Samsung, SK Hynix Push China NAND Toward 286 Layers as BIS Stays Silent on License Approval

Sergei Starostin/pexels.com South Korea’s two memory giants are quietly upgrading their Chinese NAND flash factories to produce chips with more than 200 layers — and the U.S. government has not publicly confirmed whether it has approved the specific equipment those upgrades require. Reuters reported on Wednesday that both Samsung Electronics and SK Hynix are pushing layer counts well beyond their current China-factory ceilings, racing to fend off a fast-rising Chinese domestic rival while meeting the AI industry’s surging demand for enterprise storage.

The two companies together account for approximately 47 percent of global NAND shipments, according to Counterpoint Research’s Q2 2026 Memory and Storage Tracker. Their Chinese factories supply a disproportionate share of that total: Samsung’s Xi’an plant alone is responsible for roughly 40 percent of the company’s worldwide NAND output, and SK Hynix’s Dalian facility accounts for an estimated 35 to 40 percent of its combined NAND production capacity including Solidigm, according to TrendForce NAND output data.

That China dependency sits at the center of a regulatory question that neither company nor the U.S. Bureau of Industry and Security has fully resolved in public. The bureau announced in August 2025 that it was terminating the Validated End-User program — a Biden-era mechanism that had allowed Samsung and SK Hynix to receive most U.S.-origin chipmaking equipment at their Chinese fabs without individual export licenses. The bureau’s announcement explicitly stated that while it intended to grant license applications to keep existing fabs running, it “does not intend to grant licenses to expand capacity or upgrade technology at fabs in China.”

Both companies have now confirmed that upgrades are underway — and that they have conducted those upgrades “in consultation with Washington.” Reuters reported it could not determine the extent to which U.S. authorities had granted export licenses for the equipment those upgrades specifically require.

Samsung’s Xi’an facility in northwestern China is the company’s only overseas memory production base, currently producing chips at approximately 200 layers (its eighth-generation V8 NAND at 236 layers, with older 128-layer lines still being phased out). Under an internal initiative codenamed “Monami” — from the French for “my friend” — Samsung is converting Xi’an to manufacture its 286-layer V9 NAND. According to one source cited by Reuters, Samsung is targeting monthly output of 5,000 wafers by December 2026, with plans to scale that to 40,000 wafers per month by December 2027.

Samsung confirmed to Reuters that the upgrade is underway, describing it as compliant with U.S. export regulations and “aimed at serving growing demand from global customers, including those in the U.S.” The company also stressed that it is not expanding capacity or building new facilities at Xi’an — language that maps directly onto BIS’s stated licensing criteria.

Outside China, Samsung currently sells NAND chips with up to 286 layers and in August unveiled a prototype exceeding 400 layers. The 400-layer project carries a notable disclosure: Samsung licensed hybrid bonding patents from YMTC in early 2025 specifically to produce that next-generation stack. Hybrid bonding — a wafer-to-wafer stacking process that eliminates conventional solder bumps and reduces electrical path length — is the architecture YMTC pioneered under its Xtacking brand and has been shown by TechInsights to be essential for NAND beyond 300 layers. That Samsung turned to a Chinese company’s patents to build its next-generation NAND illustrates how the competitive dynamic between the Korean giants and YMTC runs deeper than market-share percentages. Samsung licensed those hybrid bonding patents from YMTC in early 2025 for its 400-layer V10 generation.

SK Hynix’s upgrade centers on its Dalian plant in northeastern China — acquired when it purchased Intel’s NAND business in 2021 for approximately $8.8 billion in two tranches. The plant currently produces 192-layer NAND using a floating-gate architecture inherited from Intel — a different cell design than SK Hynix’s mainstream charge-trap NAND, which reaches 321 layers in its current leading-edge products.

Dalian remains the sole NAND manufacturing site for Solidigm, SK Hynix’s U.S.-based subsidiary, which is preparing for an initial public offering as early as 2027, with Goldman Sachs and Morgan Stanley named as lead banks and a potential valuation of up to $150 billion. That pending IPO gives the Dalian licensing question a direct financial dimension: if BIS revokes or meaningfully restricts equipment licenses for Dalian, Solidigm’s entire manufacturing base becomes legally precarious.

According to three sources cited by Reuters, SK Hynix is authorized to produce chips with up to 240 layers at Dalian and is now executing that transition from the current 192-layer baseline. SK Hynix confirmed to Reuters it “has been installing equipment at its Dalian factory in northeastern China, in consultation with the US government,” and that it is not expanding production capacity. The company may also seek permission from Washington to conduct research and development on chips with more than 300 layers at Dalian — work that would, if approved, lay the groundwork for potential production in the United States, where SK Hynix is in discussions with the Trump administration about constructing a new NAND factory.

