China NAND upgrades: what comes after the equipment arrives

Reuters reported October 21 that Samsung and SK hynix confirmed equipment updates at Chinese factories and consultation with the U.S. government. Sources described Samsung’s plan to upgrade Xi’an toward 286-layer NAND and SK hynix’s ambition for 240-plus-layer products in Dalian. Xi’an accounts for roughly 40% of Samsung’s worldwide NAND output. The companies did not confirm all reported production details, and the full scope of U.S. approvals remained unclear.¹

The significance is not simply a higher layer count. For chip buyers, the practical questions are what usable storage will arrive, when, and at what cost. Several steps still need verification between installing equipment, establishing a process and delivering an order.

AI-generated editorial illustration: NAND manufacturing upgrades and delivery, not an actual scene.

Approval, upgrading and expansion are not interchangeable

The U.S. Bureau of Industry and Security's August 2025 policy removed certain foreign-owned Chinese fabs' license-free arrangements and distinguished existing operations from expansion and technology upgrades. That historical policy is not a substitute for current equipment-specific approvals.²

Equipment installation therefore does not establish that every subsequent step is authorized. Conversely, a general policy alone does not prove that disclosed equipment activity is unlawful. Evaluating the project's room to proceed requires checking the permitted equipment, uses and process scope, rather than turning a statement about consultations into a conclusion about comprehensive approval.

Wafer input and eventual storage capacity must also be separated. Even without more wafers being processed each month, greater usable capacity per chip could change product supply. If the process is still ramping, however, a theoretical density advantage may not immediately become saleable output. Calling every technical update capacity expansion obscures these different possibilities.

Layer counts matter, but they are not a complete scorecard

Samsung's 2024 ninth-generation V-NAND release described double stacking and channel-hole etching; SK hynix's 321-layer announcement that year explained joining and aligning multiple hole sections. These provide technical context, not evidence of progress at the Chinese sites.³⁴

Vertical scaling is consequently more than piling up finished chips. The manufacturing structure has to be stable enough to produce qualifying devices repeatedly. Layer count describes one dimension of a design; yield and reproducibility determine whether that design can become dependable supply.

A useful comparison would examine qualifying chips, usable capacity per chip and the resources consumed to produce them together. Looking only at layers can overlook the cost of added complexity; looking only at wafer volumes can miss the effect of improved density. Public layer counts do not provide enough evidence to calculate either factory's yield or cost, and no such estimate follows from this report.

From flash to a server, product choices still intervene

Solidigm's technical brief distinguishes bits per cell from whole-drive endurance and discusses drive writes per day and cumulative bytes written as lifetime measures.⁵ Layer count, capacity and tolerated write workload are not substitute specifications.

A predominantly read-oriented archive and a frequently rewritten dataset may call for different product combinations. The value of a manufacturing change therefore needs to be tested against the customer's task: required space, acceptable waiting, expected writes and reliable supply after qualification. Not every workload labeled AI uses storage in the same way, and a process improvement does not by itself establish that every SSD will become cheaper.

Deliverable output is the next meaningful evidence

The report signals equipment renewal. The evidence that would materially change the supply assessment comes later. Approval boundaries determine which plans can proceed; process ramp-up determines qualifying output; product qualification and actual orders determine whether capability becomes revenue. Those activities may run in parallel, but none can stand in for another.

Industrially, upgrading an existing plant offers a way to improve what an installed asset can supply without relying solely on a larger building footprint. Its economics nevertheless depend on investment, interruption and ramp-up costs, and what customers will pay for qualified products. The defensible conclusion now is that supply capability is being worked on, not that the improvement has already been delivered. Stable shipments and product mix will be more informative than the next layer-count record.

Source notes

1. 2026-10-21 · Reuters · Original title: Samsung, SK Hynix upgrading China plants to make more advanced chips, people say. ↩

2. 2025-08-29 · BIS · Original title: Department of Commerce Closes Export Controls Loophole for Foreign-Owned Semiconductor Fabs in China. ↩

3. 2024-04-23 · Samsung · Original title: Samsung Electronics Begins Industry’s First Mass Production of 9th-Gen V-NAND. ↩

4. 2024-11-21 · SK hynix · Original title: SK hynix Starts Mass Production of World’s First 321-High NAND. ↩

5. 2023-07-17 · Solidigm · Original title: QLC NAND Technology Is Ready for Mainstream Use in the Data Center. ↩