China's Memory Localization and Korea: Decomposing Samsung, SK Hynix and Micron's China Exposure from Disclosures

Sat, 18 Jul 2026 15:00:00 +0900

Context This post is a follow-up to CXMT IPO and Memory Price Risk, which laid out the frame that “HBM risk and client DRAM risk must be separated.” This post quantifies that risk by decomposing the three companies’ China exposure. It pairs well with Is China’s AI API Pricing Sustainable, China’s Open Models and Value-Chain Redistribution, and semiconductor fair value. Related hubs are the AI HBM Hub and the Exclusive Analysis Hub.

TL;DR

The hypothesis that “Korea’s top export item gets displaced by domestic suppliers in its own top market” is directionally sound. But the current stage looks less like a 2026 earnings risk and more like a structural risk that lowers commodity memory ASP, share and fair multiples from 2028 onward.
The most dangerous path is not the disappearance of China revenue itself. The bigger second-order effect is the global ASP decline triggered by surplus volume redirected to other markets.
China exposure as seen in disclosures uses different bases across companies, so a full apples-to-apples comparison is not possible. Samsung Electronics’ 30.1% (company-wide basis), SK Hynix’s 24.3% (sales-subsidiary location) and Micron’s 10.1% (customer headquarters location) are each proxies that measure something different.
Two conventional wisdoms need correcting. The 45% China-plus-Hong-Kong share of semiconductor exports is on a customs-destination basis, so accounting for Hong Kong re-export could mean it overstates true China end demand. “HBM cannot get into China at all” is also too strong a claim: there is a conditional exception for products below a memory bandwidth density of 3.3GB/s/mm².
What makes CXMT different from the past is not the size of the funding but its structure. It is raising about RMB 57.9bn ($8.1bn), with buyers coming in as shareholders. Funding support and customer acquisition are proceeding at the same time. [Analysis scope]

Key Framing

China's memory localization is not a single trigger to sell Samsung Electronics and SK Hynix today. It is the single most important discount variable to build into 2028 earnings and the terminal multiple. The signal that decides it is not China market share, but whether that share gain transmits into a decline in global contract prices.

1. The Hypothesis and the Verdict

First, let’s state the hypothesis precisely. Semiconductors were 38.7% of Korea’s H1 exports, and China is that industry’s largest market. HBM is blocked by sanctions, so most exports to China are commodity DRAM and NAND, precisely the segment ChangXin Memory (CXMT) and Yangtze Memory (YMTC) are targeting directly. Exposure is larger still if you include local production and sales inside China that export statistics do not capture. With Chinese policy capital pouring in on top of that, localization will keep advancing.

The direction is sound. But timing and transmission path need to be separated.

The most dangerous path is not China revenue disappearing.

Chinese suppliers replace imports in China's domestic market
→ Samsung, SK Hynix and Micron divert surplus volume to other markets
→ Global DRAM/NAND ASP declines
→ High fixed costs turn that into a sharp profit hit

This second-order effect is larger than the first-order one. Volume that cannot be sold in China does not vanish; it goes to other markets. Memory has a high fixed-cost ratio, so even a small ASP slip swings profit sharply. [Inference: transmission-path analysis]

2. The Three Companies’ China Exposure: Decomposed from Disclosures

Company	China-related revenue share per disclosure	Disclosure basis	China production base	Read-through
Samsung Electronics	2025: 30.1%	China revenue KRW 71.6tn ÷ standalone revenue KRW 238.0tn	Xi’an NAND	Smartphones, appliances and semiconductors are all bundled together, so the memory-only share is undisclosed
SK Hynix	1Q 2026: 24.3%, 2025: 19.7%	China sales-subsidiary location basis	Wuxi DRAM, Dalian NAND	The most directly verifiable China memory exposure of the three
Micron	FY2025 China 7.1% + Hong Kong 3.0% = 10.1%	Customer headquarters location basis	China runs back-end operations only	Already sharply down from 16.4% in 2024

[Fact: company disclosures]

Samsung Electronics: The Number Is Large, but the Memory Slice Is Unknown

2025 China revenue was KRW 71.6tn, or 30.1% of standalone revenue. But the company does not disclose China revenue for memory alone. Consolidated memory revenue that same year was KRW 104.1tn, but the two figures cannot be cross-referenced to compute China memory revenue. Dividing numbers that use different bases produces a wrong answer. [Blocked]

SK Hynix: The Most Direct Read, but It Rebounded

The trajectory of the China sales-subsidiary revenue share is worth a closer look.

