Korea Memory Pair — SK Hynix vs Samsung Electronics

Investment Thesis · As-of 2026/06/03

Underlying analyzed: SK Hynix (000660.KS) and Samsung Electronics (005930.KS) Expression vehicles in HK: CSOP 2x SK Hynix Daily (7709.HK) and CSOP 2x Samsung Electronics Daily (7747.HK) — see §6 for the L&I product note. Currency convention: KRW unless noted. KRW/USD ≈ 1,380 used for USD reconciliation.

1. Snapshot

SK Hynix (000660.KS) Samsung Electronics (005930.KS)
Last price 2,360,000 KRW 360,500 KRW
Day chg -0.1% +3.3%
Market cap ₩1,675T (~US$1.21T) ₩2,367T (~US$1.72T)
52w range 196,900 – 2,363,000 (now 0% off high) 53,900 – 360,500 (now 0% off high)
% above 200-DMA +199% +144%
ATR(14) daily ~144,000 KRW (~6%) ~21,600 KRW (~6%)
Beta (5y) 2.03 1.33
Forward P/E (Yahoo) 6.2x 6.5x
Run-rate P/E (Q1×4) 10.4x 12.8x
Net cash +₩32.6T +₩119.3T
FY25 OpMargin 48.6% (memory-pure) 14.5% (mixed)
Q1 26 OpMargin 71.5% 42.8%
Sell-side: Strong Buy/Buy/Hold/Sell/SS 10/26/2/0/0 11/25/0/0/1
Mean target / upside 2,077,000 / -12% 397,800 / +10%
Next earnings 2026-04-22 (Q2 26) 2026-07-29 (Q2 26)
Headline read: the easy-money phase is behind both names — they sit at 52w highs after going up ~12x (Hynix) and ~7x (Samsung) off lows. But Q1 26 results just printed historic operating margins (Hynix 71.5%, Samsung 42.8%), so run-rate P/Es are still 10-13x on annualized Q1, not the >100x trough P/Es of FY24. Whether you go long here is a question about HBM cycle longevity, not about price chasing.

2. The Core Bull Thesis (what the price reflects)

One-line: AI training and inference demand has structurally repriced HBM (high-bandwidth memory), and Korea's two memory makers have become the most direct, liquid expressions of the AI capex cycle outside Nvidia/TSMC.

Five supporting pillars:

  1. HBM is supply-constrained, multi-year. Each Nvidia/AMD AI accelerator carries 6–8 HBM3E/HBM4 stacks (vs 0 in a CPU server). 2026 hyperscaler capex is tracking >US$500B; HBM bit demand growing >65% YoY, supply growing ~50% YoY. SK Hynix has held a >55% HBM share through HBM3E.
  2. Pricing power is real, not just unit growth. Hynix's gross margin went from 50% (FY25) to 79% in Q1 26 (gross profit ₩41.7T on revenue ₩52.6T). Samsung's group OpInc went from ₩6.7T in Q1 25 to ₩57.2T in Q1 26 — an 8.5x ramp in four quarters. This is the pricing-power signature of a structurally tight market.
  3. Capacity discipline. Hynix capex FY25 was ₩28.6T (vs ₩16.7T in FY24, ₩8.8T in FY23 — the cycle bottom). Samsung capex FY25 ₩52.2T (FY24: ₩53.7T). They are not building like 2022; they are building behind validated Nvidia/AMD/Google qualifications.
  4. Cycle has further to run. Memory cycles run 2–3 years from trough to peak; this one bottomed Q1 23 (Hynix lost ₩9.1T in FY23) and is ~3 years in. HBM4 (Hynix 5th gen) qualification at Nvidia is the next leg, expected to ramp through 2H26–2027.
  5. Re-rating room remains, mathematically. At ~10-13x run-rate Q1 EPS, even modest earnings persistence gets you to a fair multiple of 12–15x and a flat tape — the upside case is run-rate earnings holding through 2026 + a multiple expansion to 15-18x as the cycle is recognized as durable.

