SK Hynix

Phase A — Understand the business

Lens 1 · Company Overview (re-run — DRAM-crown + sold-out status)

World's #1 memory maker by DRAM revenue (re-took the crown from Samsung in 2026, first time in 26 years ) and #1 in HBM — the stacked-DRAM that gates every AI accelerator. DRAM ~70–75% of revenue (incl. HBM, server DRAM, DDR5, LPDDR), NAND/eSSD ~20–25%, HBM the margin engine within DRAM. Crown jewel of SK Group (SK Square holds ~20.07%; Chairman Chey Tae-won). Defining fact (held): sold out its entire 2026 output — DRAM, NAND, HBM — to customers including NVIDIA, tight into 2027, contracted a year-plus ahead with prepayments. New this pass: a strategic pivot to prioritize high-margin conventional DDR5 over maximum HBM4 ramp speed — a pricing-power flex, not a demand problem.

Lens 2 · Supply Chain (carried — unchanged; see deep-dive-2026-06-09.md)

Double chokepoint: gated upstream by ASML EUV + TSMC (HBM4 logic base die moved onto TSMC 12nm under the "One Team" alliance), itself the binding constraint downstream on NVIDIA. Effective HBM ceiling = intersection of SK Hynix stacking yield × TSMC CoWoS packaging. Chokepoint stack: HBM sev 9, CoWoS sev 9, EUV sev 7. No material change in 4 weeks.

Lens 3 · Competitive Advantages (re-run — HBM4 now multi-qualified)

Lead is real but its character changed this month: it is now a share-and-yield lead, not an exclusivity lead. HBM share ~57–62% (1Q26 SK Hynix cited 57% on its own call; TrendForce/Counterpoint ~60–62%), Micron #2 (~21%, overtook Samsung), Samsung #3 (~17%). Nvidia Vera Rubin HBM4 allocation ~SK Hynix 60–70% / Samsung 25–30% / Micron the remainder — but with all three now Nvidia-certified (Jun 2), the moat is yield-at-volume + a multi-generation Nvidia relationship, not being the only qualified supplier. Bargaining power: sold-out + prepaid + ~20% HBM3E price hikes. Vulnerability (sharper this pass): Micron is qualified and its HBM ASP is rising fastest (~22% in 2026); Samsung's 1c yield (~50%) is the only thing keeping it a distant #3. Allocation can shift at the margin — but the near-term protection is that Hynix is voluntarily leaving HBM4 volume on the table for DDR5 margin.

Lens 4 · Segments (carried — unchanged; not separately disclosed)

DRAM (incl. HBM) ~70–75% ↑; NAND/eSSD ~20–25% ↑ (fully booked 2026); HBM within DRAM not separately disclosed (n/a — the #1 value driver is modeled, not reported). HBM3E 12-Hi 36GB ≈ ~2/3 of 2026 HBM shipments; HBM4 16-Hi 48GB ramps H2→Q4 2026 (now throttled for DDR5 margin). HQ Korea (Icheon/Cheongju); China NAND fabs (Wuxi/Dalian); Yongin (KR) + planned Indiana (US) packaging. See deep-dive-2026-06-09.md for the full breakout — no reported change.

Phase B — Measure performance

Lens 5 · Earnings Result (re-run — 1Q26 still the latest actual; Q2 not yet printed)

1Q26 (reported Apr 23 2026) remains the last actual — Q2 2026 has NOT printed (SK Hynix reports Q2 ~late July; Samsung's Jul-7 preliminary is the sector read-through). 1Q26 was a record: Revenue ₩52.58T (~$35.5B) — first quarter ever above ₩50T, +60% QoQ, +198% YoY; operating profit ₩37.61T, op margin ~72% (record); net ₩40.35T. HBM 57% share cited on the call. Capex guided to rise "significantly" above 2025's ~₩30.2T (analysts model ~₩50T, +20–30%) — M15X, Yongin, ASML EUV. New sector read-through (Jul 7): Samsung's preliminary Q2 op profit surpassed ₩100T (ex-bonus) — the supercycle is intact into the quarter Hynix is about to report. 2026 full-year Street projection for Hynix: ~₩355T ($231B) revenue, ~₩221T ($144B) net income — +265%/+415% YoY. Caveat: these are analyst estimates, not filing-read — see provenance note.

Lens 6 · Earnings Calls (sentiment trend) (re-run — 1Q26 call + Computex)

Tone remains "structurally short into 2027/2030." Threads (from 1Q26 call + Chey Tae-won at Computex): (1) "sold out" — DRAM+NAND+HBM booked through 2026, demand exceeding capacity for ~3 years; (2) HBM4 + the TSMC One-Team pivot, HBM4E mass production 2027, move into client-specific custom HBM; (3) heavy capex with a balance-sheet guardrail (goal: net cash >₩100T while raising returns) — "fund the ramp from the cash flush." CFO: memory has entered "a new paradigm." New wrinkle: the deliberate DDR5-over-HBM4 prioritization is a fresh strategic signal — management is optimizing margin capture, not chasing HBM share at any cost.

