Analysis date · 10 July 2026
SKHY
SK hynix Inc. · NASDAQ ADS (1 ADS = 1/10 share)
Seoul primary: 000660.KS · Icheon, est. 1983
$149.00ADS offer price · set 9 Jul
~$1.09TMarket cap · post-deal
AI Memory · HBM Memory Supercycle US Listing · Largest Ever
Price = ADS offer + Seoul close, web-sourced this run (prod → Alpaca) · when-issued SKHYV trades 10 Jul, regular SKHY from 13 Jul · research only, not financial advice
The Verdict

The world's best memory business is listing in America during the best quarter memory has ever had — and it wants you to pay a premium to a home market that just fell 27%.

The business is not the question. SK hynix owns 56% of HBM — the memory NVIDIA cannot build AI chips without — and just printed a 72% operating margin, a number no memory maker has ever seen. The question is the cycle. Prices rose 60–75% in a single quarter; every producer is now racing to add supply for 2027–28; the sector has already started selling off on record results; and the company itself is raising $26.5B at the top to fund the arms race. At $149 the deal is priced almost exactly at our probability-weighted fair value — meaning you are paid nothing for taking cycle risk.

How strong is the caseStrong — best asset in memory
How sure is the outcomeLow-medium — the cycle decides
Confidence in this readMedium-high · 7/10
Type of betCyclical near a possible crest — buy the pullback, not the debut
Bottom line — Hold. Do not buy the debut; stalk the pullback.

Skip day one entirely — thin books, index flows and IPO adrenaline make the first prints unreliable, and you would be paying above what Seoul was paying on Thursday. The Seoul line is the price anchor; arbitrage will keep the ADS tied to it. The plan: first tranche $128–138, deeper add $119–126, invalidation below ~$112. Q2 results on 29 July are the first real test — watch whether prices are still rising quarter-on-quarter.

Verdict Scorecard

Eleven calls, one glanceEach line is a call, in plain words.

Valuation
Fairly valued — priced for the supercycle to persist
21x trailing earnings vs a 9x five-year norm; 5.5x forward. The forward multiple is the market saying "these are peak earnings."
Growth
Explosive — but price-driven, not volume-driven
Revenue +198% in a year. Consensus sees 2026 sales more than tripling to ~$235B. Almost all of it is chip prices, not more chips.
Profitability
Record-shattering
72% operating margin, 77% net margin in Q1 — all-time highs for the entire memory industry. Three years ago this company lost money.
Financial health
Fortress
₩35T (~$23B) net cash before the deal, debt-to-equity 12%, plus $26.5B fresh proceeds landing now.
Share dilution
Moderate — a one-off 2.5% for the war chest
17.79M new shares (~2.5% of the count), plus a possible greenshoe. Historically a clean share count; this is growth funding, not survival funding.
Edge / moat
The HBM crown — for now
56.4% of HBM, first to HBM4, 2026 capacity sold out. But Samsung has shipped HBM4 since February and Micron and China's CXMT are coming.
Management
Made the memory bet of the decade
Bet on HBM early when rivals cut back — that call built the moat. Chaebol structure (SK Square holds 20%) is the standing governance caveat.
Price momentum
Parabolic — and now cracking
+~700% in a year, +223% this year — then −27% off the late-June top in two weeks. The short-term regime has flipped.
Catalysts
Dense — the next three weeks decide the tone
Regular trading 13 Jul, expected SOX index inclusion flows, Q2 results 29 Jul, HBM4E samples in H2.
Risk level
High
Memory cycles end. A 2027–28 supply wave is being built right now — partly with this IPO's money. Plus China exposure and efficiency-algorithm shocks.
Overall
World-class business, late-cycle price
Own it — but on your terms, not the syndicate's. Hold at $149; accumulate $128–138.
Why "fairly valued" and not "cheap" at 5.5x forward earnings

A 5.5x forward P/E looks like a gift — until you remember memory investing's oldest rule: these stocks look cheapest at the top and most expensive at the bottom. The market refuses to capitalise peak-cycle earnings because it knows they mean-revert. In 2023 this company posted an operating loss; the honest question is not "what does it earn in 2026" but "what does a normal year look like after the 2027–28 supply lands" — and at $149 you are paying full price for a generous answer to that question.

