AI is pushing the memory industry toward differentiation.
China is pushing parts of it back toward commoditization.
For Samsung Electronics and SK Hynix, the tension between these two forces may become one of the defining investment questions of the next decade.
The AI boom has increased the strategic value of high-bandwidth memory, advanced DRAM and increasingly sophisticated packaging. Performance, power efficiency, thermal management and customer qualification now matter more than they did in the traditional commodity-memory model.
At the same time, China is building a domestic memory ecosystem of its own. That does not mean Chinese producers are about to displace the technology leaders at the frontier. Perhaps they do not need to.
China only needs to become sufficiently competitive across large portions of conventional memory to change the supply economics facing Samsung, SK Hynix and Micron Technology.
That creates a new semiconductor order for South Korea.
At the high end, AI is making memory more differentiated. At the other end, China's expansion threatens to make conventional memory more competitive.
The future profitability of Korean memory producers may depend increasingly on which side of that divide generates the next dollar of earnings.
China Is No Longer Just a Demand Market
For decades, China played two obvious roles for Korean semiconductor companies.
It was a huge end market. And it became an important manufacturing base. Now a third role is becoming increasingly important: a market destroyer. The concern is not merely technological catch-up. China has repeatedly demonstrated across industries that massive capacity expansion, aggressive pricing and domestic industrial policy can fundamentally alter global supply economics. Memory semiconductors may increasingly face the same pressure.
ChangXin Memory Technologies, or CXMT, has emerged as China's most important DRAM challenger. The significance of CXMT should not be exaggerated.
Technical analysis published by TechInsights in 2026 found that CXMT's G4 DDR5 technology still trailed the three leading DRAM producers — Samsung Electronics, SK Hynix and Micron Technology — by roughly three years.
But the same analysis points to something equally important: CXMT has continued advancing its DDR5 technology despite restrictions on access to semiconductor equipment and materials. https://www.techinsights.com/ko/node/59051
The commercial footprint is expanding as well. TrendForce reported in 2026 that Samsung, SK hynix, Micron and CXMT increasingly dominate the mobile DRAM market, describing an emerging “Big Four” structure. CXMT has strengthened its position particularly through LPDDR4X while attempting to move into more advanced products. https://www.trendforce.com/research/download/RP260527WN
This matters because China does not need technological leadership across the entire DRAM market to affect global economics.
Memory remains a supply-sensitive industry.
A new producer capable of supplying meaningful volumes of competitive memory can pressure pricing in the segments where it participates, even if it remains behind at the technological frontier.
That leads to a crucial distinction:
China does not need to beat Samsung or SK hynix at HBM4 to matter. It only needs to become large enough elsewhere.
The Memory Market Is Beginning to Split
AI is simultaneously pushing the opposite end of the industry toward greater specialization.
HBM is not simply conventional DRAM sold at a higher price. It requires advanced DRAM processes, stacking, through-silicon vias, packaging expertise, thermal management and increasingly close coordination with accelerator platforms.
That technological differentiation creates barriers to entry.
Geopolitics is reinforcing them.
In December 2024, the U.S. Department of Commerce's Bureau of Industry and Security introduced new export controls covering high-bandwidth memory and additional semiconductor manufacturing equipment.
BIS explicitly linked HBM to advanced computing and AI applications and imposed controls intended to limit China's ability to produce advanced semiconductors.
The result is not simply semiconductor “decoupling.” Something more complicated is emerging: fragmentation by technology tier and geopolitical access.
At one end sits an increasingly sophisticated AI-memory ecosystem built around leading accelerators, HBM, advanced packaging and cutting-edge manufacturing.
At another sits a rapidly developing Chinese semiconductor ecosystem attempting to localize more of its own memory supply.
Those ecosystems are not completely separated.
But neither are they operating under the frictionless global model that defined much of the previous semiconductor era.
Export Controls Cut Both Ways for Korea
For Korean semiconductor companies, U.S. export controls create an unusual strategic problem.