3. Hynix vs Samsung — They Are Not the Same Trade

This is the key intellectual point: Samsung is not a 2x lagged Hynix.

Dimension SK Hynix Samsung
Business mix ~95% memory (DRAM + NAND) ~50% memory; ~30% mobile/CE; ~20% foundry/system LSI/display
HBM position Market leader, Nvidia primary Late entrant, qualifying HBM3E at Nvidia 2H26
Operating model Pure cycle-leveraged Diversified — memory upside diluted by foundry losses, mobile flat
Q1 26 OpMargin 71.5% 42.8%
FY25 ROE 61.2% 18.9%
Net cash ₩32.6T ₩119.3T
5y beta 2.03 1.33
3-month return +151% +85%
Investor archetype Cycle/HBM purist Quality compounder + cycle option

The bull case for Hynix is "the cycle is real and I want max torque" — clean memory P&L, no business-mix dilution, and the operating leverage to ride peak pricing from peak share.

The bull case for Samsung has a different shape: (a) it's a Hynix-style memory cycle play plus (b) HBM3E qualification at Nvidia 2H26 closes the gap and brings a non-linear pricing/share gain, plus (c) you are paid for foundry optionality (TSMC alternative for AI ASICs) plus (d) ₩119T of net cash and an under-utilized buyback authorization makes downside softer. Samsung is the lower-beta way to express the same cycle, with a back-end catalyst the market is still pricing skeptically.

Pair conclusion: if the cycle continues into 2H26 as the bull case requires, both work but Samsung outperforms because (a) HBM3E ramp at Nvidia would be a step-function for them while it's incremental for Hynix, (b) margin expansion is earlier-stage (42.8% vs 71.5% — more room to run), and (c) sell-side is still net positive on Samsung (10% upside to mean target) while net negative on Hynix (-12% to mean target). If the cycle rolls over, Hynix loses harder (full-cycle exposure, beta 2.0+), Samsung's diversification and cash absorb some of the drawdown.

4. Thesis Pillars Scorecard

SK Hynix — HIGH conviction long, but late-cycle

Pillar Original Expectation (Q1 25) Current Status (Q1 26) Trend
HBM share leadership 55%+ in HBM3E Held; HBM4 sampling on schedule ✅ Stable
OpMargin >40% "Possible at peak" 71.5% in Q1 26 ⚠️ Above peak — sustainability is the question
FCF turn positive After ₩4.5T burn FY23 ₩24.8T FCF FY25; Q1 26 alone ₩18.5T ✅ Inflected
Capex discipline Stay <₩30T ₩28.6T FY25 ✅ Holding
Customer concentration Nvidia >50% of HBM Validated; pricing intact ⚠️ Concentration risk
Stock vs target Re-rate to ₩2.0–2.4M Hit 2.36M (top of band) ⚠️ Mean target now -12% below spot

Samsung — HIGH conviction long, with two-handed catalyst path

Pillar Original Expectation Current Status Trend
Memory recovery Margin lift in 2H25 Q4 25 OpMargin 21%; Q1 26 43% ✅ Inflected hard
HBM3E at Nvidia Q4 25 → 1H26 qualification Reportedly closing — confirm at Q2 26 print 7/29 ⏳ Catalyst pending
Foundry losses narrowing <₩2T/qtr loss target Improving but not yet break-even ⚠️ Watch
Buyback execution ₩10T+ authorization ₩8.2T repurchased FY25 ✅ Active
Net cash optionality Maintain >₩100T ₩119T at Q1 26 ✅ Strengthening
Stock vs target Re-rate above ₩300K Hit 360K, mean target 397K ✅ Sell-side still has +10% room

