Global memory market Report Interpretation

Overview

This report examines why global memory-market fundamentals and memory share prices have diverged. Nomura argues that AI data-center demand, supply constraints and the emergence of long-term agreements support a longer and more stable upcycle than investors currently price into Korean memory makers.

Core views

Nomura argues that the memory shortage is intensifying rather than easing. Data-center memory demand, which began to surge in 3Q25, is expected to remain elevated through 2027F. It estimates that only 70%-80% of data-center customer demand and about 50% of non-data-center demand can be fulfilled. Customers therefore may have to reduce memory content per device or curtail end-product shipments. For cloud service providers, however, unfulfilled memory demand is more likely to be deferred because memory capacity directly constrains data-center capacity. Nomura also notes that if Nvidia and Google had retained their initially contemplated 16hi/12hi HBM specifications, 2027F HBM demand would have risen by more than 80% year on year and commodity DRAM output would have declined year on year. The report sees long-term agreements, or LTAs, as a structural change in the industry. Nomura assumes they will cover roughly 50%-70% of total DRAM including HBM, NAND and SSD sales volumes, principally involving 10-20 counterparties such as cloud service providers and major US technology companies. It expects open-market prices to remain the benchmark, with tight supply keeping open-market prices higher for longer and an LTA ceiling at or up to about 20% above current prices. Typical agreements are assumed to run about five years, with firmer pricing for the first three years and more flexibility in the final two. Prepayments are estimated at roughly 20%-30% of expected contract revenue, while customers that decommit may forfeit about one year's purchase value. Nomura believes these provisions improve revenue visibility, make decommitment more difficult and give suppliers one to two years of warning to adjust supply, although incomplete contract disclosure means consensus forecasts are less reliable. Nomura contends that the required supply response is far larger than the market assumes. With memory demand growth moving from a historical 15%-20% range to more than 30%-40%, raising supply by 15 percentage points would require wafer-capacity growth above 20%, versus roughly 5% recently. Starting from combined DRAM and NAND capacity of about 3.6mn wafers per month, capacity would need to double to 7.2mn within four years and rise to 11mn within six years—an increase of almost 6.4mn wafers per month, equivalent to about three Pyeongtaek-scale fabs every year. Nomura therefore expects tight conditions through 2028F and considers a broad supply-demand soft landing unlikely before 2029F. Chinese suppliers' expansion is not viewed as an immediate solution: their likely addressable market is estimated at around 15% of global memory demand in 2027F because US big-tech data centers are expected to represent 75% of total demand, rising to 85% when Apple and Samsung mobile-device demand are included. On demand, Nomura believes consensus understates cloud-service-provider capital expenditure and operating cash flow. While Street consensus expects CSP capex to grow 30%-40% year on year in 2027F, hardware orders to Nvidia and memory vendors are showing at least 60% growth; Nvidia has guided to at least 70% growth in 2027E, constrained by component supply. Nomura expects TPU demand to grow even faster than Nvidia demand in 2027F. It argues that improving cloud-service pricing should allow EBITDA and operating-cash-flow margins to grow faster than operating profit if big-tech operating margins hold, with future earnings releases serving as a catalyst as investment converts into revenue and profitability is demonstrated. The market has nonetheless priced memory stocks as though earnings are near a conventional cyclical peak, in Nomura's view. After falling roughly 50% from their peak and then rebounding about 20% from the trough, memory stocks remained 37% below the peak and Korean memory makers averaged about 3x 2027F P/E on Nomura estimates. The report sees investors as divided between concerns over slowing second derivatives of earnings growth and confidence in durable cash generation. Nomura expects a growth-rate soft landing rather than a sharp negative-growth phase, followed by re-rating as earnings growth drives shareholder returns and investors gain greater clarity on LTA terms and CSP investment profitability. For Samsung Electronics, Nomura reiterates Buy and its KRW670,000 target price. The target is based on 4.3x target P/B, reduced from 5.0x, applied to 12-month-forward BVPS of KRW153,584; the lower multiple reflects an expected ROE soft landing from deleveraging and a higher risk-free rate. The current price is KRW250,500, with implied upside of 167.5%. Nomura expects 3Q26F operating profit of KRW10.7tn, slightly below its prior estimate, and forecasts a KRW1.3tn operating loss for the MX division. It raises its commodity DRAM price-growth estimate for 4Q26F to 10% quarter on quarter from 5%, lifts 2027F commodity DRAM price growth to 28% year on year from 18%, and raises its 2027F HBM price-growth forecast from 72% to 97%. Nomura forecasts Samsung operating profit of KRW64.0tn in 2027F and KRW81.0tn in 2028F, up 69% and 26% year on year, respectively. It expects 2026F/27F shareholder-return yields of 8%/14%, and estimates that maintaining the existing return framework could take shareholder returns above KRW20tn after 2027, nearly double the 2026F level. For SK Hynix, Nomura reiterates Buy and its KRW4,700,000 target price. The target is based on 5.6x target P/B, reduced from 6.0x, applied to 12-month-forward BVPS of KRW836,330; this implies 8.1x target P/E against a current 12-month-forward P/E of 2.8x. The current price is KRW1,613,000, with implied upside of 191.4%. Nomura lowers its 3Q26F operating-profit estimate to KRW7.7tn from KRW8.6tn because of KRW appreciation, but expects stronger memory prices to leave the broader earnings outlook little changed. It raises Hynix's 2027F commodity DRAM ASP by 8% to 28% year-on-year growth, while revising 2027F HBM price growth to 79% year on year. It forecasts 2027F/28F operating profit of KRW47.3tn/KRW60.8tn, up 73%/28% year on year. Nomura expects shareholder returns above KRW7.4tn in 2026F and KRW15.8tn in 2027F, implying respective shareholder-return yields of 4% and 9%.