Snapshot

The numbers, read aloud

ADS offer price
$149.00
Set 9 Jul at a 3.1% premium to Seoul — you start by paying more than the home market.
Seoul close · 9 Jul
₩2,186,000
≈ $144.50 per ADS-equivalent. This line is the anchor — arbitrage keeps the ADS tied to it.
52-wk range (Seoul)
₩245K–2,987K
A 12x range in a year. Already −27% off the late-June top. Entry price is everything here.
Market cap
~$1.06T
Crossed $1T in May. Korea's #2 company; out-earned Samsung in 2025 for the first time ever.
Deal size
$26.5B
177.9M ADS, 7x oversubscribed — the largest foreign US listing in history, beating Alibaba 2014.
P/E · trailing
~21x
More than double the 9x five-year norm. The cycle has been re-rated, hard.
P/E · 2026E
~5.5x
The tell. Markets never pay up for peak memory earnings — this multiple is the top warning.
Dividend yield
~0.1%
Not why anyone owns it. Cash goes to fabs, not shareholders — for now.
Next earnings
29 Jul 2026
Q2 — first print as a US-listed name, sixteen days after the debut. The first real test.
What it is

The company that feeds NVIDIA's chips

SK hynix makes memory — the chips that store and move data. It is #1 in the world in HBM (56.4% share), the stacked high-bandwidth memory bolted directly onto AI processors, #1 in DRAM (29.1%) and #2 in NAND flash storage (18.5%, boosted by buying Intel's NAND unit, now Solidigm). Customers include NVIDIA, Google, Microsoft and Apple. Every serious AI accelerator needs HBM, and more than half of it comes from this company.

Founded 1983 as Hyundai Electronics, nearly killed by the 2000s memory glut, rescued and renamed when SK Group took control in 2012 — then transformed by one decision: betting early and heavily on HBM while rivals cut back. When AI demand exploded, SK hynix was the only supplier ready at scale. It has traded in Seoul since 1996; this is a dual listing, not a company going public.

The ADS mechanics matter: one Nasdaq ADS (ticker SKHY, from Monday 13 July; when-issued as SKHYV today) equals one-tenth of a Seoul share, with Citibank as depositary. The Seoul line — 712.7M shares, SK Square holding 20% — sets the price. If the ADS runs ahead of Seoul, institutions arbitrage the gap closed. UBS flags that a modest structural ADS premium is possible (US holding is cheaper and easier), but the ADS cannot durably decouple from the home market.

Theme Fit

Where it sits in the Alpha Intelligence map

ThemeFitRead
AI infrastructure bottleneckCoreHBM is the scarcest input in AI compute. This is the purest large-cap expression of the memory chokepoint.
Memory supercyclePure-playGartner sees the memory market tripling from $216B (2025) to $633B (2026). SKHY is the biggest single beneficiary — and the biggest casualty when it turns.
AI/semis selloff · rotation thesisDirect conflictThis is the highest-beta version of the trade the house view says to fade. The −27% drawdown, Samsung selling off on record guidance and KOSPI entering a bear market all argue the rotation view is currently winning. That is exactly why the entry zone sits below the deal price, not above it.
Financial Forensics · Q1 2026

Quarter ended 31 Mar 2026Reported 22 Apr 2026 (preliminary, K-IFRS) · every number, in plain words, with what to do about it

Revenue₩52.58T (~$35.5B)+198% YoY · +60% QoQ
What it means

First ₩50T quarter in company history — in what is normally the weakest quarter of the year. Small footnote with a big message: it still missed the ₩53.55T consensus. Even blowouts now disappoint someone. Expectations are the real competition.

Operating profit₩37.61T · 72% margin~5x YoY
What it means

The margin was 42% a year ago, 58% last quarter, 72% now — an all-time record for any memory maker. One quarter of operating profit exceeded the entire FY2024. This is what a genuine shortage looks like. It is also what a peak looks like.

The engine: chip prices, not chip volumesDRAM ASP +~65% · NAND +~75%QoQ · shipments ≈ flat
What it means

DRAM shipments were roughly flat versus Q4; NAND shipments actually fell ~10%. Nearly all the growth is price. That cuts both ways with total symmetry: when prices stop rising, growth stops instantly — no volume cushion. Watch the ASP direction in Q2 above everything else.

Net profit₩40.35T · 77% marginnet > operating
What it means

Net profit exceeding operating profit is unusual — the gap is non-operating income (interest on the giant cash pile, FX and other gains). Real money, but not repeatable core earnings. Anchor on operating profit, not the net line.

The full-year staircaseFY23 → FY24 → FY25₩32.8T → ₩66.2T → ₩97.1T rev
What it means

Revenue tripled in two years; FY25 operating profit of ₩47.2T (+101%) out-earned Samsung for the first time in history. But hold the FY23 line in view: an operating loss of ₩7.7T, thirty months ago. Same company, same industry. That is the amplitude of this business — in both directions.

2026 consensus~$235B revenue (LSEG)>3x FY25
What it means

Analysts expect 2026 sales to more than triple again, with Gartner forecasting DRAM prices up 136–198% year-on-year through H1. If it lands, this is one of the fastest earnings explosions in corporate history. The risk is not that 2026 disappoints — it is that 2026 is the top of the mountain.

