' questions about individual export licenses and merely pointed to its 2025 statement. Samsung says it complies with export controls and SK hynix has acknowledged installing equipment at Dalian, but that alone does not show whether licenses have been granted for all the U.S.-made equipment needed for the upgrades. At this point, it cannot be said that the U.S. government has eased its rules, nor that the companies are violating them.
Organizing the report and the companies' announcements, evaluating the supply impact of the technology upgrades requires distinguishing the following three indicators.
| Information | Subject and timing | Caution when evaluating supply |
|---|---|---|
| V9 monthly output to rise from 5,000 to 40,000 wafers | Reuters report dated October 21, 2026. Targets for V9 at Xi'an for December 2026 and December 2027 | Indicates the scale of V9 production, not the increase rate of the fab's total wafer processing capacity |
| Bit density improved by about 50% | Samsung announcement of April 2024. Comparison of ninth- and eighth-generation V-NAND products | Not a value measured at Xi'an, and not an increase rate for shipped capacity |
| Export license policy differs for maintaining existing fab operations versus technology upgrades | BIS statement of August 2025. Policy on China fabs formerly covered by VEU | The licensing status of the individual equipment needed for this upgrade has not been disclosed |
The V9 production target at Xi'an, the density improvement between product generations, and U.S. equipment export licenses each indicate something different. No single figure can determine the supply increase rate for the China fabs as a whole. That said, if memory density can be raised without expanding wafer processing capacity, the storage capacity obtained from the same number of wafers may increase. Actual supply also depends on yield and product mix, so it cannot be calculated from the increase in layer count alone. Nor does this plan necessarily lead directly to lower SSD prices; the first priority is whether new-generation products can be stably mass-produced at the existing fabs.
The near-term point to watch is whether Samsung's Xi'an fab can achieve its target of 5,000 V9 wafers per month in December 2026. Also still in focus is how far export licenses will be granted for the U.S.-made equipment needed for the upgrades. As progress in mass production and details of the licenses become clear, it will be possible to assess how much the China fabs' technology upgrades contribute to expanding supply of AI storage, based not only on layer counts but also on actual output and supply capacity.