5. Catalysts & Calendar

Date Event Stock Expected Impact
2026-04-22 Hynix Q2 26 earnings (Yahoo's listed next print — actually likely Q1 reported in late April) 000660.KS Already reported — Q1 EPS 56,670 KRW set the run-rate. Q2 print in late July will be the next reading.
2026-07-29 Samsung Q2 26 earnings 005930.KS Pivotal. Watch (a) memory OpMargin trajectory above Q1's 42.8%, (b) explicit HBM3E shipment commentary, (c) foundry segment narrowing of losses.
Q3 26 HBM4 mass-production ramp at Hynix 000660.KS Validation of 2027 share retention
2H 26 Samsung HBM3E qualification at Nvidia (if not already) 005930.KS Step-function — HBM ASP/share/mix
Ongoing DRAM contract pricing prints (DRAMeXchange / TrendForce) Both Monthly read on cycle health
Ongoing Hyperscaler capex commentary (MSFT/META/GOOGL/AMZN) Both Demand-side bellwether

6. Expressing The Trade in HK — 7709 & 7747 (CSOP 2x Daily Leveraged)

These are NOT shares — they are 2x daily-reset leveraged ETFs issued by CSOP and listed on HKEX. Use them only with their mechanics in mind.

7709.HK (Hynix 2x) 7747.HK (Samsung 2x)
Last price 146.2 HKD 238.5 HKD
Today's move +5.1% +5.9%
Underlying today -0.1% (Hynix) +3.3% (Samsung)
3-mo underlying ret +151.3% +84.8%
Naive 2x cumulative +302.7% +169.6%
3-mo ETF actual +368.9% +189.7%
Realized leverage 2.44x 2.24x

Observation: in this rally, both products have exceeded naive 2x cumulative because the underlying trended (low return-reversal). Daily-reset leverage compounds in trending markets and decays in chop — the 3-month tape was strongly trending up, so compounding worked in the holder's favor.

This is not a structural feature, it is regime-dependent. If Korea memory chops sideways for two weeks at this elevated vol (Hynix daily ATR ≈ 6%, Samsung ≈ 6%), the ETFs will visibly bleed vs naive 2x. Volatility decay scales with σ²/2 per day — at 6% daily, that's ~18 bps/day of expected drag relative to a frictionless 2x position, or ~4–5% per month in pure choppy regime.

Sizing rules for L&I exposure:

  1. Short holding period — days to a few weeks, not months. Re-up the position rather than buy-and-hold across the cycle.
  2. Treat as a directional bet on the next 1–2 catalysts (e.g. Samsung's 7/29 print), not as a way to "leverage your portfolio's exposure to Korea memory."
  3. Position size based on 2.5x effective beta to underlying, not on listed leverage. A 7709 position behaves like ~5x beta to KOSPI semi index given Hynix's own beta to KOSPI is ~2x.
  4. The HK secondary market is reasonably liquid (7709 traded HK$4.7B turnover today; 7747 HK$2.1B) but mind the spread on volatile days — these can widen to 50–100bps.
  5. The product currency is HKD; underlying P&L is KRW. KRW/HKD basis rarely moves more than 1–2% in a fortnight, so FX is a second-order risk for sub-month holds.
  6. There is no inverse / 2x short product on these specific names — to fade the move you'd need to short the L&I product itself (borrow availability + cost is the constraint), or short Korean ADRs / use KRW-denominated futures.

7. Risks That Would Invalidate The Thesis

Cycle-rollover risk (both, sequenced):

Hynix-specific:

Samsung-specific:

Cross-cutting:

8. Recommendation

View Conviction Position size guide
SK Hynix (000660.KS) Long, take some risk off into strength High Trim 20–30% from full position; redeploy on 10%+ pullback or Q2 26 confirming print
Samsung (005930.KS) Long, prefer over Hynix from here High Hold full / add into 7/29 setup if we get a -5% pullback
Pair Long Samsung / underweight Hynix High Pair sized for 1.0x dollar neutrality, expecting Samsung to close the relative gap
7709.HK Tactical only; not an investment vehicle n/a Use for pre-catalyst trades, ≤2 weeks; size at half what listed 2x leverage suggests
7747.HK Tactical for Samsung 7/29 print setup n/a Same sizing discipline; preferred L&I expression for the next 8 weeks

Stop-loss triggers:

Sources & Data Vintage

Marked [UNSOURCED] items: none in numerical claims; all figures above sourced from yfinance MCP unless explicitly framed as sell-side / cycle commentary.