Cash generation+₩19.4T cash QoQafter capex
What it means

Cash and equivalents jumped to ₩54.3T in a single quarter — after heavy investment. At this run-rate the company generates a meaningful slice of its own market cap in cash across 2026–27. This harvest is the bull case's hard core: even a cycle turn leaves the balance sheet transformed.

Balance Sheet & Dilution

A fortress — funding an arms raceThe money is real. So is what everyone plans to do with it.

Net cash₩35T (~$23B)debt/equity 12%
What it means

₩54.3T cash against ₩19.3T debt, and debt fell ₩2.9T in the quarter. This company enters whatever comes next with the strongest balance sheet in its history. Downturns kill leveraged memory makers — this one is bulletproofed.

The raise$26.5B primary~2.5% dilution
What it means

17.79M new shares on a 712.7M base — modest dilution for a war chest this size, with a greenshoe that could add ~15% more. All primary: no insiders cashing out, which is the good version of an IPO. But note the reflex: memory companies raise the most money at cycle tops, never at bottoms.

Where the money goes₩45.5T fabs + ₩11.9T EUV
What it means

Yongin Fab 1 and the Cheongju P&T7 advanced-packaging plant through 2030, plus EUV lithography scanners delivered by end-2027, plus a $4B Indiana packaging site. Necessary to defend the crown — and it is precisely this industry-wide capacity, arriving 2027–28, that seeds the next downturn. The IPO funds both the moat and the glut.

Who anchored the bookBaillie Gifford · Coatue · Situational Awareness
What it means

Cornerstone investors — Situational Awareness alone reportedly committed up to $7B — with the book 7x covered. Serious long-horizon AI money endorsing the deal is a genuine signal. It also means the easy demand is already in the price.

Catalysts & Timeline

Three weeks that set the toneDense calendar — each item has a direction attached.

EventDetailWhat it means
10 Jul · todayWhen-issued trading, ticker SKHYVFirst US price discovery. Thin, unreliable prints — watch, do not touch.
13 JulRegular trading begins as SKHY · settles 14 JulThe real book opens. The spread vs Seoul (÷10, in USD) is the honesty meter for early prices.
Index inclusionExpected SOX entry → passive flowsSemiconductor index funds become forced buyers — a mechanical bid under the ADS in the early weeks.
29 JulQ2 2026 resultsThe first real test. One question: are prices still rising quarter-on-quarter? Samsung already hinted the pace is slowing.
H2 2026HBM4E samples ship · mass production 2027The next leadership test. Samsung has shipped HBM4 since February — the crown gets defended chip by chip, socket by socket at NVIDIA.
Standing2026 capacity sold out · SK chair sees wafer shortage possibly to 2030Near-term revenue is effectively contracted. The fight is entirely about 2027 pricing — negotiations land in H2 news flow.
What to watch next

Two numbers on 29 July: sequential ASP direction (still rising, or flattening?) and any language about 2027 HBM contract pricing. Those two lines are worth more than everything else in the release combined.

Both Sides

The bet, argued

▲ Bull case — "HBM is a franchise now, not a commodity"
Contracted, custom, scarce. 56.4% HBM share, 2026 sold out, HBM4E built with customer-specific dies. This looks more like a design win business than spot-market DRAM.
Demand is not done. Gartner: memory market $216B → $633B → $748B by 2027. Agentic AI shifts load to inference — which eats even more memory per query.
The cash harvest. On consensus, 2026–27 could generate a meaningful fraction of the market cap in free cash. Even a hard landing leaves a transformed balance sheet.
The multiple gives cover. At ~5.5x forward, earnings could halve in 2027 and it would still trade at ~11x — hardly demanding.
Flows and re-rating. 7x oversubscribed, SOX inclusion, and HSBC notes Micron carried an average 35% premium over 13 years purely for being US-listed. That gap is now SKHY's to close.
The Street agrees. 37 analysts, 35 buys, average target ≈ ₩3.17M ≈ $210/ADS (+41%); UBS at ₩3.2M; high estimate ₩4.7M ≈ $311.
▼ Bear case — "peak of the sharpest cycle ever"
It is still memory. Prices +60–75% in one quarter is the definition of unsustainable. FY2023 was an operating loss. Amplitude works both ways.
The supply wave is funded — partly by this deal. SK, Samsung, Micron and CXMT are all adding 2027–28 capacity. Morningstar's analyst is explicit: pricing power "only exists because of deep undersupply" — and expects a cycle turn.
The tape already turned. −27% off the June top, Samsung −7% on record guidance, KOSPI in a bear market, Micron sliding. Good news is being sold — the classic late-cycle signature.
Efficiency shocks are real. Google's TurboQuant (6x memory compression) knocked 5–7% off the sector in a day in March and is flagged in SK's own F-1. Jevons paradox is an argument, not a law.
End-demand is five capex budgets. Meta reportedly reselling compute and Apollo warning on AI monetization — if hyperscaler capex digests, memory feels it first and hardest.
China cuts both ways. CXMT climbing the stack (Apple is testing its chips) while SK's own Wuxi DRAM fab sits exposed to export-control whiplash.
Valuation

