UBS reiterates Buy on SK Hynix: Significant de-rating is unwarranted

Overview

This report is UBS's earnings review and updated investment view on SK Hynix. The core conclusion is that the recent sharp share price correction and valuation de-rating are unwarranted: although UBS cut some DRAM ASP assumptions, 2027/2028 operating profit, and the target price, it still believes agentic AI is expanding memory demand beyond HBM, and that the long-term supply-demand structure, LTA signings, free cash flow, and potential buybacks will support SK Hynix in sustaining profitability above historical cycle levels.

Core views

UBS maintains its Buy rating and believes SK Hynix's fundamentals remain solid. First, agentic AI is not only driving HBM but also boosting DDR5/LPDDR5, AI server CPU head nodes, NAND KV Cache, and storage demand. Second, most incremental DRAM wafer capacity is being redirected to HBM, while outside China there is little new NAND capacity, making it difficult for supply to catch up with demand in time. Third, SK Hynix is signing long-term agreements more quickly; this may cap some ASP upside in the short term, but should help stabilize margins and ROE over the long term. Fourth, strong FCF may support the launch of share buybacks in 2H26 and gradually move toward returning 50% of FCF through dividends and buybacks combined.

Analysis framework

The report uses a top-down memory industry supply-demand assessment, segment-level forecasts for SK Hynix, earnings forecast revisions, and a P/BV valuation framework. UBS first evaluates AI's impact on DRAM, NAND, and HBM demand, then combines LTA pricing, capex expansion, HBM capacity, and market share assumptions to assess operating profit, EPS, FCF, and shareholder returns, ultimately deriving the target P/BV and target price from long-term ROE and cost of equity.

Methodology notes

• Valuation methodsP/BV-ROE-CoE

  Target price-to-book valuation based on long-term ROE and cost of equity

  Based on a long-term ROE forecast of 40.2% and a cost of equity of 11.5%, UBS applies a 3.65x NTM P/BV valuation to SK Hynix and derives a 12-month target price of Won3,000,000.

• Industry supply-demandMemory supply-demand analysis

  Memory semiconductor supply-demand and bit demand forecasting

  UBS assesses whether the industry will remain in undersupply through analysis of DRAM, NAND, and HBM bit demand, wafer capacity allocation, HBM ramp-up, and incremental supply constraints.

• Scenario analysisUpside/Downside Spectrum

  Upside, base, and downside scenarios based on ASP and operating profit

  The report presents an upside scenario of Won3.30m, a base scenario of Won3.00m, and a downside scenario of Won800k, with DRAM/NAND ASP growth and operating profit as key value drivers.

• Investment frameworkUBS Thesis Map

  Organizing the investment case around key questions

  The report structures its argument around three key questions: whether the memory undersupply will continue through 2028, whether SK Hynix can maintain its HBM leadership, and whether it will implement meaningful shareholder returns.

Asset mapping & comparison

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

• SK Hynix (000660.KS)

  Covered company; UBS maintains Buy rating

  Strengths

  HBM leadership, AI-driven demand for DRAM and NAND, LTAs improving earnings stability, strong FCF with potential buybacks.

  Weaknesses

  Short-term DRAM ASP upside is dragged by fixed-price LTAs and product mix; capex continues to rise; 2027/2028 operating profit forecasts have been cut.

  Comparison

  The current 1.66x NTM P/BV implies a long-term ROE of about 18.8%-18.9%, below UBS's forecast average ROE of 40.2% for 2027-2031, and also close to the historical framework of 17.7% average ROE after DRAM consolidation in 2012 and before AI.

  Risks

  Memory affordability, customer capex pressure, cyclical corrections, NAND industry structure, volatility in smartphone and tablet demand, volatility in enterprise spending, capital expenditure intensity, and substitution risk from new memory technologies.

• DRAM

  Core SK Hynix business and valuation driver

  Strengths

  Rising demand from AI servers, DDR5/LPDDR5, and HBM, with HBM consuming more front-end DRAM capacity and helping ease supply pressure in traditional DDR.

  Weaknesses

  2Q26 DRAM ASP rose only 30% QoQ, below more bullish scenarios; the fixed-price portion of LTAs suppresses short-term pricing elasticity.

  Comparison

  UBS expects DRAM bit end-consumption growth to increase from 22% in 2026 to 36% in 2027.

  Risks

  Price cycles, constrained short-term ASP upside due to higher LTA coverage, and worsening customer affordability.

• NAND

  Important SK Hynix business line and beneficiary of AI storage demand

  Strengths

  KV Cache and storage demand are driving NAND flash, and there is little new NAND capacity outside China.