Samsung and SK Hynix are major participants in the U.S.-centered technology ecosystem. But both also built significant manufacturing operations inside China, though they have been trying to escape by lowering down manufacturing capacity.
These factories created a problem that did not exist when semiconductor globalization was at its peak. How can Korean companies continue operating large Chinese fabs when the most advanced semiconductor equipment increasingly falls under U.S. export restrictions?
For a time, Washington created a workable bridge. In October 2023, BIS updated Validated End-User authorizations for Samsung's and SK hynix's Chinese semiconductor facilities, allowing approved items to reach those fabs under a general authorization rather than requiring repeated individual licenses.
But that arrangement changed materially in 2025. In August 2025, BIS announced that it would remove the special VEU treatment enjoyed by certain foreign-owned semiconductor fabs in China. Samsung China Semiconductor and SK Hynix Semiconductor (China) were subsequently removed from the VEU list.
The distinction in the new policy is critical.
BIS said it intended to approve licenses necessary for former VEU participants to operate their existing fabs, but did not intend to approve licenses that would allow them to expand capacity or upgrade technology at those Chinese facilities.
That changes the strategic meaning of Samsung's and SK Hynix's Chinese manufacturing footprints.
The issue is no longer simply whether those fabs can continue producing memory. The more important question is: How far can those fabs move up the technology curve?
China Is Four Things at Once
This is why conventional geopolitical narratives often miss the complexity of South Korea's semiconductor position.
China is not simply a rival.
For Samsung and SK Hynix, China is simultaneously:
a market
a manufacturing base
a competitor (or market destroyer)
and a geopolitical constraint.
The market relationship remains economically significant.
Official Korean government data show that Korea's ICT exports to China including Hong Kong reached $23.08 billion in the first half of 2026, up 192.4% from a year earlier. Korea's semiconductor exports overall also reached record levels as AI and server-memory demand supported both volumes and prices. https://english.motir.go.kr/eng/article/EATCLdfa319ada/2688/view
This is why framing Korea's semiconductor strategy as a simple choice between Washington and Beijing is misleading.
South Korea's challenge is not merely to choose one market over another. It is to manage increasingly different economic roles inside increasingly segmented technology ecosystems.
CXMT Does Not Need to Win HBM to Change the Cycle
The same distinction matters when evaluating China's HBM ambitions.
There is evidence of progress. TrendForce reported in early 2026 that Chinese memory companies were accelerating HBM development, with CXMT pursuing more advanced generations and packaging technologies.
But the same research described Chinese HBM development as remaining in relatively early verification stages, with technical barriers and equipment restrictions limiting its immediate impact on the global market. https://www.trendforce.com/research/download/RP260120JO3
This is exactly where investors should avoid a binary question.
The relevant question is not:
Can CXMT beat SK Hynix in HBM tomorrow?
That sets the bar too high.
The more important question is:
Can China absorb an increasing share of the conventional memory market while gradually narrowing the technology gap?
Evidence from mobile DRAM suggests that the process is already underway.
TrendForce has highlighted CXMT's growing position in LPDDR4X and its efforts to move toward higher-capacity DDR5 and eventually HBM. https://www.trendforce.com/research/download/RP251112MX
That creates a potentially uncomfortable future for the incumbent leaders. Samsung, SK Hynix and Micron Technology can devote more leading-edge capacity to AI and server memory.
China can capture more demand in mature or increasingly localized segments.
The global memory market then becomes less homogeneous. And that may make the old concept of a single memory cycle increasingly incomplete.
China May Actually Accelerate Korea's Move Up the Value Chain
There is an important paradox here.
China's rise is clearly a competitive threat. But it may also reinforce the strategic direction Korean memory producers are already taking.
If Chinese capacity compresses margins in more commoditized memory products, Samsung and SK Hynix have an even stronger incentive to migrate capital and engineering resources toward areas where barriers to entry are higher:
HBM
advanced server DRAM
high-capacity enterprise memory
advanced packaging
custom memory solutions
and products more deeply integrated into AI systems.
