BofA lowers HBM/LTA enthusiasm, more optimistic on traditional memory and the upcycle in Korean substrates/MLCC

Overview

This is a weekly thematic research report by Bank of America on global memory technology. The report's key shift is that memory makers are becoming less enthusiastic about HBM capacity expansion and long-term LTA sales, while the outlook for traditional DRAM, NAND, Korean substrates, and MLCC remains favorable. The authors believe that earnings growth in 2H 2026 relative to 2Q may come more from traditional memory rather than HBM4.

Core views

First, traditional DRAM, especially LPDDR5, is more profitable than HBM, leading some capex originally intended for HBM to shift toward traditional DRAM and even NAND. Second, HBM4 order growth appears somewhat moderate, while the growth of NAND participation in AI inference KV cache memory solutions weakens the single-track HBM expansion narrative. Third, if memory shortages persist into 2027-2028, long-term LTAs may limit price and profit upside; if the industry turns down, such long-term contracts also carry breach risk. Fourth, Korean substrates and MLCC may enter a new upcycle due to AI server demand, specification upgrades, and capacity utilization.

Analysis framework

The report combines memory makers' capacity plans, price trends, Korean semiconductor exports, Chinese IC imports, DRAM/NAND spot and contract prices, and semiconductor supply chain valuations to judge the direction of industry conditions. The focus is not on single-company ratings, but on thematic judgments around memory supply and demand, price elasticity, capex allocation, and beneficiary links in the AI server supply chain.

Methodology notes

• Industry supply-demand analysisMemory supply-demand and price cycle framework

  Assess the cycle position of DRAM/NAND through wafer capacity, product mix, spot prices, contract prices, and end demand.

  The report argues that incremental wafer capacity may be more constrained in 2028, with each new fab adding at most about 50-60k wpm, implying only single-digit percentage wafer capacity growth for the global DRAM industry and thus supporting the long-term shortage view.

• Capex allocation analysisInvestment substitution relationship between HBM and traditional memory

  Compare the profitability and order visibility of HBM, traditional DRAM, and NAND to determine producers' capex direction.

  Because traditional DRAM is more profitable and HBM4 order growth is slightly slower, some unused HBM capex may shift to traditional DRAM and NAND.

• Price momentum analysisCross-validation of spot and contract prices

  Use changes in DRAM/NAND spot and contract prices to validate the strength of industry conditions.

  16Gb DDR5 spot and contract prices are around US$30-40, significantly above the historical norm of US$3-5; NAND prices are also multiple times above their 2025 lows, though some spot prices began correcting in late March and April.

Asset mapping & comparison

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

• SEMCO / Samsung Electro-Mechanics (009150 KS, SMSGF)

  Beneficiary of Korean MLCC and semiconductor substrates

  Strengths

  May benefit from AI server demand from major U.S. technology companies, rising MLCC prices, new semiconductor substrate orders, and glass substrate commercialization opportunities.

  Weaknesses

  Exposed to internal demand from Samsung Electronics semiconductors and smartphones, among others.

  Comparison

  The report assigns a target price of W1,100,000 based on 2027-28E P/E of 27x, above the long-term historical average, reflecting expectations for above-trend EPS growth.

  Risks

  Downside risks include weaker internal demand and aggressive price cuts by Japanese and Taiwanese competitors in MLCC, substrates, and camera modules.

• LG Innotek (011070 KS, XLGQF)

  Beneficiary related to semiconductor substrates and the Apple supply chain

  Strengths

  May benefit from stronger iPhone demand, a more premium camera module mix, and growing new orders for semiconductor substrates and automotive components.

  Weaknesses

  The Apple business still carries seasonal procurement and end-demand volatility.

  Comparison

  The report assigns a target price of W750,000 based on 2026-27E P/E of 18x, above the historical peak of about 15x during 2020-2025.

  Risks

  Downside risks mainly come from weaker iPhone demand, significant market share losses, or failure to meet iPhone 18 specification upgrades.

• Traditional DRAM / DDR5

  One of the core drivers of earnings growth in 2H 2026

  Strengths

  More attractive profitability relative to HBM, stable DDR5 demand, and strong Korean export data.