To understand what these upgrades mean in practice, it helps to know what NAND layers are. In 3D NAND flash memory, “layers” refer to actual physical tiers of memory cells stacked vertically above one another in a single chip. Each layer contains a complete plane of charge-trap (or floating-gate) memory cells. Adding more layers is equivalent to building taller: more layers mean more data per unit of die area, directly reducing the cost per gigabyte.

The math is roughly linear at each generation: Samsung’s 286-layer V9 NAND packs approximately 43 percent more cells per die than its 200-layer predecessor, at roughly the same physical die footprint. For an enterprise SSD, that translates into more storage capacity per drive, lower cost per terabyte, and fewer physical drives needed to serve a given workload — reducing rack space and power consumption in AI data centers.

AI inference workloads are now the defining demand driver for NAND. As the industry shifts from training to inference — the phase where large language models actually process queries — demand has surged for storage that can hold key-value caches and model weights at high bandwidth and low power. Enterprise SSDs accounted for 48 percent of NAND bits shipped in the second quarter of 2026, nearly double the 26 percent share seen a year earlier, according to Counterpoint Research, and the firm expects that share to surpass 50 percent before year’s end.

Read more: Samsung Shipped $12.7B More Chips to China Than US in H1; Xi’an Fab Annual US License Deadline Looms

For Samsung and SK Hynix, standing still at 192 or 200 layers while their Chinese rival races past them in-market would mean ceding technological credibility and commercial relevance in one of the world’s largest memory markets.

That question is the article’s central unresolved issue.

BIS stated publicly in August 2025 that it intended to grant licenses sufficient for former VEU participants to “operate their existing factories in China” — but explicitly said it did not intend to grant licenses to “expand capacity or upgrade technology.” According to a notice reviewed by koreapro.org, BIS “does not plan to authorize capacity expansions or technology upgrades.”

Both Korean companies are arguing that their layer-count transitions are technology refreshes — replacing older-generation equipment within an existing physical footprint — rather than technology upgrades or capacity expansions. Samsung is not adding buildings. SK Hynix is not adding wafer starts beyond its current ceiling. But a transition from 192-layer to 240-layer NAND almost certainly requires new etching chambers, deposition tools, and metrology equipment — hardware supplied by U.S. companies including Lam Research, Applied Materials, and KLA.

BIS “referred Reuters to its 2025 statement” rather than confirming or denying whether the specific upgrades are authorized. As one analyst noted, “the fact that the building has existed for some time does not by itself settle whether additional equipment would count as a capacity expansion under export controls.”

The distinction matters enormously. If BIS determines that installing new, more advanced etching tools to enable 286-layer production counts as a “technology upgrade” — not mere “operation” — it can demand the equipment be removed or deny future license renewals. The outcome would directly affect the supply of AI enterprise SSDs and Solidigm’s IPO valuation.

Yangtze Memory Technologies Co. (YMTC), the Wuhan-based NAND maker that provides the competitive pressure driving these upgrades, captured 14 percent of global NAND bit shipments in the second quarter of 2026, placing it third in the world behind Samsung (25 percent) and SK Hynix (22 percent including Solidigm), according to Counterpoint Research. That share rose from 9 percent in the first quarter of 2025 to 14 percent in the first quarter of 2026, fueled substantially by AI-related demand, with Counterpoint Research expecting YMTC’s revenue to grow roughly six times faster than the broader industry in 2026. On YMTC’s home turf, the pressure is sharper: Omdia data cited by Reuters showed YMTC holding 38.5 percent of China’s domestic NAND market in the second quarter of 2026, overtaking Samsung within China.

Independent analysis by TechInsights of actual ZhiTai SSD products confirmed YMTC’s Xtacking 4.0 generation achieves 294 layers — split across two decks of 150 and 144 physical layers respectively. YMTC’s Xtacking architecture fabricates memory cells and peripheral circuits on separate wafers, then bonds them via copper wafer-to-wafer hybrid bonding — a technique that achieves greater than 90 percent NAND array efficiency compared to approximately 65 to 75 percent for conventional approaches. That architecture gave YMTC foundational hybrid bonding patents it now licenses to Samsung for 400-layer NAND production.

