On October 21, 2026, Reuters published an exclusive report with a finding that read like a contradiction. Samsung Electronics and SK Hynix, the two largest memory chip makers in the world, were upgrading their factories inside China to produce more advanced chips. Samsung was converting its Xi'an NAND plant from roughly 200-layer technology to 286-layer V9 NAND, under an internal project named "Monami." SK Hynix was upgrading its Dalian NAND plant toward 240 layers and above. Both companies confirmed the upgrades were underway and said they were in close consultation with Washington.
The contradiction lies in the policy, not the reporting. In 2025, the U.S. Bureau of Industry and Security stated it did not plan to approve licenses for technology upgrades and capacity expansion at foreign-owned wafer fabs in China. The export controls imposed since 2022 require annual licenses for the installation of U.S.-made equipment — etch tools, deposition systems, inspection gear — at those facilities.
The upgrades were happening anyway. Reuters could not determine to what extent U.S. authorities had approved the equipment licenses required to make them possible.
That gap between stated policy and observable outcome is the story. It is not a story about Samsung or SK Hynix defying Washington. It is a story about what happens when an export control regime collides with a market where the customer and the competitor are in the same place.
What the Upgrades Actually Are
The scale of the two companies' China operations makes the question of upgrades consequential.
Samsung's Xi'an plant accounts for approximately 40 percent of the company's global NAND output. It currently produces chips at roughly 200 layers. The Monami project is converting that line to 286-layer V9 NAND, with a target of 5,000 wafers per month by December 2026 and 40,000 wafers per month by December 2027 — an eightfold increase in advanced-node output in twelve months.
SK Hynix's Dalian plant, acquired from Intel in 2021, currently produces 192-layer NAND. The upgrade targets 240 layers and above. Separately, SK Hynix's Wuxi DRAM plant received 581.1 billion won in investment in 2025, up 102 percent year over year, and completed a process migration from 1z to 1a — the fourth-generation 10-nanometer-class node — with 90 percent of capacity converted. The Wuxi plant can now mass-produce DDR5 and LPDDR5X, and it carries 30 to 40 percent of SK Hynix's global DRAM output.
Those numbers describe facilities that are not marginal to the two companies' global supply chains. They are core assets. The question is not whether Samsung and SK Hynix want to upgrade them. The question is how they are doing so under a licensing regime that was designed to slow the transfer of advanced manufacturing capability into China.
The Licensing Mechanism
The export control framework that governs these fabs has a specific structure, and the structure is where the ambiguity lives.
U.S. rules restrict the export of certain semiconductor manufacturing equipment to China. For foreign-owned fabs operating in China, the relevant mechanism has been a system of annual licenses — sometimes described as "validated end user" authorizations — that allow the installation of controlled equipment under conditions the U.S. government sets.
The policy signal in 2025 was restrictive. BIS indicated it did not intend to approve licenses for technology upgrades or capacity expansion at these facilities. The stated rationale was that advanced chipmaking capacity in China, even when operated by allied companies, contributes to the technological capability the controls are meant to constrain.
The 2026 reality is different. Samsung and SK Hynix are upgrading. Reuters reported that both companies are "in close consultation with Washington," a phrase that suggests the upgrades are not covert. It also suggests that whatever consultation is occurring has not produced a public prohibition.
There are two possible readings of that gap. The first is that Washington's posture has shifted — quietly, without a public announcement — toward allowing upgrades that keep allied fabs competitive against Chinese rivals. The second is that the upgrades are proceeding within the narrow space the licensing system permits, using equipment that falls below the control thresholds or authorizations granted for specific process steps rather than wholesale capacity expansion.
Reuters did not resolve which reading is correct. What it did establish is that the upgrades are real, the companies acknowledge them, and the U.S. government has not stopped them.
The Competitive Logic
The reason Samsung and SK Hynix are upgrading their China fabs is not complicated. It is the same reason any manufacturer invests in a facility near its largest customers: the alternative is to lose the market to someone who does.
China is the world's largest market for memory chips used in AI data centers, smartphones, and consumer electronics. It is also the home of the competitor that is closing the technology gap fastest. CXMT, the Chinese DRAM maker, released its first 4F² DRAM product — a server DDR5 RDIMM — at the end of 2026. Its G5 platform entered mass production in September with a 11.95-nanometer active-area half-pitch, according to CXMT disclosures. Its share of global DRAM revenue stood at 7.6 percent in the first quarter and hit 9.5 percent in the second, a 99.3 percent jump in sequential revenue.
YMTC, the Chinese NAND maker, is the reason Samsung is converting Xi'an to 286-layer V9 rather than waiting. If Samsung's most advanced NAND available inside China remains at 200 layers while YMTC ships denser products, Samsung loses the domestic Chinese customer to a local supplier — not because the customer prefers Chinese chips, but because the customer needs the density and cannot import the alternative.
That is the trap the export controls create for allied manufacturers. Restricting the flow of advanced equipment into China does not preserve the allied manufacturer's position. It cedes the Chinese market to the Chinese competitor, who is not subject to the same restrictions and who can iterate faster because it is closer to the customer.
Samsung and SK Hynix are not upgrading their China fabs to help China. They are upgrading them to avoid losing China to a competitor that Washington's controls do not reach.
The Enforcement Boundary
The Reuters report exposes the boundary between what export controls can and cannot do.
A control regime can restrict the export of a specific piece of equipment. It can require a license for a specific process step. It can threaten penalties for violations. What it cannot easily do is tell a Korean company that it must abandon its Chinese customers to a Chinese competitor in order to serve a strategic objective defined in Washington.
