App Economy Insights — SK Hynix Bets AI Broke the Cycle

The US listing — largest ADR in history (SKHY)

• SK Hynix is trying to pull off the largest ADR listing ever: over $28B (45T won), edging past Alibaba's ~$22B 2014 New York debut, with trading expected ~July 10. "The funny part is that the company doesn't really need the money."
• An ADR (American Depositary Receipt) lets US institutions own the Seoul-listed, won-denominated shares in dollars on Nasdaq: the company issues new shares, parks them with a custodian bank, and a US depositary issues the receipts. Up to 178M ADRs (~2.5% of the company); each Seoul share (~$1,580) splits into ten ADRs, so one ADR is referenced near $158, priced finally on July 10, listing on the Nasdaq Global Select Market under SKHY.
• Every dollar goes into capacity, split four ways: Yongin Y1 (its next major DRAM/HBM fab), Cheongju P&T7 (advanced HBM packaging / die-stacking), ASML EUV scanners, and a $4B Indiana packaging plant (its first US fab).

Raising from strength, not need (SPCX contrast)

• Contrast with the SpaceX IPO: SpaceX is raising for its CapEx ramp with $10B negative free cash flow in Q1 alone. SK Hynix raises from strength — $24B net cash and an order book sold out through 2028.
• SK Hynix first floated a ~$10B raise; the board settled on $28B. "You don't do that unless internal demand forecasts have moved well past the old memory-cycle playbook."
• A control wrinkle: SK Square (the holding company owning ~20%) must keep its stake above 20% under Korea's holding-company rules, so the deal issues new shares sized to protect that floor rather than selling treasury stock. Some Korean shareholders dislike the ~2.5% dilution; for US investors, "the dilution is the price of admission."

How SK Hynix makes money — the HBM bottleneck

• History: started 1983 as Hyundai Electronics; a 1999 forced merger with LG Semicon buried it in debt, a ~80% memory-price collapse in 2001 made it a "national symbol of corporate failure." Micron offered ~$3.5B for its memory business in 2002 (rejected); SK Group bought control in 2012 and gave a serial cycle-survivor its first patient parent — funding an early bet on HBM.
• Two chip families: DRAM (fast volatile memory feeding processors) and NAND flash (persistent SSD storage), plus a small foundry arm and Solidigm (enterprise SSDs). Q1 2026 IDC share: #2 DRAM (29.1%), #2 NAND (18.5%), and #1 HBM (56.4%) — "the last one is the whole story."
• HBM (high-bandwidth memory) is DRAM stacked vertically beside an AI GPU; as models grew, the constraint shifted from raw compute to how fast you can feed the chip. NVIDIA buys ~two-thirds of its HBM4 from SK Hynix for Vera Rubin. The edge is TSV (through-silicon vias), packaging, yield and "never burning the customer." HBM climbed to 12% of DRAM revenue (double a year ago) at the richest margins in the business.

Q1 FY26 in numbers — a record on every metric

• Revenue +198% Y/Y to $35.5B (52.6T won) — the first quarter ever above 50T won, more than all of 2023. DRAM ASP up mid-60% Q/Q; NAND ASP up sharply with shipments −10% Q/Q as mix moved to high-value products.
• Operating margin 72% (+13pp Q/Q), an all-time high above NVIDIA's or TSMC's most recent prints; operating profit $25.4B (+405% Y/Y). Net margin 77%, net profit $27.3B (+398%). Cash ~$37B, net cash ~$24B.
• The read: it's price, not volume — DRAM ASP jumped mid-60% against roughly flat shipments; HBM is the margin engine (strip it out and these are "good-but-ordinary" numbers); seasonality didn't matter (Q1 is normally weak); demand exceeds three years of capacity; and $24B net cash "flips the model" from the debt-laden SK Hynix of past downturns. But "margins are already peak-like" — the question every buyer must answer is how much of a 72% operating margin survives when supply catches up.

What you're paying — parity with Micron (MU)

• SK Hynix comes to market priced almost exactly like Micron — both near ~7x forward earnings and ~18x trailing EV/EBIT. "The so-called Korea discount for weak governance and limited access is gone." The market is treating the richest margins in memory history as close to a new baseline.
• Bull case: AI changed the pattern — HBM sold out through 2028, priced under multi-year contracts, tied to demand that looks structural. Bear case: "memory is memory, and every past boom claimed to be different, too."
• The ADR's real function is access: US institutions can finally own the HBM leader directly, in dollars, with the liquidity and index eligibility a Nasdaq line brings — that changes who can buy but "doesn't make the stock cheap." At parity with Micron, SK Hynix "requires making the same peak-cycle bet on the same memory boom," the only difference being it leads HBM.

Competition & risks (Samsung, CXMT, YMTC, NVDA)

• Samsung is back: passed NVIDIA's HBM4 qualification and begun shipping; Counterpoint projects ~28% of the 2026 HBM4 market vs SK Hynix's ~54%. The near-monopoly on early HBM4 supply "could shrink from roughly six months to a single quarter."
• China at the low end: CXMT (DRAM) and Yangtze Memory (NAND) scale fast with heavy state backing but trail badly at the high end — "neither is a real HBM threat in 2026." Standard DRAM is fungible; the real pressure lands in 2027+.
• Customer concentration: ~two-thirds of NVIDIA's HBM4 is also an exposure to one company's roadmap and appetite for second sources (NVIDIA already pulling Samsung and Micron in). Plus the cycle ("overbuild, glut, crater, repeat") and dilution/FX (won-dollar exposure, governance/geopolitical factors).

What to watch

• The HBM4 split with Samsung — what share of NVIDIA's Vera Rubin HBM4 SK Hynix actually wins; watch formal allocation and pricing (a narrowing split shows a narrower moat in real time).
• The ADR debut as a signal — pricing and the first weeks of trading are "the most informative read available": a strong open that closes the Micron gap validates the re-rating; a weak one shows US investors still see a Korean memory cyclical.
• CapEx discipline and Yongin timing (Y1 pulled forward to early 2027, M15X ramping, EUV on order) — can it add capacity fast enough to hold the lead without flooding a market it wants tight? And the pricing curve into 2027 — the first quarter of decelerating DRAM/NAND ASPs settles the "structural vs cyclical" debate.

Personal take — IPO = "It's Probably Overpriced"

• The author's golden rule: "IPO stands for It's Probably Overpriced." The twist here — SK Hynix is already public and wildly profitable and raising only to build, so survival was never the risk. "The real question is the price you pay for peak-cycle earnings, and a market that keeps getting more crowded."
• Bull case: the HBM leader is sold out through 2028, on $24B net cash, with a margin profile that "would embarrass most of the Mag 7," now accessible to the deepest pool of AI capital on earth. Bear case: "you're buying peak margins at peak narrative" — Samsung is back, China grinds up, and the bull case rests on "the one thing memory has never delivered: a cycle that doesn't turn."
• Verdict: "SK Hynix belongs on every serious investor's watchlist… But the listing arrives at record margins treated as a new baseline. I'd rather watch the ADR trade through a quarter or two than pay up for permanence the industry has never delivered."