Lens 7 · Comps & Valuation — THE PAIR-CALL CORE (re-run, primary-reconciled)

Same-source (stockanalysis.com, so the methodology is consistent across the pair — this is the clean comparison the whole call turns on), as of Jul 6–7, 2026:

The reconciliation that flips the pair lean. The freshest Micron dossier (2026-06-29) built part of its case on "MU trades at an EV/EBITDA premium to SK Hynix the leader — 18.4x vs 10.7x, a yellow flag." That 10.7x Hynix figure does not hold on current same-source data. Today: Hynix EV/EBITDA 17.85x vs MU 15.95x — the two are near-parity, and if anything MU is marginally cheaper on EBITDA, while Hynix is cheaper on forward P/E (6.06x vs 6.86x) and on the two metrics that reward the leader — it carries ~3× the HBM share and just re-took the overall DRAM crown at the same ~$1.1T cap. The "priced-in incumbent vs catching-up value" framing behind the original "long MU / underweight Hynix" lean inverts: you are being asked to pay the same ~$1.1T for either (a) the 21%-share #3 whose own primary-grounded dossier says the bull is in the price at NEUTRAL/WATCHING, or (b) the 57–62%-share #1 leader, at a lower forward multiple, with a dated ~$29B re-rating catalyst 3 days out. On through-cycle earnings power the leader with 3× the share at a lower forward multiple is the better risk/reward long leg. The valuation gap has closed to near-zero and slightly favors Hynix.

Other peers (context, ): Samsung 005930.KS ~$1.3–1.5T cap, record Q2, but a conglomerate (foundry + memory + devices) — not a clean memory comp; Nanya 2408.TW small-cap DRAM laggard ~6x fwd.

Lens 8 · Stock-Price Catalysts (re-run — new moves)

High-beta memory-cycle + AI-narrative + Korea-macro instrument. ATH ₩2,407,000 on Jun 2, 2026 (cap $100B→$1T in ~16 months). New moves this pass: (a) Jun 6: −9.92% on a soft Broadcom outlook; (b) Jul 2: −14.5% on the Meta Compute demand scare (Samsung −9.1%, Kospi −7.9%) — a sector rout from Wall Street, not memory-specific; (c) ~20% off the June peak into Jul 7 (Samsung's record Q2 print triggered profit-taking, not a rally — "sell the news"). The pattern (sharpened): the equity is a leveraged bet on one demand narrative (it fell ~10% on a Broadcom wobble, ~15% on a Meta headline — neither its own results), which is exactly why it trades at a cyclical multiple despite oligopoly economics. Imminent binary: the Jul 10 ADR debut is the next catalyst — a successful mega-listing into AI-hungry US money is the re-rating case; a soft debut into a fearful tape is the "sell-the-news" risk. Fortune framed the listing itself as a market referendum: "could signal whether the market can still boom — or is headed for a bust."

Phase C — Judge people & books

Lens 9 · Management & Governance (re-run only on the listing size)

CEO Kwak Noh-jung; CFO Kim Woohyun (romanization varies — hygiene note). Controlled via SK Square (~20.07%) under the SK Group pyramid; Chairman Chey Tae-won drives strategy (Computex 2026: pledged to double capacity in 5 years, shortage through 2030). Foreign ownership 56%. The governance event is now here and bigger: a ~$29B Nasdaq ADR listing (~Jul 10) — up from the "$14B / 2–3% float" the June dossier had — with the explicit rationale of forcing a re-rating by putting SK Hynix in direct comparison with Micron. Chaebol tells / red flags (held, and the bigger raise amplifies the dilution concern): investor groups oppose dilutive new-share issuance (want buybacks); a ₩100T shareholder/employee reward "burden"; SK pyramid cross-shareholding governance discount. A ~$29B raise is ~1.7% of a ~$1.1T cap if fully primary — the dilution is modest in %, but the size + the fearful tape make the debut's pricing the swing variable. Structure (primary/secondary split, exact new-share count) not confirmed from a filing this pass — flag.

Lens 10 · Forensic Red Flags (re-run — peak-cycle + concentration)

(1) Peak-cycle accounting — 72% op margin / 61% ROE are cycle-peak; memory mean-reverts violently (operating losses in 2023); forward P/E ~6× is the market pricing non-durability. Watch inventory build + contract-price roll-over (1Q26 balance-sheet detail n/a — the scaffold is empty; not pulled from a primary filing this pass). (2) NVIDIA concentration — HBM ~2/3 to one buyer's roadmap; the Jul-2 −14.5% on a Meta headline shows the coupling. (3) HBM not separately disclosed — the #1 value driver is modeled, not reported. (4) Capex vs durability — ~₩50T + ~$8B EUV + Indiana/Yongin underwritten by "sold out 3 years"; net-cash guardrail mitigates; a cycle turn mid-build is the canonical memory trap. (5) Chaebol/related-party — SK pyramid, ₩100T reward, now a ~$29B dilutive listing opposed by minorities. Regulatory findings: NOT independently verified this pass — SK Hynix has null CIK (no SEC EDGAR EFTS coverage), fetch-regulatory-findings.ts yields no SEC LR/AAER for a non-US filer, and the regulatory/ dir is empty. No material Korean (FTC/FSC) or US enforcement surfaced in web scanning, but this is not an audited clean-bill — label: "no findings surfaced via web; SEC EFTS N/A for a foreign private issuer; not independently verified."