What you're paying for

⚖️ Valuation verdict — fairly valued at $149; the deal is priced at scenario EV

Base fair value: ~$140/ADS. Method: normalised mid-cycle earning power (structurally higher than the past thanks to the HBM mix — roughly $50–60B/yr) at 14x, plus today's net cash and IPO proceeds, plus the retained 2026–27 supercycle cash harvest, gives ~$129–148 per ADS. Probability-weighting the three scenarios below lands at ~$151 — within a dollar of the $149 offer. Translation: the syndicate priced this almost perfectly. You are paid nothing for cycle risk at the deal price. What to do: let the price come to you — the zone where the odds tilt in your favour is $128–138.

If…What has to happenRoughly worth
Goes well · 25%2027 HBM contracts hold pricing; HBM4E lands first at NVIDIA; US listing closes the Micron premium gap$230 — the Street's target zone (avg $210, high $311)
Middle · 50%2026 delivers, the cycle crests in 2027 but soft-lands; normalised earnings settle far above old cycles$140 — mid-cycle value plus the cash banked on the way
Goes badly · 25%2027–28 supply lands into digesting demand; ASPs roll; classic memory de-rate off the ₩2.99M peak$95 — a standard −50%+ cycle drawdown, cushioned only by cash
The multiple paradox is the whole story

Trailing P/E ~21x (vs 9x norm) says the market has already re-rated the cycle upward. Forward P/E ~5.5x says the market refuses to believe 2026 earnings persist. Both are rational at once only near a top. The stock is a buy when one of them breaks: either 2027 pricing holds (own it for the re-rate) or the price falls to where peak-and-fade is fully paid for (own it for the value). $149 is neither.

The Trade

Entry, hold, exit — one glanceConcrete levels, anchored to the Seoul line. A plan, not advice.

Entry zone
$128–138
The retest of the early-July low (Seoul ₩1.93–2.08M). Deeper add $119–126 at the ₩1.8–1.9M shelf. Nothing on debut day; don't chase above $160.
How to enter
In thirds
First tranche in the zone, second on the deeper shelf or after Q2 confirms ASPs, never all at once.
Hold for
12–24 mo
Through the 2026 harvest into the 2027 pricing verdict. Cycle-aware position, not a forever hold.
Sell · good case
$200–230
Trim hard into the Street's target zone. Let a rump ride only if 2027 HBM pricing is confirmed holding.
Sell · bad case
Below ~$112
Seoul below ₩1.7M after entry, or any invalidation trigger fires (below). Sell regardless of price.
Your Action Plan

What to watch, when to buy, when to sellThe bet, in one line: HBM scarcity holds 2027 pricing up long enough for the cash harvest to pay you before the cycle turns.

◐ Watch — every quarter

Four things: sequential ASP direction (DRAM and NAND), 2027 HBM contract news, industry capex announcements (the glut-o-meter), and the ADS↔Seoul spread. Q2 lands 29 July.

✓ If this happens — the bet is working, add

Q2 shows ASPs still rising sequentially · 2027 HBM volumes and prices reported pre-sold · HBM4E wins the lead slot at NVIDIA's Rubin · Seoul line builds a base above ₩2.0M. Add the second tranche.

✕ If this happens — the bet is broken, sell

First quarter of sequential DRAM ASP decline · 2027 HBM contract prices reported cut · Samsung or Micron displaces SK at a flagship NVIDIA socket · hyperscaler capex guidance rolls over hard. Sell, whatever the price.

Confidence & Data Gaps

What's solid, what's stubbed

Solid: Q1 2026 and FY2025 financials (company release, K-IFRS, 22 Apr — noting Q1 figures are preliminary/pre-audit), F-1 market-share and capex data, deal terms and pricing (Reuters / Bloomberg / Renaissance Capital, 9–10 Jul).

Secondary-sourced (verify before acting): analyst consensus (₩3.17M avg across 37 analysts), forward P/E (~5.5x, FactSet via press), FX conversion (~₩1,510/USD implied by deal pricing), SOX-inclusion expectation.

Timing caveats: written on debut day — Friday's Seoul close and the first SKHYV when-issued prints were not yet confirmed at write time; all levels are anchored to Thursday's Seoul close and the $149 offer.

Not wired — honest stubs (never-fake): insider transactions, forum/retail sentiment, congressional trades, short interest on the new ADS line, and the verified Alpaca close (no ADS price history exists yet; this run is flagged web_manual_on_demand).