Period	Share
2023	30.9%
2024	23.5%
2025	19.7%
1Q 2026	24.3%

It fell, then climbed back. But that swing is mixed in with the effect of the US share rising on the back of HBM, so it should not be read purely as a change in China exposure itself. [Inference: mix effect]

Micron: A Case Already Lived Through

The China-plus-Hong-Kong share fell from 16.4% in 2024 to 10.1% in 2025. That already reflects the 2023 ban by China’s cybersecurity authority on Micron purchases by critical-information-infrastructure operators. China has no front-end fab, only assembly and test facilities.

The three figures use different bases, so a full apples-to-apples comparison is not possible. Company-wide basis (Samsung), sales-subsidiary location (SK Hynix) and customer headquarters location (Micron) each measure something different. They are, precisely, China-exposure proxies. [Inference: differing bases]

3. Korea’s Export Data: What’s Confirmed and What Needs Caution

What is confirmed: of Korea’s total H1 exports of $496.7bn, semiconductors accounted for $192.4bn. Do the math and it comes to exactly 38.7%. [Fact: Ministry of Trade, Industry and Energy]

$192.4bn ÷ $496.7bn = 38.7%

What requires caution: the 45% China-plus-Hong-Kong share of semiconductor exports is on a customs-destination basis. Hong Kong is a re-export and distribution hub, so reading that figure directly as “45% China end demand” could overstate it. [Inference: statistical basis reading]

That correction does not change the direction, though. China remains Korea’s largest demand and assembly hub for memory.

4. HBM Controls: “Cannot Get In at All” Is Too Strong

The claim that “HBM cannot get into China because of US sanctions” is a bit too strong. Advanced HBM currently requires a license to export to China and is in practice tightly restricted, but there is a conditional exception for some products with a memory bandwidth density below 3.3GB/s/mm². [Fact: US BIS rules]

To be precise:

Product	China export status
HBM3E/HBM4-class	Effectively restricted
Lower-spec HBM (bandwidth density below 3.3GB/s/mm²)	Possible on a limited basis

As a result, revenue directed at China is likely weighted more heavily toward DDR4, DDR5, LPDDR, commodity NAND and SSDs. The core of the hypothesis, that exports to China concentrate in commodity products, still holds after this correction. But company-level product-by-region cross-tables are not disclosed. [Blocked]

5. How China’s Localization Differs from the Past

This is where the real change lies. CXMT has moved past being a simple government-backed project.

Set to raise about RMB 57.9bn, $8.1bn based on final offering terms as of July 2026
Strategic investors include Alibaba Cloud, Xiaomi, ZTE, Transsion, TCL, NIO and Chery, among others
Not just suppliers, but a structure in which actual buyers come in as shareholders

[Fact: Shanghai Stock Exchange CXMT IPO materials]

That last point is decisive. It means funding support and customer acquisition are advancing simultaneously. If China’s past semiconductor support meant “the money is there but there is nowhere to sell,” this time the shareholders are the customers. [Inference: implication of the shareholder roster]

YMTC has also placed Gen4/Gen5 TLC/QLC 3D NAND in its official product lineup. In the China consumer SSD, mobile and embedded markets at least, it appears to have moved past being merely a “cheap substitute.” [Fact: YMTC product list]

6. Company-by-Company Impact

Samsung Electronics: Largest Exposure, but Also the Largest Absorption Capacity

Samsung likely has the largest absolute exposure of the three in both commodity DRAM and NAND. With a NAND fab in Xi’an, it is exposed to both China localization and US equipment controls.

But holding HBM, server DRAM, NAND, foundry and mobile all at once also gives it the largest shock-absorption capacity. China risk is less a threat of a Samsung-wide earnings collapse than a variable that lowers the ceiling on peak memory margins and multiples. [Inference: portfolio effect]

SK Hynix: HBM Is the Shield, but Front-End Assets Are Exposed

In 2026-2027, a high HBM profit share should make SK Hynix the best defended of the three against commodity-memory displacement. But its China revenue share has climbed back to 24.3%, and it also holds front-end assets in Wuxi DRAM and Dalian NAND.

US BIS has laid out a principle of licensing equipment for existing China fab operations while not allowing capacity expansion or technology upgrades. [Fact: BIS rules] If Wuxi and Dalian stay locked into older process nodes while competing with CXMT and YMTC, asset efficiency could deteriorate over the long run. [Inference: asset-aging path]

Micron: Low Direct Exposure, but Not Safe

Micron has the lowest direct China revenue and the lowest China front-end asset risk of the three. That does not mean it is safe, though.