The annual licensing system was designed to create a chokepoint. In practice, it created a negotiation. Samsung and SK Hynix need U.S. equipment to upgrade their China fabs. The U.S. government wants to limit the transfer of advanced capability. The outcome is a bargain in which some upgrades proceed, some equipment is approved, and the public posture remains restrictive while the private accommodation is pragmatic.
That is not a failure of enforcement in the sense of rules being broken. It is a failure in the sense of rules being unable to produce the outcome their architects intended. The controls were meant to slow the advancement of China's domestic memory capability. They have not done so. CXMT's 4F² product is on schedule. YMTC is shipping denser NAND. The allied fabs in China are upgrading to stay competitive with them. The controls have changed the cost and the paperwork of operating in China. They have not changed the direction of the industry.
What the Upgrades Signal
The more consequential signal is not about Samsung or SK Hynix. It is about the structure of the global memory industry.
Memory manufacturing is a scale business. The companies that win are the ones that spread fixed costs across the largest possible output. Samsung's Xi'an plant and SK Hynix's Wuxi plant are not peripheral assets that can be left at older nodes while the companies' advanced capacity moves elsewhere. They are integral to the scale economics that make the companies competitive.
Leaving them at older nodes would raise the companies' average cost per chip, reduce their ability to fund the next generation of process technology, and hand the Chinese market to competitors who are already closing the technology gap. No memory executive would choose that outcome voluntarily. The export control regime would have to force it, explicitly and publicly, and the regime has not done so.
The result is a quiet accommodation. The upgrades proceed. The licenses are granted for some steps and not others. The public posture remains restrictive. The industry keeps moving.
Reuters reported what the companies confirmed. The companies confirmed what the market required. The market required what the controls were designed to prevent. The controls did not prevent it.
What This Means
The China fab upgrades are not a loophole in the export control system. They are a demonstration of its limits.
A control regime can restrict the movement of goods across a border. It cannot easily restrict the decisions of companies whose survival depends on serving the market on the other side of that border. When the largest customer and the fastest-growing competitor are in the same country, the rational strategy for an allied manufacturer is to stay close to both — even if that means operating inside a licensing framework that was designed to slow it down.
Samsung and SK Hynix are not the story. The story is the structure of the industry they operate in. Memory is a scale business, China is the scale market, and the competitor Washington is trying to slow is inside that market. The controls can make the upgrades more complicated. They cannot make them unnecessary.
The Reuters report did not resolve how much Washington approved. It did not need to. The upgrades are underway. The companies said so. The market required it. The controls were supposed to stop it.
Sources: Reuters exclusive by Hyunjoo Jin, Liam Mo, and Karen Freifeld, published October 21, 2026, via The Edge Malaysia, Yahoo Finance, Investing.com, and Euronext; TrendForce News (October 21, 2026); DigitalToday (October 22, 2026); TweakTown (October 21, 2026); Yonhap News via Reuters (October 21, 2026); Asia Economy Daily (October 22, 2026); Ming Pao (October 21, 2026); 163.com (October 21, 2026); Baidu Baike entry on SK Hynix Wuxi DRAM fab (July 21, 2026); Asia Economy Daily on SK Hynix Wuxi investment (March 25, 2026); China Energy News on 2025 China investment by Samsung and SK Hynix (March 27, 2026); C114 on combined 1.5 trillion won investment (March 25, 2026); China Star Market on SK Hynix Wuxi and Dalian investment (April 1, 2026); TrendForce 2Q26 DRAM revenue ranking (September 2026); CXMT 4F² announcement via Shanghai Securities News (October 10, 2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on Reuters reporting and company disclosures available as of the article date.
License approvals and equipment-installation details were not confirmed by the companies or by the Bureau of Industry and Security.
Wafer-per-month targets are company plans and may be revised.
License terms can change with policy, and the enforcement boundary described here reflects current practice.
Reporting standards differ across jurisdictions; figures may not be directly comparable.
Market size and contract value figures come from third-party research and are estimates.
This sector is changing quickly; claims can become outdated soon.
Sources
- Reuters exclusive by Hyunjoo Jin, Liam Mo, and Karen Freifeld, published October 21, 2026, via The Edge Malaysia, Yahoo Finance, Investing.com, and Euronext
- TrendForce News (October 21, 2026)
- DigitalToday (October 22, 2026)
- TweakTown (October 21, 2026)
- Yonhap News via Reuters (October 21, 2026)
- Asia Economy Daily (October 22, 2026)
- Ming Pao (October 21, 2026)
- 163.com (October 21, 2026)
- Baidu Baike entry on SK Hynix Wuxi DRAM fab (July 21, 2026)
- Asia Economy Daily on SK Hynix Wuxi investment (March 25, 2026)
- China Energy News on 2025 China investment by Samsung and SK Hynix (March 27, 2026)
- C114 on combined 1.5 trillion won investment (March 25, 2026)
- China Star Market on SK Hynix Wuxi and Dalian investment (April 1, 2026)
- TrendForce 2Q26 DRAM revenue ranking (September 2026)
- CXMT 4F² announcement via Shanghai Securities News (October 10, 2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on Reuters reporting and company disclosures available as of the article date.; License approvals and equipment-installation details were not confirmed by the companies or by the Bureau of Industry and Security.; Wafer-per-month targets are company plans and may be revised.; License terms can change with policy, and the enforcement boundary described here reflects current practice.; Reporting standards differ across jurisdictions; figures may not be directly comparable.; Market size and contract value figures come from third-party research and are estimates.; This sector is changing quickly; claims can become outdated soon.