JPMorgan: YMTC capex expansion has limited near-term NAND supply-demand impact, while HBM and server memory remain relatively tight

Overview

This JPMorgan memory market update focuses on three themes: YMTC's potential NAND expansion, tighter supply-demand from traditional memory price rises, and evolving HBM competitive dynamics. The report believes that while YMTC’s Phase 3 expansion and its planned two new greenfield fabs require ongoing monitoring, near-term NAND supply growth is still constrained by capex discipline from leading vendors. At the same time, AI server demand is lifting HBM, server DRAM, and enterprise SSD, keeping the memory industry’s supply-demand outlook still tight.

Core views

Key views include: first, YMTC’s Phase 3 expansion pace is slightly slower than JPMorgan’s own total-capacity assumptions, and some cleanrooms may be used for low-power DRAM R&D and production, so the impact on NAND supply-demand over the next 12-18 months is limited. Second, Samsung P5 may prioritize DRAM cleanroom allocation first, and if one quarter is allocated to NAND, wafer output may not materialize until 2029. Third, traditional DRAM prices have risen significantly, with tight server demand flowing through to B2C applications like PCs and smartphones, though excessive price increases could curb end demand. Fourth, Samsung is advancing quickly on HBM4 qualification and volume production, leading to a potential upward revision in NVDA share, while overall HBM supply-demand remains constrained due to stronger HBM3E and ASIC demand. Fifth, although memory stocks have already recovered in April, the report remains constructive on a longer-term cycle driven by AI demand.

Analysis framework

The report combines news flow, channel checks, JPMorgan capacity and capex assumptions, CoWoS model updates, price trends, and downstream demand observations to assess NAND, DRAM, HBM, and eSSD across supply-demand balance, price elasticity, and stock catalysts.

Methodology notes

• Industry Supply-DemandS-D Supply-Demand Analysis

  Determine price direction through wafer capacity, bit supply, capex priority, and end-demand.

  The report uses expansion pace, server demand, B2C procurement divergence, and inventory/price signals to determine whether traditional DRAM, NAND, and HBM remain tight.

• Capex TrackingCapacity Expansion and Cleanroom Allocation Analysis

  Analyze YMTC Phase 3, subsequent new plants, and Samsung P5 cleanroom allocation impacts on future output.

  The report emphasizes that expansion announcements are not equivalent to immediate supply release; wafer start timing, cleanroom usage, full-capacity scale, and actual output year all matter.

• Value Chain ModelCoWoS and HBM Procurement Model

  Evaluate HBM supply-demand using GPU output, HBM sourcing, and packaging capacity changes.

  The Rubin GPU output cut is attributed to HBM sourcing issues, but upward revisions to HBM3E and ASIC demand keep overall HBM supply-demand constrained.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Samsung Electronics

  A major supplier of HBM4, DRAM, and NAND that could benefit from higher HBM4 share and tight server-memory conditions.

  Strengths

  Samsung leads in HBM4 ramp progress, and 1cnm DRAM core die, 4nm base die, and back-end yield improvements support higher NVDA share.

  Weaknesses

  P5 capacity allocation still requires balancing DRAM and NAND; the Rubin GPU production cut could affect HBM4 procurement timing.

  Comparison

  The report expects Samsung’s NVDA HBM share could rise to around 35%, up from the prior expectation of about 25%.

  Risks

  HBM qualification may miss expectations, customer procurement plans may be revised, and timing of NAND capacity release may change.

• YMTC

  A NAND-supply-side variable; expansion pace influences global NAND supply-demand expectations.

  Strengths

  Phase 3 wafer starts may be pulled forward, and plans for two additional greenfield fabs indicate expansion intent.

  Weaknesses

  Phase 3 ramp is slightly slower than JPMorgan’s existing capacity assumptions, and some cleanrooms may be repurposed to low-power DRAM R&D and production.

  Comparison

  Compared with top-tier global memory vendors, YMTC expansion is viewed as having limited supply impact over the next 12-18 months.

  Risks

  If subsequent rounds of investment are accelerated or exceed scale expectations, the NAND supply-demand balance could shift.

• SK hynix / Micron / other memory suppliers

  They benefit from tight supply-demand and rising prices in DRAM, HBM, and server memory.

  Strengths

  Strong AI server demand and relatively low incremental flexibility in capex discipline and cleanroom additions support cycle durability.

  Weaknesses

  Excessively high pricing in PC and smartphone products could suppress downstream demand.

  Comparison

  The report says that over the past 16 months, pure NAND, pure DRAM, and blended DRAM/NAND stocks diverged, while memory shares outperformed SOX clearly in April.

