SK hynix
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
The HBM pace-setter inside the three-player DRAM oligopoly: SK hynix took the lead in AI memory first and still ships over half of HBM by revenue, but HBM leadership has changed hands before, the product is still memory, and its contracts are newer and less disclosed than Micron's.
SK hynix's edge is being first and best at stacking DRAM for AI accelerators — a real, measurable lead that sits on top of a commodity-cycle business:
- HBM Scale and Yield Lead: SK hynix held about 56% of HBM revenue in Q1 2026 (IDC) and is expected to supply more than half of NVIDIA's total HBM this year. Its MR-MUF packaging process and years of high-volume HBM3/HBM3E yield learning are why it shipped first; Samsung only cleared NVIDIA HBM4 qualification in Q1 2026, and TrendForce expects the lead to narrow from 59% to roughly 50% of bits in 2026.
- Long-Term Agreements, Not Yet a Fortress: Management has completed long-term agreements with about ten customers, typically five-year terms with deposits and price mechanisms. That is the same direction as Micron's take-or-pay SCAs, but SK hynix has not disclosed a contracted backlog or revenue coverage figure, so the floor under a downcycle cannot yet be measured.
- Still a Memory Oligopoly: Three companies make essentially all of the world's DRAM, which keeps entry off the table. But customers dual- and triple-source, prices are set by supply and demand rather than by the vendor, and a 76% operating margin is what a shortage looks like, not what the business earns through a cycle.
Moat Verdict
SK hynix is a direct AI beneficiary: its HBM scale and yield lead is the one strong pillar, and talent, process IP and the new long-term agreements are intact. Durability depends on keeping roughly half of HBM as Samsung ramps HBM4 and on the agreements proving they can hold a floor through a memory downcycle; nothing here compounds independently of the hardware cycle.
76.0 resilient · 65.0 vulnerable · 80/20 = 73.8 · = 74
Open a moat to read its note.
SK hynix sells memory components to OEMs and hyperscalers; there is no user interface to learn.
Memory chips carry no vendor-owned business logic; process know-how is rated under proprietaryData and scaleEconomics.
SK hynix does not derive any advantage from controlling access to public data.
HBM stacking, TSV and advanced-packaging engineers are scarce, and SK hynix's Icheon and Cheongju teams have the longest high-volume HBM record. Samsung and Micron recruit from the same small pool, so this is a deep bench rather than an irreplaceable one.
DRAM, HBM and NAND are bought separately and customers source each from several vendors; there is no product-suite lock-in to rate.
MR-MUF packaging process IP and years of HBM yield-learning data are genuine and are why SK hynix qualified first, but Samsung and Micron have both now shipped HBM4, so the lead is in years of learning rather than an uncopyable dataset.
Korean and US subsidies support SK hynix's fabs but create no switching barrier for customers, and US export controls on its China fabs are a constraint rather than a moat.
Memory has no network effect; customers buy on specification, qualification and price.
About ten customers have signed long-term agreements, typically five-year terms with deposits and price mechanisms, and HBM is co-qualified with each accelerator generation. Unlike Micron, SK hynix has not disclosed a contracted backlog or revenue coverage, so this is real embedding that cannot yet be measured.
Memory is a component input; SK hynix is not the source of truth for any customer process.
SK hynix held about 56% of HBM revenue in Q1 2026 (IDC) and 62% of HBM shipments as of Q2 2025 (Counterpoint), and is expected to supply over half of NVIDIA's HBM in 2026. That volume lead gives it the most yield learning per wafer in the highest-margin memory product, which Samsung spent two years failing to close.
Sells a specified component to engineers; buyers pay the market price, not a brand premium.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
The HBM pace-setter inside the three-player DRAM oligopoly: SK hynix took the lead in AI memory first and still ships over half of HBM by revenue, but HBM leadership has changed hands before, the product is still memory, and its contracts are newer and less disclosed than Micron's.