  Weaknesses

  Prices are already at historically abnormal highs, leaving room for a pullback.

  Comparison

  DDR5 is cheaper than DDR4 while offering higher specifications; the report believes DDR5's relative outperformance versus DDR4 may continue through June.

  Risks

  If elevated spot prices fall too quickly, industry earnings expectations may be compressed.

• NAND

  Beneficiary of AI inference KV cache and rising memory prices

  Strengths

  NAND contract and spot prices are multiple times above their 2025 lows, and SSD product prices rose significantly in April.

  Weaknesses

  Some NAND spot prices declined slightly in the second half of March and in April.

  Comparison

  The report believes NAND's role in AI inference memory architecture reduces market dependence on a single HBM expansion path.

  Risks

  If prices fall back from abnormally high levels, or if AI inference memory demand disappoints, the durability of favorable conditions may be affected.

Key data

• South Korean semiconductor exportsUp 182% YoY in the first 20 days of April 2026; value about US$18.3bn, down 2% MoM but still near historical highsGrowth was mainly driven by DDR5, HBM, and NAND.
• DDR4 spot pricesDown 5-8% this weekDDR4 carries a premium of more than 50% versus DDR5, and spot demand has weakened.
• 16Gb DDR5 pricesSpot and contract prices around US$30-40Historical norm is about US$3-5, so current levels are abnormally high.
• China IC importsReached US$49.8bn in March 2026, up 54% YoYChina's memory chip imports rose about 130-200% YoY in January-March.
• China memory importsMemory imports in March 2026 were about US$23bn, accounting for 47% of total IC importsThis shows that both memory demand and pricing are driving import value higher.
• Assumption for incremental DRAM capacityEach new fab in 2028 can add at most about 50-60k wpmEven if cleanroom space can support higher capacity, execution risk at advanced nodes makes expansion more cautious.
• LG Innotek valuation basisTarget price W750,000, based on 2026-27E P/E of 18xAbove the previous cycle peak of 15x, reflecting semiconductor substrate growth beyond the Apple business.
• SEMCO valuation basisTarget price W1,100,000, based on 2027-28E P/E of 27xThe report believes EPS CAGR in 2026-2028 may exceed 50%.

Impact & implications

The investment implication is that the market needs to shift from a single HBM capacity expansion narrative to a more balanced framework of memory shortages and AI server supply chain beneficiaries. Traditional DRAM and NAND may contribute the main incremental earnings in 2H 2026; Korean substrates and MLCC may see tighter supply, better pricing, and improved orders due to AI server specification upgrades, larger size per unit, and longer manufacturing cycles. At the same time, the high DDR4 premium and elevated memory spot prices imply that price correction risk has already emerged.

Risks

• DRAM and NAND prices are at abnormally high levels, and spot prices already corrected in late March and April, with possible declines in 2H 2026.
• Slightly weaker HBM4 order growth may weaken market expectations for HBM expansion and equipment demand.
• If LTAs lock in prices during a shortage cycle, they may limit suppliers' price and profit upside; if the industry weakens, there is also contract performance risk.
• There is execution risk in ramping advanced nodes at new wafer fabs, and actual expansion timing may be slower or faster than expected.
• SEMCO and LG Innotek face company-specific risks including customer concentration, weaker end demand, competitor price cuts, and failure in specification upgrades.
• The report discloses that LG Innotek and SEMCO have investment banking or non-investment banking relationships with BofAS and its affiliates, so potential conflict-of-interest disclosures should be monitored.

What to watch

• Whether DDR5's price performance relative to DDR4 continues through June 2026.
• Whether traditional DRAM and NAND can continue driving sequential earnings growth in 2H 2026.
• Whether HBM4 order growth and HBM-related capex continue to cool.
• Whether the signing volume of new LTAs comes in below market expectations.
• Whether Korean semiconductor exports and Chinese memory imports maintain high growth.
• New orders from major U.S. technology companies for Korean substrates and MLCC.
• The actual pace of wafer equipment installation at new fabs in South Korea and the United States in 2028, as well as TCB equipment demand.