Analysis framework

Nomura first assesses supply-demand conditions across DRAM, HBM and NAND, then models the effect of AI data-center demand, capacity constraints and long-term agreements on prices and profitability. It compares these fundamentals with market valuations and investor expectations, before applying company-specific earnings forecasts, P/B-based target prices and shareholder-return assumptions to Samsung Electronics and SK Hynix.

Methodology notes

• Industry AnalysisSupply-demand framework

  Memory supply-demand analysis

  Nomura compares AI-led demand growth with wafer-capacity, utilization, bit-per-wafer and inventory forecasts to assess the persistence of shortages and pricing power.

• Industry AnalysisVolume-price decomposition

  Memory ASP, shipment and margin forecasting

  The report separates expected price, shipment and cost changes for commodity DRAM, HBM and NAND to derive revenue, operating profit and margin forecasts.

• Valuation methodsPB valuation

  Target P/B valuation

  Nomura derives Samsung Electronics and SK Hynix target prices by applying target P/B multiples to 12-month-forward BVPS, adjusting multiples for expected ROE soft landing and higher risk-free rates.

• Corporate Fundamentals and FinanceROIC–WACC spread

  ROE and risk-free-rate adjustment to target multiples

  The report links lower target P/B multiples to an expected soft landing in ROE and a higher risk-free rate when assessing fair valuation.

Asset mapping & comparison

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

• Samsung Electronics (005930 KS)

  Covered memory producer expected to benefit from sustained memory-price strength and shareholder returns.

  Strengths

  Buy maintained; Nomura expects 2027F/28F operating profit growth of 69%/26%, higher DRAM and HBM price assumptions, and 2026F/27F shareholder-return yields of 8%/14%.

  Weaknesses

  KRW strength is expected to reduce near-term earnings and pressure the MX, smartphone and home-appliance businesses.

  Comparison

  Nomura cites 12-month-forward P/B of 1.6x and 2027F P/E of 2.9x versus its target P/B of 4.3x.

  Risks

  US semiconductor tariffs, data-center construction delays due to power shortages, and slower AI capex in a higher-rate environment.

• SK Hynix (000660 KS)

  Covered memory producer expected to benefit from tight DRAM and NAND supply, higher commodity DRAM ASPs and sustainable shareholder returns.

  Strengths

  Buy maintained; Nomura forecasts 2027F/28F operating profit growth of 73%/28% and shareholder returns above KRW7.4tn/KRW15.8tn in 2026F/27F.

  Weaknesses

  KRW appreciation reduces KRW-denominated revenue while much of the cost base remains KRW-denominated; 3Q26F operating profit was cut to KRW7.7tn from KRW8.6tn.

  Comparison

  The target P/B is 5.6x versus current 12-month-forward P/E of 2.8x; target P/E is 8.1x.

  Risks

  US semiconductor tariffs, data-center construction delays due to power shortages, and slower AI capex in a higher-rate environment.

Key data

• LTA volume coverage50%-70%Nomura estimate of total DRAM including HBM, NAND and SSD sales volumes ultimately covered by LTAs.
• Data-center demand fulfillment70%-80%Expected fulfillment rate for data-center customers through 2027F; non-data-center fulfillment is estimated at about 50%.
• Samsung Electronics target priceKRW670,000Buy maintained; current price KRW250,500 and implied upside 167.5%.
• SK Hynix target priceKRW4,700,000Buy maintained; current price KRW1,613,000 and implied upside 191.4%.
• Samsung 2027F/2028F operating profitKRW64.0tn / KRW81.0tnNomura forecasts growth of 69% / 26% year on year.
• SK Hynix 2027F/2028F operating profitKRW47.3tn / KRW60.8tnNomura forecasts growth of 73% / 28% year on year.
• Samsung 2027F commodity DRAM price growth28% year on yearRevised from 18%; 4Q26F growth assumption raised to 10% quarter on quarter from 5%.
• Memory-stock valuation~3x average 2027F P/ENomura estimate for Korean memory makers after the sector remained 37% below its peak.

Impact & implications

Nomura believes the market is discounting a traditional cyclical downturn even though the report's supply-demand analysis points to multi-year pricing and cash-flow support. It sees clearer LTA disclosures, proof of CSP investment profitability and sustained shareholder returns as the main paths to a re-rating of Samsung Electronics and SK Hynix.

Risks

• A macro shock, especially sustained global interest-rate increases, could weaken highly financed AI investment.
• Rapid KRW appreciation could reduce profitability and price competitiveness for Korean memory makers; Nomura estimates a 10% appreciation could cut profits by about 12%.
• US semiconductor tariffs could raise costs for cloud providers and end users, reducing memory demand per unit of investment.
• Data-center construction could be delayed by power-supply shortages.
• Chinese producers could become more disruptive if US restrictions loosen or equipment localization accelerates, weakening industry discipline.

What to watch

• Disclosure of more specific LTA pricing, volume, prepayment and decommitment terms, potentially from year-end onward.
• Quarterly earnings from large cloud-service providers for evidence that AI capex is producing revenue and sustained profitability.
• CSP capex orders and AI hardware demand, including Nvidia and TPU demand trends.
• Memory wafer-capacity additions, especially from Chinese producers and post-2029 expansion plans.
• Global interest-rate trends, KRW appreciation and the timing or scope of potential US semiconductor tariffs.