Growth Score
Q2 2026 revenue was a record ₩79.3T (+257% YoY, +51% QoQ) with ₩60.5T of operating profit at a 76% margin, after ₩52.5T of Q1 revenue. DRAM bit shipments grew high-single digits sequentially while ASPs rose about 30%. HBM4 mass shipments began in Q2 and ramp through the second half, although the quarter missed estimates on slower HBM4 shipments and conservative pricing. 2026 capex is guided to the upper end of the ₩40T range against ₩30.2T in 2025, and management says it will add capacity in stages. SK hynix listed ADRs on Nasdaq (SKHY, ten ADSs per share) on July 10, 2026; the ADR has traded at a large premium to the Seoul line. It is a defendant, with Samsung and Micron, in the June 2026 US DRAM price-fixing class action.
Valuation Score
At ₩1,776,000 (September 30 close, ~₩1,260T market cap), SK hynix sits between the ₩1,000,000 bear and ₩2,000,000 base, about 78% of the way from bear to base, for a valuation score of 71. Q2 2026 operating profit of ₩60.5T annualises to roughly ₩266,000 of after-tax EPS, so the stock trades near 7× run-rate earnings — the same single-digit band MU sits at, which is what memory always trades at near a peak.
First Across the HBM Line
SK hynix's edge is being first and best at stacking DRAM for AI accelerators — a real, measurable lead that sits on top of a commodity-cycle business:
- HBM Scale and Yield Lead: SK hynix held about 56% of HBM revenue in Q1 2026 (IDC) and is expected to supply more than half of NVIDIA's total HBM this year. Its MR-MUF packaging process and years of high-volume HBM3/HBM3E yield learning are why it shipped first; Samsung only cleared NVIDIA HBM4 qualification in Q1 2026, and TrendForce expects the lead to narrow from 59% to roughly 50% of bits in 2026.
- Long-Term Agreements, Not Yet a Fortress: Management has completed long-term agreements with about ten customers, typically five-year terms with deposits and price mechanisms. That is the same direction as Micron's take-or-pay SCAs, but SK hynix has not disclosed a contracted backlog or revenue coverage figure, so the floor under a downcycle cannot yet be measured.
- Still a Memory Oligopoly: Three companies make essentially all of the world's DRAM, which keeps entry off the table. But customers dual- and triple-source, prices are set by supply and demand rather than by the vendor, and a 76% operating margin is what a shortage looks like, not what the business earns through a cycle.
Moat Verdict
SK hynix is a direct AI beneficiary: its HBM scale and yield lead is the one strong pillar, and talent, process IP and the new long-term agreements are intact. Durability depends on keeping roughly half of HBM as Samsung ramps HBM4 and on the agreements proving they can hold a floor through a memory downcycle; nothing here compounds independently of the hardware cycle.
76.0 resilient · 65.0 vulnerable · 80/20 = 73.8 · = 74
Open a moat to read its note.
SK hynix sells memory components to OEMs and hyperscalers; there is no user interface to learn.
Memory chips carry no vendor-owned business logic; process know-how is rated under proprietaryData and scaleEconomics.
SK hynix does not derive any advantage from controlling access to public data.
HBM stacking, TSV and advanced-packaging engineers are scarce, and SK hynix's Icheon and Cheongju teams have the longest high-volume HBM record. Samsung and Micron recruit from the same small pool, so this is a deep bench rather than an irreplaceable one.
DRAM, HBM and NAND are bought separately and customers source each from several vendors; there is no product-suite lock-in to rate.
MR-MUF packaging process IP and years of HBM yield-learning data are genuine and are why SK hynix qualified first, but Samsung and Micron have both now shipped HBM4, so the lead is in years of learning rather than an uncopyable dataset.
Korean and US subsidies support SK hynix's fabs but create no switching barrier for customers, and US export controls on its China fabs are a constraint rather than a moat.
Memory has no network effect; customers buy on specification, qualification and price.
About ten customers have signed long-term agreements, typically five-year terms with deposits and price mechanisms, and HBM is co-qualified with each accelerator generation. Unlike Micron, SK hynix has not disclosed a contracted backlog or revenue coverage, so this is real embedding that cannot yet be measured.
Memory is a component input; SK hynix is not the source of truth for any customer process.
SK hynix held about 56% of HBM revenue in Q1 2026 (IDC) and 62% of HBM shipments as of Q2 2025 (Counterpoint), and is expected to supply over half of NVIDIA's HBM in 2026. That volume lead gives it the most yield learning per wafer in the highest-margin memory product, which Samsung spent two years failing to close.