  Risks

  End-demand weakness, unsustainable price hikes, and limited visibility into 2027 demand.

• NAND Market

  Affected by YMTC and Samsung expansion news, but near-term supply-demand remains constructive.

  Strengths

  Leading vendors’ capex is more DRAM-biased, and NAND bit supply growth is limited.

  Weaknesses

  If greenfield capacity planned for late 2028 or 2029 is released in a concentrated way, longer-term supply pressure may rise.

  Comparison

  Compared with HBM and DRAM, NAND is more sensitive to expansion-news shocks, while eSSD demand remains strong.

  Risks

  De-specification at smartphone/PC levels, earlier-than-expected greenfield release, and demand compression from price hikes.

• HBM Market

  A premium memory segment driven by AI GPU and ASIC demand, still in shortage.

  Strengths

  HBM3E demand is rising, Blackwell-tier procurement plans are strengthening, and ASIC production plans are stronger.

  Weaknesses

  Rubin GPU cuts show that supply-chain constraints are still affecting customer production planning.

  Comparison

  Unlike traditional memory, HBM supply-demand is more affected by AI compute-platform and advanced packaging constraints.

  Risks

  GPU shipment cuts, lower HBM4 procurement plans, and changes in pricing negotiations or customer shares.

Key data

• YMTC Phase 3 Capacity RampRamp advanced by end-2026, reaching about 50K wfpm in 2027According to press reports; JPMorgan views the pace as somewhat slower than its assumption of YMTC total capacity rising from 160K by end-2025 to 180K by end-2026 and 240K by end-2027.
• YMTC Greenfield ExpansionTwo new plants of approximately 100K wfpm full-scale eachThe report says supply impact may emerge no earlier than the second half of 2028 or 2029, and investment progress needs to be tracked continuously.
• Traditional DRAM Pricing1Q26 up about 70% sequentially; market expects another 30%-50% rise in 2QTight supply-demand is transmitting from server demand to B2C applications such as PCs and smartphones.
• JPMorgan 2026E End-Market Sell-Through AssumptionSmartphones -12% YoY, PC -11% YoYFurther price increases may create pressure on end demand.
• Samsung NVDA HBM Share ExpectationAround 35%, above prior initial expectation of around 25%Based on Samsung’s HBM4 ramp, 1cnm DRAM core die, SF 4nm base die, and back-end yield improvements.
• Rubin GPU Output ForecastRevised down for 2026E from 2.8 million units to 2.2 millionPrimarily due to HBM sourcing issues.
• Initial Rubin-Class HBM4 Procurement Plan10-11bn GbThe Rubin GPU cut implies downside risk to this procurement plan.
• Memory Stock Performance3月-7% vs SOX -2%; 4月+27% vs SOX +18%The April rebound came from improved risk sentiment and EPS beat-and-raise.
• NAND Bit Supply CAGR2025A-2028E about 21%JPMorgan sees leading memory vendors prioritizing capex for DRAM, limiting NAND bit growth.

Impact & implications

From an investment standpoint, the report is more supportive of a longer-term setup of continued supply constraints, rising prices, and margin expansion across memory. HBM, server DRAM, and enterprise SSD remain the strongest demand themes, while NAND expansion has limited near-term supply-side pressure. Potential risks include fast price increases for PC and smartphone memory causing de-specification or demand compression, and Rubin GPU output cuts affecting part of HBM4 procurement plans.

Risks

• Continued price increases in PC and smartphone DRAM/NAND could suppress end demand and trigger de-specification.
• If YMTC or Samsung releases new capacity faster than expected, NAND supply-demand balance could shift.
• Rubin GPU output cuts may weaken the initial Rubin-class HBM4 procurement plan.
• Memory equities have already rallied sharply in the short term; if 2027 demand visibility is weak, valuations may come under pressure.
• If AI server demand or CSP capex slows, the price-upside case for HBM, server DRAM, and eSSD could be affected.

What to watch

• The actual ramp pace of YMTC Phase 3 capacity, cleanroom usage, and the investment timeline for two subsequent greenfield plants.
• How Samsung P5 allocates cleanroom capacity between DRAM and NAND, and whether NAND wafer output appears only by 2029 as expected.
• Whether traditional DRAM in 2Q26 rises a further 30%-50% sequentially.
• Whether brand purchasing patterns in PC and smartphones continue to diverge, and whether end users show de-specification.
• Changes in Samsung HBM4 qualification, ramp yield, and NVDA supply share.
• Whether memory suppliers provide 2027 sell-down guidance over the coming months and whether HBM pricing negotiations begin in advance.
• Whether long-term supply agreements, capital allocation policy, and shareholder-return updates become catalysts for memory equities.