In other words:
China may compress margins in commodity memory while simultaneously accelerating Korea's migration toward differentiated AI memory.
The Korean memory industry's future will depend less on defending every segment of the old market and more on maintaining technological leadership where memory becomes a system-level constraint.
AI and China Are Pulling Memory in Opposite Directions
This brings the argument back to the central tension of this series.
AI is pushing memory toward differentiation. China is pushing important portions of memory toward greater commoditization and domestic substitution.
The forces work in opposite directions.
AI can increase:
product differentiation
customer qualification requirements
technological barriers
pricing power
and potentially
sustainable return on equity.
Chinese expansion can increase:
available capacity
price competition
domestic substitution
and eventually cyclical pressure in conventional memory.
Neither force is likely to eliminate the other. That means asking whether “the memory cycle” is dead misses the more interesting development.
There may increasingly be more than one memory cycle.
An AI-memory cycle shaped by accelerator roadmaps, hyperscaler investment, qualification and advanced packaging.
And a more traditional memory cycle shaped by capacity, inventory, consumer electronics and price competition.
The same company can participate in both.
The New Korean Semiconductor Order
This is the environment Samsung Electronics and SK Hynix now face.
Their competitive advantage no longer comes simply from producing DRAM and NAND at enormous scale.
Scale still matters. Manufacturing efficiency still matters.
But the value of technological position is increasing.
SK Hynix's recent financial performance illustrates how powerful that transition can become.
In 2025, HBM revenue more than doubled year over year, helping the company achieve record annual revenue of KRW 97.15 trillion and operating profit of KRW 47.21 trillion.
At the same time, the competitive pressure beneath the premium tier is not disappearing.
CXMT is expanding. China is localizing.
U.S. restrictions are fragmenting technology access.
And Korean companies' own Chinese fabs face tighter constraints on future technological upgrading.
The new order therefore does not eliminate the old cycle.
It stratifies it.
At the top, memory increasingly behaves like a differentiated strategic technology. Further down the stack, it can remain brutally cyclical.
Why This Changes the Valuation Question
This distinction matters enormously for investors.
(refer to the Part II: https://www.geomarketsignal.com/2026/08/part-ii-memory-cycle-is-not-dead-why-pe.html)
If Samsung Electronics or SK Hynix earns a growing proportion of its future profits from technologically differentiated AI memory, those earnings should not necessarily be valued in the same way as profits generated by a commodity DRAM shortage.
But neither should investors simply assign an “AI multiple” to the entire company.
China is a reminder of why. Conventional memory still faces supply competition. Capital expenditure still matters. Technology diffuses. Competitors catch up. And extraordinary margins attract extraordinary investment.
So the key valuation question becomes more precise:
What proportion of future earnings comes from commoditized memory — and what proportion comes from differentiated AI memory?
That distinction may ultimately determine how much of today's profitability deserves a higher structural valuation and how much should still be discounted as cyclical.
It also suggests why Samsung Electronics and SK hynix may no longer deserve the same valuation framework.
Their product mixes are different.
Their AI-memory positions are different.
Their capital structures are different.
And their exposure to other semiconductor businesses is very different.
The next step, therefore, is not to declare one company cheap and the other expensive.
It is to measure how these differences translate into sustainable returns.
In Part IV, we return to valuation and ask whether Samsung Electronics and SK hynix deserve a new P/E–P/B equilibrium — and whether that equilibrium should be the same for both companies.
The AI Memory Cycle & Korean Semiconductors
This article is Part III of the GeoMarketSignal AI Memory Cycle series.
← Previous: Part II — The Memory Cycle Is Not Dead: Why P/E and P/B Send Opposite Signals
Next → Part IV: Samsung vs. SK Hynix: Has AI Created a New Valuation Equilibrium?