The Global Memory Semiconductor Race: SK hynix, Samsung, Micron and China’s CXMT

Semiconductor Intelligence · Special Report · October 31, 2026

The Global Memory Semiconductor Race: SK hynix, Samsung, Micron and China’s CXMT — market structure, AI-driven HBM demand, competitive dynamics, export controls, and the evolving balance of global supply.

Special Report

The memory industry has split into two concurrent contests. One is the race among SK hynix, Samsung, and Micron for leadership in high-bandwidth memory that powers AI accelerators. The other is the rise of China’s CXMT as a fourth-scale supplier of conventional DRAM, expanding into the space the three leaders partially vacated while they prioritized HBM.

Both contests are shaped by the same forces: multi-year demand visibility from hyperscalers, record capital expenditure, long-term supply agreements, and export-control constraints on technology and equipment. This report examines the market structure, competitive dynamics, and forward implications without treating any single company profile as the primary subject.

1. Global DRAM and NAND Market Structure

As of the second quarter of 2026, the DRAM revenue ranking stood approximately as follows (TrendForce / Counterpoint):

Samsung — ~39.4% (clear #1)

SK hynix — ~24.9%

Micron — ~23.3%

CXMT — ~9.5–10% (fourth, and the only Chinese supplier at meaningful scale)

Samsung remains the only company ranked first in both DRAM and NAND. NAND is still dominated by Samsung, SK hynix, Micron, and Kioxia, with China’s YMTC as a growing fourth player. The critical structural change is the divergence between conventional DRAM (where CXMT is gaining share) and HBM (still a three-player market).

2. AI-Driven HBM Demand

High-bandwidth memory has become the binding constraint and the highest-margin product in the memory complex. Hyperscaler and AI accelerator demand (notably platforms associated with Nvidia’s Vera Rubin architecture and successors) has outstripped available supply. Both SK hynix and Samsung have described HBM volumes as effectively sold out for multi-year periods under long-term agreements.

SK hynix stated that DRAM and HBM output is effectively sold out through 2029 when existing LTAs and volumes under negotiation are combined; LTAs already cover more than half of 2027–2028 capacity with 14 key customers. Samsung reported that its planned 2027 HBM supply is sold out at target pricing, with additional customer requests still arriving. Customer order volumes for HBM have been described as exceeding available supply by more than two times at the leading suppliers.

Management commentary from both Korean leaders points to supply tightness extending into 2028–2030. Independent verification will depend on actual shipment and ASP data in coming quarters.

3. Competitive Dynamics among the Major Manufacturers

HBM ranking (Q2 2026 revenue share, Counterpoint): SK hynix 50%, Samsung 33%, Micron 18%. Samsung has narrowed a once-dominant SK hynix lead (64% a year earlier) through aggressive HBM4 ramp. Some third-party estimates for Q3 2026 suggest Samsung may have overtaken SK hynix on absolute HBM revenue in that quarter; these remain analyst estimates pending full-year and Q3 share data.

Product transition: HBM4 is becoming the flagship. SK hynix expects HBM4 to surpass HBM3E in both shipments and revenue in Q4 2026. Samsung reported HBM4 revenue more than tripling quarter-on-quarter in Q3 and accounting for roughly 60% of its HBM sales. Both are advancing HBM4E toward 2027 mass production. Micron remains the third qualified supplier.

Conventional DRAM: While the three leaders reallocate capacity and engineering toward HBM, CXMT has expanded its share in mobile and commodity server DRAM. That reallocation is one reason conventional DRAM pricing has remained elevated and why a fourth supplier now matters for global supply balance.

4. China’s Domestic Memory Manufacturing Ambitions

CXMT is the primary instrument of China’s DRAM self-sufficiency effort. It has reached commercial scale in conventional DRAM (G5 platform in mass production, ~10% global revenue share), completed a large domestic IPO, and is expanding wafer capacity toward the mid-to-high hundreds of thousands of wafers per month later in the decade.

Its HBM effort remains early-stage. Reported plans to begin limited HBM3E production in late 2026 / early 2027 come with yields cited in the 20–30% range versus roughly 70% at the leaders, and with performance in testing described as closer to prior-generation product. Limited production at low yield does not yet constitute competitive commercial HBM supply.

The strategic effect is already visible: CXMT supplies a growing share of conventional DRAM, particularly to domestic Chinese customers, while the three leaders focus capital and capacity on HBM. That divergence influences global pricing and availability in non-HBM segments.

5. Semiconductor Export Controls and Supply-Chain Dependencies

U.S. and allied export controls restrict China’s access to the most advanced lithography (especially EUV), certain advanced HBM products, and related manufacturing technology. These measures slow CXMT’s path to process parity and to competitive HBM yields. They also reinforce a bifurcated market: leading-edge HBM for non-restricted customers remains concentrated among SK hynix, Samsung and Micron; Chinese AI developers face tighter and more expensive memory supply.