Phase D — Project & stress-test

Lens 11 · Forward Projection (EPS, next 3 years) (re-run — Street FY26 now available)

Bottom-up; calendar-year; KRW. Street now models FY26 ~₩355T rev / ~₩221T net income — I anchor the base near Street and widen the tails for the cycle. Shares ~708.3M.

Read: FY26 is effectively locked at a record (sold out + prepaid). The bet is 2027–28. At ~6× forward the market is pricing a roll-over the sold-out book + ADR re-rating are set to resist — the mirror image of Micron, where ~7× is pricing the bull. Relative: Hynix's floor is arguably firmer (larger share, sold-out prepaid, DDR5-margin optionality) at a lower multiple. Log: npx tsx scripts/research/forecast.ts create --topic hardware --question "SK Hynix 2026 operating margin >= 65%" --p 0.62 --resolves 2027-02-28 --tags sk-hynix,deep-dive (not executed — dispatched/unattended pass; log on human review.)

Lens 12 · Bull vs Bear (re-run)

Bull. SK Hynix captures the single largest slice of the AI scarcity rent — ~57–62% of HBM, ~2/3 of Nvidia's HBM4, sold out and prepaid into 2027, at 72% operating margin and ~61% ROE — has just re-taken the overall DRAM crown from Samsung after 26 years, and still trades at ~6× forward. Management is so convinced of the mispricing it is executing the largest ADR listing in history (~$29B, Jul 10) specifically to force the re-rating against Micron. Rare alignment: best fundamentals in the cluster and the most explicit, dated catalyst — now days away. Bear. (1) The lead is yield-and-trust, not a monopoly — Samsung out-qualified it on HBM4 timing in Feb, and all three are now Nvidia-certified; if Samsung's ~50% 1c yield converges, Nvidia wants a credible second source and allocation shifts. (2) ~2/3 of HBM rides one customer's roadmap; a Meta headline knocked 15% off in a day. (3) It's a memory cyclical at peak margins with a chaebol governance discount the listing may not erase — and a ~$29B raise into a fearful tape risks a "sell-the-news" debut. Pre-mortem (2027): Samsung's 1c yields hit parity, Nvidia rebalances Rubin/Feynman toward two-plus suppliers, the DDR5-throttle backfires as commodity DRAM floods, a capex pause coincides with HBM4 capacity from all three landing — margins fall 72%→40%, and the ADR prices at the old cyclical multiple anyway (the "boom-or-bust referendum" resolves bust). Contrarian view: the market is fighting the last memory war (assuming mean-reversion) while the sold-out+prepaid contract structure and the One-Team TSMC coupling have made Hynix more contracted-infrastructure-supplier than spot cyclical — and the ADR is the catalyst that forces the multiple to admit it. The variant-perception edge is that the Jul-2 selloff and the DDR5-throttle are being read as weakness when they're beta and pricing-power respectively.

Lens 13 · Devil's Advocate (short-seller) (re-run)

The 2/3 HBM4 allocation is a procurement decision Nvidia re-makes each generation, and as of Jun 2 all three are certified — the exclusivity is gone; only yield-share remains. A −15% day on a Meta headline proves the equity is a leveraged bet on one customer's capex dressed as an oligopoly. The chaebol steers capital to group priorities; the **$29B dilutive listing** (opposed by minorities) and the ₩100T reward "burden" both transfer value away from the shares — and a raise that large into the worst tape in a year risks pricing the whole complex lower. 72% operating margin is the most mean-reverting number in hardware; the deliberate DDR5-over-HBM4 pivot is itself an admission that HBM4 volume economics aren't yet worth chasing. If AI capex merely decelerates (Meta Compute is the first crack) while HBM4 + DDR5 capacity floods in 2027, op margin halves and the "structural" story reverts to "Korean cyclical at the top of the cycle" — exactly what the ~6× forward multiple always said. And the relative long-Hynix case cuts the other way if the ADR sells the news: a fresh ~$29B of tradeable supply can pressure the very re-rating the thesis relies on.

Lens 14 · Management Questions (carried + 3 added)

Carried from deep-dive-2026-06-09.md (HBM-as-%-of-revenue disclosure, Samsung-parity allocation protection, listing dilution, Nvidia-%-of-HBM, One-Team margin capture, capex-vs-oversupply ASP, net-cash-vs-ramp, custom-HBM lock-in, China NAND export exposure, trough-margin defense). Added this pass: 11. Why throttle HBM4 for DDR5 now — is it margin optimization or an HBM4-yield/economics tell? At what DDR5-vs-HBM ASP spread does the priority flip back? 12. With all three Nvidia-certified for HBM4, what contractually protects your ~2/3 Rubin allocation through Feynman? 13. The ~$29B ADR — primary vs secondary split, exact new-share dilution, and use of proceeds vs the ₩100T shareholder-reward commitment?

---