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.
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.
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.
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 | Read |
|---|---|---|
| AI infrastructure bottleneck | Core | HBM is the scarcest input in AI compute. This is the purest large-cap expression of the memory chokepoint. |
| Memory supercycle | Pure-play | Gartner 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 thesis | Direct conflict | This 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. |
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.
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.
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 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.
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.
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 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.
₩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.
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.
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.
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.
| Event | Detail | What it means |
|---|---|---|
| 10 Jul · today | When-issued trading, ticker SKHYV | First US price discovery. Thin, unreliable prints — watch, do not touch. |
| 13 Jul | Regular trading begins as SKHY · settles 14 Jul | The real book opens. The spread vs Seoul (÷10, in USD) is the honesty meter for early prices. |
| Index inclusion | Expected SOX entry → passive flows | Semiconductor index funds become forced buyers — a mechanical bid under the ADS in the early weeks. |
| 29 Jul | Q2 2026 results | The first real test. One question: are prices still rising quarter-on-quarter? Samsung already hinted the pace is slowing. |
| H2 2026 | HBM4E samples ship · mass production 2027 | The next leadership test. Samsung has shipped HBM4 since February — the crown gets defended chip by chip, socket by socket at NVIDIA. |
| Standing | 2026 capacity sold out · SK chair sees wafer shortage possibly to 2030 | Near-term revenue is effectively contracted. The fight is entirely about 2027 pricing — negotiations land in H2 news flow. |
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.
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 happen | Roughly 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 |
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.
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.
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.
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.
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).