Sells a specified component to engineers; buyers pay the market price, not a brand premium.
Growth Analysis
Growth Drivers
Key Risk
If AI hyperscaler capex pauses in 2027–28 while SK hynix, Samsung and Micron all bring higher capex online, HBM and DRAM pricing can fall faster than the long-term agreements protect, as in the 2022–23 downcycle when SK hynix posted large losses. Samsung's HBM4 qualification is already expected to cut SK hynix's HBM bit share from 59% toward 50%. Falsifiable: quarterly operating margin below 40% for two consecutive quarters, or HBM revenue share below 45%, would break the leadership-through-the-cycle claim. Secondary: the June 2026 DRAM price-fixing class action remains early-stage.
Score Derivation
80.7 base + 2.7 trajectory − 10 risk = 73
Base 80.7 (12–20% CAGR, midpoint 16%, measured off the peak-cycle 2026 base, matching MU) + 2.7 trajectory (HBM and conventional DRAM accelerating; NAND/Solidigm stable) + 0 margin (stable: a 76% operating margin is the top of a pricing cycle, not a trend that can keep expanding) − 10 high risk (hyperscaler capex pause against an industry-wide capex step-up, graded the same as MU and Samsung on the same commodity exposure) = 73
HBM4 Ramp and the Capex Question
HBM4 mass shipments began in Q2 2026; ramp weighted to the second half
About ten long-term agreements signed, typically five-year terms with deposits
2026 capex guided to the upper end of the ₩40T range vs ₩30.2T in 2025
Fixed dividend of ₩1,500 per share through 2027; a further shareholder-return plan is promised later this year
Nasdaq ADR (SKHY) listed July 10, 2026 at ten ADSs per common share
Price Scenarios (12–24 Months)
Valuation Analysis
The base prices a mid-cycle EPS of roughly ₩150,000–160,000 (a little over half the Q2 run-rate, the same haircut used for MU) at ~12–13×, which gives about ₩2,000,000. The Street's ~₩3.1M mean capitalises the run-rate for longer; the base stays below it because the whole industry is raising capex into the peak. Reported Q2 net profit of ₩93.9T ran far above operating profit on non-operating items, so the run-rate here is built from operating profit, not the headline EPS. The SKHY ADR has traded well above the Seoul line in dollar terms; the scores use the KRX price. ~₩2,000,000.
Valuation Multiples
| Forward P/E (run-rate) | ~7× |
| P/E on mid-cycle EPS | ~11–12× |
| Price / Sales (run-rate) | ~4× |
| Dividend | ₩1,500 |
On run-rate earnings SK hynix looks very cheap, but that is the normal shape of memory at a peak. The bear prices a digestion-year EPS near ₩75,000 at ~13× trough earnings, giving ₩1,000,000. The bull needs the run-rate to hold through 2028 on HBM4/HBM4E and long-term agreements.
Approximate figures as of September 30, 2026.
Where We Are vs Targets
Loading live price…
Memory cycle reversion: hyperscaler capex pauses in 2027, Samsung takes back HBM share on HBM4, and the industry's higher capex turns a shortage into oversupply.
- Samsung's HBM4 ramp takes SK hynix's HBM revenue share below 45%
- Conventional DRAM contract prices fall 30%+ from the 2026 peak in a 2027–28 digestion year
- EPS falls toward ₩75,000 and the stock trades at ~13× trough earnings, giving about ₩1,000,000
HBM4 carries 2026–27, then a digestion year; long-term agreements hold mid-cycle earnings well above any prior memory cycle.
- HBM4 ramp keeps SK hynix at or near half of HBM revenue through 2027
- Long-term agreements with about ten customers cushion pricing in the down-year
- Stock trades at ~12–13× a ~₩155,000 mid-cycle EPS, giving about ₩2,000,000
AI inference creates a second HBM demand wave, so there is no digestion year and the run-rate earnings are capitalised.
- HBM4E and custom-base-die HBM keep SK hynix first to qualify at NVIDIA and the hyperscalers
- A disclosed long-term agreement backlog shows contracted revenue above a third of sales
- EPS holds near the ~₩265,000 run-rate through 2028 and the stock trades at ~11×, giving about ₩2,900,000