Korean and U.S. memory makers with fabs or operations in China face their own licensing and capacity-expansion constraints under the same regime. The net result is slower technology diffusion into China and higher strategic value attached to domestic Chinese capacity, even when that capacity trails on yield and node.

6. HBM4 and Next-Generation Memory Technologies

HBM4 is the current commercial battleground. Higher stack heights, higher bandwidth, and tighter integration with AI accelerators define the generation. HBM4E is the next step, with mass-production timelines pointed at 2027 for the leaders. Beyond HBM, process density improvements (including 4F² cell architectures under development) and advanced packaging remain central to cost and performance.

Larger die sizes and more metal layers in successive HBM generations themselves act as a physical constraint on bit output: more wafer area and process complexity per stack reduce the effective supply growth that can be achieved from a given CapEx dollar.

7. Capital Expenditure and Manufacturing Economics

CapEx is rising sharply. Samsung raised 2026 semiconductor-related spending substantially (total company CapEx guidance around KRW 84 trillion, with the large majority in semiconductors). SK hynix has signaled significantly higher 2027 CapEx to support advanced nodes, HBM, and new fab space (including Yongin). Micron continues large multi-year U.S. and other capacity programs. CXMT is funding expansion from IPO proceeds, operating cash flow, and state-linked capital.

Long lead times from fab construction to meaningful wafer output mean that even elevated 2026–2027 spending will not fully relieve tightness in 2027–2028. A large share of new spend also goes to technology conversion and higher construction costs rather than pure bit-growth capacity. Operating margins at the leaders have reached levels (70–80%+ in peak quarters) rarely seen in manufacturing, reflecting ASP strength and HBM mix.

8. The Evolving Balance of Global Memory Supply

The industry is no longer a single undifferentiated DRAM/NAND market. It is a multi-layer structure:

• Leading-edge HBM — three suppliers, multi-year sold-out conditions, extreme margins, tight customer qualification.

• High-value server DRAM — still dominated by the three, but with rising competitive pressure from CXMT on density and price.

• Mobile and commodity DRAM — where CXMT’s share gains are most visible and where the leaders’ HBM prioritization has the largest secondary effect on availability and pricing.

• NAND — still a separate contest, with its own capacity and pricing cycle, less directly tied to the HBM boom.

The near-term equilibrium is one of constrained HBM supply, elevated conventional DRAM pricing, and a Chinese supplier that is commercially relevant in conventional products but not yet in leading-edge HBM. The medium-term variables are HBM4/HBM4E execution and share among the three, the pace of CXMT yield improvement on both G5/4F² and HBM, and the duration of hyperscaler AI CapEx.

Forward-Looking Industry Implications

1. HBM leadership is contestable again. SK hynix’s earlier dominance has narrowed; Samsung’s HBM4 ramp and CapEx intensity make share rankings fluid through 2027.

2. Conventional DRAM is no longer a pure three-player game. CXMT’s scale and domestic customer base change supply and pricing dynamics even if it remains a generation behind on HBM.

3. Capacity additions lag demand signals. Fab lead times and the bit-density cost of successive HBM generations limit how quickly elevated CapEx translates into relief.

4. Geopolitics hardens market bifurcation. Export controls sustain a technology and yield gap while simultaneously creating a protected demand pool for Chinese suppliers.

5. Profitability peaks are real but cycle-sensitive. Record margins rest on ASP strength and mix; any sustained slowdown in AI infrastructure spend, or successful dual-sourcing by customers, would test the durability of those margins.

Sources & Methodology Note

Market-share figures draw on TrendForce and Counterpoint Research data for Q2 2026 (DRAM and HBM revenue shares). Company statements on sold-out status, LTA coverage, HBM4 transitions, and CapEx are taken from SK hynix and Samsung Q3 2026 earnings commentary (October 28–29, 2026). CXMT capacity, G5, and HBM status combine company announcements with contemporaneous Reuters and industry reporting. Analyst estimates of Q3 HBM revenue rankings are labeled as estimates.

Confirmed commercial production is distinguished from limited production, samples, and roadmap targets. No unsupported technology-leadership or market-share claims are asserted beyond dated third-party estimates or company-reported product transitions.

Editorial Note

This Special Report is provided for informational and educational purposes. It is not investment advice, a recommendation, or a forecast of future prices or share rankings. Competitive positions and supply-demand balances can shift with subsequent disclosures and CapEx execution.

THE CODEW · Semiconductor Intelligence · Special Report · October 31, 2026