Global NAND wafer starts are projected at approximately 1.3 million wafers per month in 2026, as Samsung and SK hynix reduce output while Chinese capacity expands.
The market is shifting from volume-led NAND expansion toward node migration, server SSD conversion and tighter supply, with the Korean and US supplier group forecast at 670,000–700,000 wafers per month through 2026 and 2027 [S3].
NAND capacity measured in monthly wafer starts
NAND wafer capacity is expressed as wafer starts per month, not finished gigabits, because output depends on die density, wafer size, layer count and process yield. A single NAND wafer can therefore produce a different amount of usable bits from one quarter to the next, especially as suppliers transition from TLC to QLC and move toward denser production nodes [S1][S3].
Samsung, SK hynix and Micron account for 670,000–700,000 NAND wafers per month, while Chinese suppliers contribute 210,000–220,000 wafers per month during 2026–2027, for a combined range of 880,000–920,000 wafers per month [S3].
That monthly framing differs from the annual figures used for individual vendors. Samsung was projected to move from 4.9 million NAND wafer starts in 2025 to 4.68 million in 2026, while SK hynix was expected to fall from 1.9 million to 1.7 million [S2]. Those annual totals cannot be added directly to a monthly run rate without checking whether the forecasts cover the same period and accounting basis.
Supplier-level changes define the 2026 balance
Samsung's 2026 NAND wafer starts are projected at 4.68 million, down from 4.9 million in 2025, while SK hynix is expected to reduce its starts by 10%, from 1.9 million to 1.7 million [S2]. These reductions apply even as AI workloads lift NAND demand and prices, because the producers are prioritizing output mix and margins over maximum wafer volume [S2].
Forecast accuracy is itself a constraint. Omdia's November figures placed Samsung at 4.72 million wafers for 2025 and SK hynix at 1.8 million, but the January 2026 estimates revised those totals upward to 4.9 million and 1.9 million [S2]. Kioxia was listed at 4.8 million NAND wafer starts in 2024 and 4.69 million in 2025 [S2]. Procurement teams should treat these figures as revision-prone estimates rather than fixed plant capacities.
SK hynix is expected to keep NAND investment restrained through 2026, concentrating on conversion projects rather than greenfield expansion as data-center SSD recovery remains stronger than mobile demand [S1]. Samsung is also shifting production from smartphone NAND toward server NAND, including QLC-based high-capacity enterprise SSDs [S1].
Chinese expansion offsets part of the supplier contraction

Chinese NAND wafer capacity is forecast to rise from 100,000–110,000 wafers per month in the fourth quarter of 2022 to 210,000–220,000 by 2026–2027, effectively doubling over the period [S3]. The expansion offsets only part of the contraction among Samsung, SK hynix and Micron, whose 2027 capacity is projected to be 40% lower than the 2022 peak [S3].
The production mix is also changing. Chinese suppliers are expanding into segments vacated by international producers, while the Korean and US integrated device manufacturers are reallocating cleanroom space and capital expenditure toward DRAM and HBM [S3]. In a comparison of expansion strategy, Chinese capacity is being added directly, Korean and US capacity is being converted or held flat, and new greenfield NAND capacity is not expected to arrive in meaningful volume until after 2027 [S3].
The result is a two-speed market. Wafer starts among the leading Korean and US suppliers are forecast to stabilize at 670,000–700,000 per month through 2026 and 2027, while Chinese suppliers move toward 210,000–220,000 per month [S3]. For buyers, this matters because the geographic gain does not translate directly into equivalent high-end enterprise SSD supply, since product positioning and technology migration differ by manufacturer.
Node migration matters more than raw wafer additions
Suppliers plan to raise NAND bit output through process transitions that improve gigabits per square millimetre rather than adding large volumes of wafer capacity [S3]. This strategy supports more finished storage from the same constrained wafer footprint, but it also creates qualification risk for buyers whose systems, controllers or firmware are tied to older NAND generations.
QLC NAND is one focus area for high-capacity enterprise SSDs, with Samsung and SK hynix positioning their production conversions around that demand [S1]. QLC stores four bits per cell, while TLC stores three, but the architectural benefit depends on controller error correction, firmware and the workload profile [S1]. A plant's nominal wafer-start capacity is therefore not enough to forecast enterprise SSD unit availability.
New fab capacity also has a long lead time. NAND capacity additions take approximately 18–24 months to come online [S5]. That delay makes 2026 supply decisions largely dependent on conversions, yield improvement, layer-count migration and disciplined wafer loading rather than on new greenfield fabs arriving within the year [S3][S5].
Price pressure can tighten even as wafer starts fall

NAND prices were projected to rise by 33%–38% in the first quarter of 2026 as supply tightened [S5]. A 30 TB TLC enterprise SSD was priced at $10,950, nearly 260% above the $3,062 level recorded nine months earlier [S2]. This demonstrates why a falling manufacturer wafer-start plan does not automatically mean lower SSD procurement costs.
Product categories can also move at different speeds. NAND wafer pricing for selected categories increased by more than 60% month over month in the March 2026 crisis update, with 512 GB TLC showing the steepest increase [S6]. A useful comparison therefore separates three variables: wafer starts measure fabrication volume, gigabits per square millimetre measure process density, and contract pricing measures how supply and demand are balanced at the transaction level.
For inventory planning, the critical signals are supplier wafer-start revisions, the pace of server NAND and QLC conversion, and spot-price movement by capacity rather than a single headline forecast. These three indicators can identify tightening supply before quarterly financial statements show the effect.
Capacity data is suitable for planning, not procurement
Annual NAND wafer-start estimates are useful for scenario planning because Samsung and SK hynix are modeled at 4.68 million and 1.7 million starts for 2026 [S2]. They are not precise monthly commitments. The published ranges mix peak-quarter monthly capacity, annual output, supplier-group forecasts and product conversions, so each number must retain its original unit and scope [S3].
Buyers should build supplier-specific scenarios around Samsung, SK hynix, Micron, Kioxia and Chinese producers [S2][S3]. The largest forecast risk is revision, demonstrated by Samsung's 2025 estimate changing from 4.72 million to 4.9 million and SK hynix's changing from 1.8 million to 1.9 million [S2].
The next trackable indicators are monthly NAND price changes, especially the more-than-60% category-level increase recorded in March 2026, and supplier conversion milestones for server NAND and QLC [S1][S6].