China's Memory Gambit Threatens Korea's Silicon Throne

The Pincer Movement

China is no longer content to specialize. For years, the division of labor within Chinese memory chipmaking was neat: CXMT made DRAM, YMTC made NAND. The two companies avoided each other, each dominating a separate lane in a market dominated globally by Samsung and SK Hynix.

That arrangement is dissolving.

CXMT is now building NAND research and production lines at its Beijing facility and has reportedly held talks with customers — including AI startup hopefuls — about integrating its NAND into enterprise storage systems. YMTC, already China’s top NAND maker, started sending low-power DRAM samples to customers as early as last April, quietly probing the market it once ceded to CXMT.

The result is a pincer movement aimed squarely at Samsung and SK Hynix, the two Korean giants that together hold nearly half the global NAND market and an even larger share of the fast-growing eSSD segment.

This is not a sudden strategic reversal born of ideological fervor. It is a calculated exploitation of a window of vulnerability — one created by an AI demand shock that has overwhelmed the existing supply landscape and given Chinese incumbents the cover they needed to diversify.

The Demand Explosion That Changed Everything

The numbers are stark. TrendForce data released in early September showed that combined revenue from the top five enterprise SSD (eSSD) manufacturers hit approximately $3.76 billion in the second quarter of 2026 — a 103.6 percent increase over the prior quarter. In three months, the market more than doubled.

The drivers are unmistakable. Generative AI agent services are multiplying data requirements. Cloud service providers are racing to build out data center infrastructure. Nvidia’s GB-series AI server racks are shipping at scale, and each one demands massive amounts of high-capacity NAND storage. Enterprise SSDs are the bottleneck, and the bottleneck is profitable.

But supply has not kept pace. Memory makers have prioritized investment in DRAM and high-bandwidth memory — the chips directly attached to AI accelerators — leaving NAND capacity expansion underfunded. TrendForce projects the supply shortfall will persist through at least the second half of next year.

In normal markets, sustained scarcity rewards incumbents. In this market, it rewards incumbents and invites challengers. CXMT is banking on the latter outcome.

Who Holds What

The current market map tells the story.

In overall NAND flash, Samsung commands 29.3 percent, SK Hynix Group (including Solidigm) holds 18.2 percent, and Micron sits at 15.1 percent. The remaining share is split among Kioxia, Western Digital, and emerging Chinese players.

In the eSSD segment specifically — the enterprise storage tier most relevant to AI workloads — the Korean advantage is even sharper. Samsung holds 35.1 percent, SK Hynix 21.1 percent, and Micron 17.1 percent. Samsung has been pushing hard on 176-layer QLC products and expanding its PCIe 5.0 portfolio. SK Hynix and Solidigm are scaling 321-layer TLC and ultra-high-capacity QLC offerings.

Neither Korean company can afford to lose ground here. eSSD margins are stronger than consumer SSD margins, and the AI server supply chain locks in long-term contracts that compound advantage.

CXMT’s entry into NAND, even at an early stage, threatens to chip away at exactly that position. And YMTC’s DRAM ambitions threaten the other half of the Korean memory business.

The American Complication

While Chinese memory makers expand, the geopolitical winds are shifting the opposite direction for Korean companies.

SK Hynix subsidiary Solidigm is reportedly considering building a NAND fab in the United States. This would be separate from the existing SK Hynix-Intel collaboration to produce memory at Intel’s Ohio facility. Solidigm currently operates only one NAND manufacturing site — its plant in Dalian, China.

The move reflects mounting pressure from the US government for foreign semiconductor companies to onshore production. Tariff risk, export controls on semiconductor manufacturing equipment, and the broader decoupling agenda make heavy reliance on Chinese fabrication increasingly untenable for companies with American customer relationships.

But building a fab in America is enormously expensive and time-consuming. Semiconductor facilities typically take years to come online, and US production costs run significantly higher than in China or Korea. There is also no guarantee that the AI-driven demand surge will sustain through the construction period.

Solidigm has not confirmed a timeline or investment figure. SK Hynix declined to comment beyond confirming that it is “reviewing multiple options.”

The Timing Problem

None of this means Korean dominance is ending tomorrow. CXMT’s NAND R&D line has not been scheduled for commercial mass production.YMTC’s DRAM samples are still samples, not volume product. Solidigm’s US factory remains a proposal. The semiconductor industry runs on multi-year cycles, and none of these projects will materially shift supply and demand before late 2027 at the earliest.

But the direction of travel is clear. China is closing the gap on both sides of the memory market simultaneously, exploiting a moment when Korean companies are being pulled in two directions: by US pressure to onshore and by AI demand that rewards continued investment in advanced nodes.

Samsung sits in perhaps the most exposed position. It is the NAND market leader, the eSSD market leader, and the largest single memory supplier to the AI infrastructure buildout. Defending that position requires holding back Chinese entrants while managing the cost of geographic diversification — a balancing act with no margin for error.

For SK Hynix, the calculus is slightly different. Its HBM business gives it a crown jewel outside the NAND arena, and the Solidigm onshoring play could eventually strengthen its US customer relationships. But losing eSSD market share to a Chinese competitor would be a strategic wound.

The Real Story

The conventional framing of this story is about market share percentages. The real story is about what happens when a second superpower decides that strategic autonomy in semiconductors requires breaking an existing supply chain rather than competing within it.

CXMT and YMTC were never going to stop at their respective lanes. They were always going to expand. The AI-driven demand surge simply gave them the economic cover to do it faster and with less immediate competitive consequence.

Korea built its memory dominance on the assumption that China would remain a mid-tier player, content to specialize and deferred to Japanese and Korean technological leadership. That assumption is now obsolete.

The companies that win this next phase will be the ones that can maintain technological leadership while navigating the geopolitical fracturing of the supply chain. No one is winning both fronts comfortably right now.

The era of Korean memory supremacy is not over. But the guard is changing.