Despite its shipment ranking, YMTC’s revenue position remains weaker — it placed fifth globally in NAND revenue in the second quarter of 2026, behind Micron and Kioxia — because its product mix is still concentrated in lower-margin consumer electronics rather than high-value enterprise SSDs. That is a gap YMTC is actively working to close, and the Korea Herald reported YMTC has already launched an enterprise eSSD product with specifications nearly identical to Samsung’s PM1753. The company is also eyeing a Shanghai STAR Market IPO.

Both companies have reached the same conclusion: not trying to upgrade their China fabs is a larger commercial risk than asking Washington’s permission. Samsung and SK Hynix together told Reuters that upgrades are underway and are being done in close consultation with Washington. Shares in both companies ended 6 percent higher on Wednesday after the Reuters article’s publication.

The situation illustrates a tension at the heart of U.S. semiconductor export policy. Controls intended to prevent China from gaining advanced chip manufacturing capability apply equally to trusted U.S. allies whose Chinese production serves global supply chains, including American AI companies. Samsung and SK Hynix are simultaneously: (1) upgrading their China fabs to remain competitive; (2) arguing those upgrades comply with U.S. licensing requirements; and (3) cultivating goodwill in Washington through discussions about building new NAND capacity on U.S. soil — SK Hynix has discussed constructing a new American NAND factory with the Trump administration, with expanded China production framed as the financial bridge to make U.S. investment viable.

Lee Jong-hwan, a professor of system semiconductor engineering at Sangmyung University, has noted that NAND flash is “less dependent on leading-edge lithography than DRAM or foundry processes” — making it “a relatively more accessible area for Chinese firms to advance under US equipment sanctions” than other chip categories. That comparative accessibility is why YMTC has been able to close the gap despite export restrictions — and it is why Samsung and SK Hynix cannot afford to treat their Chinese fabs as maintenance-only operations.

Whether BIS ultimately approves the tools required for 286-layer production at Xi’an or 240-plus-layer production at Dalian remains an open question — and the bureau’s silence on the specifics is itself the answer a semiconductor supply chain running on single-digit license renewals cannot fully plan around.

Read more: Washington Demands Samsung, SK Hynix Build US Memory Fabs After Korea’s $577B Chip Bet

When the Bureau of Industry and Security terminated the Validated End-User program for Samsung and SK Hynix in August 2025, it said publicly that it “does not intend to grant licenses to expand capacity or upgrade technology at fabs in China.” Samsung’s shift from 200-layer to 286-layer NAND and SK Hynix’s shift from 192-layer to 240-plus-layer NAND both require new U.S.-made etching and deposition equipment. Whether installing that equipment constitutes a “technology upgrade” — which BIS said it would not license — or a “technology refresh within an existing operational footprint” — which the companies argue it is — has not been publicly settled. BIS declined to answer Reuters directly; it cited its own 2025 statement instead.

In 3D NAND flash, layers are literal physical tiers of memory cells stacked vertically in a chip — essentially building storage upward like a skyscraper rather than outward. Each additional layer multiplies the amount of data a single chip can store without increasing its footprint. More storage per chip means cheaper, denser SSDs for AI data centers, which need to hold large model weights and key-value caches at high speed for AI inference workloads. Enterprise SSDs now account for nearly half of all global NAND shipments, driven almost entirely by AI infrastructure buildout.

Standard 3D NAND builds memory cells and the logic circuits that control them on the same wafer, which forces engineers to compromise on density. YMTC’s Xtacking fabricates the two components on separate wafers using their respective optimal processes, then bonds them face-to-face using copper connections with no solder bumps — achieving greater than 90 percent memory array efficiency vs. roughly 65 to 75 percent for conventional designs. TechInsights confirmed the approach gives YMTC a measurable density advantage at equivalent layer counts. For 400-plus-layer NAND, the hybrid bonding technique is now essential — and YMTC holds foundational patents for it. Samsung licensed those patents from YMTC in early 2025 for its 400-layer V10 generation. That IP relationship exists regardless of what U.S. export controls do to YMTC’s equipment access.

Solidigm — the U.S.-based NAND company SK Hynix created from its 2021 acquisition of Intel’s flash memory business — manufactures all its NAND at SK Hynix’s Dalian, China facility. Goldman Sachs and Morgan Stanley are leading a planned IPO as early as 2027 with a potential valuation of up to $150 billion. If BIS determines that Dalian’s layer-count transition constitutes a “technology upgrade” requiring a license it will not grant, Solidigm’s entire manufacturing base becomes legally precarious. The connection between export control licensing ambiguity and the largest potential semiconductor IPO in history is not coincidental — it is the financial reason SK Hynix has framed China production expansion as a prerequisite for U.S. factory investment.