  Weaknesses

  The improvement in NAND industry structure is less pronounced than in DRAM, and it remains affected by demand from smartphones, tablets, and enterprise SSDs.

  Comparison

  UBS expects NAND demand growth to rise from 20% in 2026 to 23% in 2027.

  Risks

  Industry competition, 3D NAND capacity strategy, end-demand volatility, and price cycles.

• HBM

  Leading SK Hynix product line supporting the AI memory theme

  Strengths

  UBS expects SK Hynix's 2026 HBM bit shipment share to be 48%, maintaining industry leadership; HBM4 has already begun volume shipment, and negotiations for 2027 and beyond have progressed.

  Weaknesses

  Its 2027 share is expected to decline to 39%, slightly below Samsung's 41%.

  Comparison

  UBS expects industry HBM shipments of 17.2bn Gb in 2026 and 23.0bn Gb in 2027, representing YoY growth of 37% and 34%, respectively.

  Risks

  Competition from Samsung and Micron, timing of HBM capacity build-out, and changes in customer orders and price negotiations.

Key data

• 12-month ratingBuyUBS reiterates its Buy rating.
• 12-month target priceWon3,000,000Cut from the previous Won3,200,000.
• Current priceWon1,401,000As of July 29, 2026.
• Forecast total return115.8%Includes forecast share price upside of 114.1% and forecast dividend yield of 1.6%.
• Share price drawdownDown 52% from the June 22 peakDespite still being up 115% year to date.
• Current valuation1.66x NTM P/BVUBS believes this valuation implies a long-term ROE of about 18.8%-18.9%, below its long-term ROE forecast of 40.2%.
• DRAM bit demand growth2026E 22%, 2027E 36%UBS expects agentic AI to accelerate demand.
• NAND bit demand growth2026E 20%, 2027E 23%NAND demand is supported by AI-related use cases such as KV Cache and storage.
• 2026/2027/2028 capex forecastWon47tn / Won62tn / Won67tnRaised from previous forecasts of Won45tn, Won60tn, and Won63tn, respectively.
• 2027/2028 operating profit forecastWon505tn / Won544tnCut by 19% and 18%, respectively, from previous forecasts.
• 2026/2027/2028 FCF forecastWon188tn / Won320tn / Won374tnUBS believes strong free cash flow can support potential buybacks and higher shareholder returns.
• HBM share forecast2026 48%, 2027 39%UBS expects SK Hynix to maintain the No.1 HBM bit shipment share in 2026, with 2027 slightly below Samsung's 41%.

Impact & implications

The report has a positive investment implication: the short-term earnings forecast cuts mainly reflect higher LTA coverage and lower DRAM ASP assumptions, rather than a breakdown in the demand thesis. If AI-driven memory demand, HBM capacity allocation, and shareholder returns materialize, the current valuation may underestimate SK Hynix's structural earnings power; however, if memory price increases impair customer affordability, hyperscaler capex comes under pressure, or the HBM competitive landscape weakens, there is downside risk to the target price and earnings forecasts.

Risks

• Rising memory product prices may weaken customer affordability, especially by further increasing hyperscaler capex.
• DRAM and NAND still have commodity characteristics, and cyclical corrections may still occur.
• Higher LTA coverage may cap near-term ASP upside.
• Capex forecasts continue to rise, and greater capital intensity may affect free cash flow sensitivity.
• The HBM competitive landscape may change, with UBS expecting SK Hynix's 2027 HBM share to be slightly below Samsung.
• The improvement in NAND industry structure is weaker than in DRAM, and it is affected by volatility in smartphone, tablet, enterprise SSD, and server demand.
• Over the long term, new memory technologies may emerge to replace or complement the current Flash and RAM architecture.
• The report discloses that the covering analyst or members of their team/household hold a long position in the company's common stock, UBS Securities LLC is a market maker in the company's securities or ADRs, and UBS holds a long position of 0.5% or more in the company's listed shares.

What to watch

• Whether share buybacks will begin in 2H26, and whether the 3Q26 earnings call will provide a more complete shareholder return policy.
• Progress in HBM negotiations for 2027 and beyond, HBM4 shipment pace, and major customer orders.
• Changes in DRAM LTA coverage and their impact on ASP, margins, and ROE.
• Yongin 1 clean room equipment move-in in February 2027 and progress on the second clean room.
• Whether the M17 NAND project will begin ramping in 2029.
• Whether hyperscaler capex comes under pressure due to rising memory costs.
• Whether 2026-2028 FCF will approach UBS's forecasts of Won188tn/Won320tn/Won374tn.
• Whether industry HBM capacity reaches the forecast path of 230k wpm by end-2026 and 270k wpm by end-2027.