Micron has already been excluded from China’s critical-infrastructure market, and because DRAM and NAND account for such a pure share of its business, earnings sensitivity is high if Chinese-made volume presses down on global ASP. Micron itself states in its 10-K that Chinese government support for CXMT and YMTC could trigger DRAM/NAND oversupply and aggressive price competition. [Fact: Micron 10-K]

Even without selling into China, there is no escaping it if China-made volume presses down global prices. Micron has the lowest direct exposure of the three, yet it may be the most vulnerable on the second-order effect.

7. Timeline Judgment

Period	Expected China impact	Investment read
2026-2027	Expanding replacement of commodity memory in China’s consumer, mobile and PC segments	Caps the multiple more than it breaks earnings
2027-2028	CXMT/YMTC capacity additions and yield stabilization	Whether it transmits to global ASP is the key question
2028 and beyond	Possible expansion into server, automotive, high-performance SSD and HBM qualification	Structural share risk for Samsung, SK Hynix and Micron

[Inference: timeline estimate]

Chinese companies still have to clear qualification and yield barriers in high-performance server DRAM for global hyperscalers, enterprise SSD, automotive-grade reliability and advanced HBM. So the fundraising itself is a warning sign, not a confirmed signal of oversupply.

8. Final Judgment: How to Price It Into the Stock

The hypothesis is directionally correct. But translating it into the current stock price requires making distinctions.

Item	Judgment
2026-2027 EPS	Not yet enough direct evidence for a collapse
2028 EPS	Needs to conservatively reflect China share and global ASP
Fair multiple	A basis for why peak earnings may no longer command as high a multiple as in the past
Most important confirming signal	Whether CXMT/YMTC share gains stop at China’s domestic market or transmit into a decline in global DRAM/NAND contract prices

In the end, the right move is to elevate China’s memory localization from a single trigger for selling Samsung Electronics or SK Hynix today into the single most important discount variable to build into 2028 earnings and the terminal multiple. [Inference: overall judgment]

This conclusion dovetails exactly with the structure computed in semiconductor fair value. That post found the stock price equal to 2026-2028 FCFE plus a normalized 2029 terminal value, and China’s memory localization is exactly the variable that raises the discount rate on that terminal value.

9. Signals to Watch

Global contract-price transmission: does CXMT/YMTC volume press down DDR5, LPDDR and consumer NAND contract prices outside China?
CXMT qualification expansion: does it win qualification with global server and automotive customers beyond China’s domestic market?
SK Hynix Wuxi and Dalian: do they stay locked into older process nodes under BIS rules, and how does asset efficiency change?
Samsung Xi’an: does the double squeeze of equipment controls and China localization show up in results?
Micron’s China share: does it fall further below 10.1%, and does its global-ASP sensitivity rise instead?

Closing

The core of the hypothesis is correct. A structure is beginning in which Korea’s top export item gets displaced by domestic suppliers in its own top market. Buyers joining CXMT’s shareholder roster is the evidence.

But investment judgment needs to separate timing and transmission path. There is not yet enough evidence that China revenue collapses in 2026 and breaks earnings. The more precise risk is the second-order effect in 2027-2028, when volume displaced from China crosses into the global market and presses down ASP, and that shows up first not in earnings but in the durability of earnings and the terminal multiple from 2028 onward.

So what to watch now is not China market-share headlines, but global contract prices. Until transmission is confirmed, China’s memory localization is a discount variable, not a locked-in loss.

This post is analysis compiled from company disclosures (Samsung Electronics’ 2025 annual business report, SK Hynix’s F-1/A, Micron’s FY2025 10-K), Ministry of Trade, Industry and Energy import-export data, Shanghai Stock Exchange CXMT IPO materials, YMTC product materials, and US BIS rules. The three companies’ China revenue shares use different disclosure bases (company-wide, sales-subsidiary location, customer headquarters location) and are proxies that do not allow a full apples-to-apples comparison. Samsung Electronics’ memory-only China revenue and company-level product-by-region cross-tables are undisclosed, and the timeline judgment and transmission-path analysis are estimates as of the time of writing. The stocks mentioned are examples used to illustrate industry structure, not a recommendation to buy or sell any specific stock. Investment decisions and their outcomes are the sole responsibility of the investor.

China Localization on Korea Invest Insights