The setup
SK hynix walked in as the HBM leader of the three-maker DRAM oligopoly and the most direct pure-play on the AI-memory cycle — and with the highest bar. The stock had run hard into the print, the sell-side had marched estimates up toward a KRW 64T operating-profit number, and days earlier SK Group and NVIDIA had unveiled a $500 billion-plus comprehensive partnership spanning AI-factory construction and next-generation memory. When the setup is that loaded, a merely record quarter can still disappoint the tape — and this one did on the expectations line even as it set company records on the fundamental one. The result is the cleanest case study of the quarter in the gap between "beat" and "priced in."
SK hynix is the memory maker most levered to HBM and AI-server DRAM, with a NAND business (including the former Intel/Solidigm enterprise-SSD franchise) tilted to high-capacity data-center storage. The second quarter (ended June 2026) is the one in which its HBM4 moved into mass production, its long-term contract book reached roughly half of sales, and — the same week — its largest customer wrote it into a half-trillion-dollar infrastructure commitment. The fundamentals and the structure both stepped up; the stock repriced expectations anyway.
The numbers
Consolidated revenue was KRW 79.3T (~$54B), up 51% sequentially and 257% year-on-year — a record, though below a consensus that had reached roughly KRW 84T (~$57B) (unverified estimate). Operating profit was KRW 60.5T (~$41B), up 61% sequentially and 557% year-on-year, at an all-time-high 76% operating margin (up four points sequentially) — also a record, and also short of a stretched ~KRW 64T (~$43B) Street bar by roughly 5%. (USD conversions throughout are approximate at the prevailing rate near ₩1,480 to the dollar, and unverified to the rate.) That "record that missed" is the whole tension: on any absolute measure this is one of the most profitable quarters a memory maker has ever printed; against the bar the market had set, it landed light.
The bottom line needs a caveat, and it has a name: Kioxia. Reported net profit was KRW 93.9T (~$63B) — larger than operating profit and larger than the quarter's revenue — because it is inflated by a large, one-time non-operating gain on SK hynix's long-held stake in the Japanese NAND maker. Through a Bain-led special-purpose vehicle, SK hynix holds roughly a 14% interest in Kioxia (its second-largest shareholder), acquired for about KRW 3.9T (~$2.7B) in 2018 in a mix of equity and convertible bonds. After Kioxia's December-2024 Tokyo listing and the AI-driven rally in its shares since, SK hynix recognized on the order of KRW 40T (~$27B) of Kioxia-related non-operating income this quarter — a realized gain from shares the first vehicle (SPC1) sold last month, plus a fair-value revaluation on the remaining holding and its convertible bonds (SPC2). That accounts for most of the ~KRW 62T (~$42B) gap between pre-tax profit (~KRW 122.7T, ~$83B) and the KRW 60.5T operating line; the balance is other non-operating items the release did not itemize and the audited report will formalize. Treat the headline net figure the way you would a one-off asset sale: the clean, repeatable read of the quarter is the KRW 60.5T (~$41B) operating line at a 76% margin — the Kioxia windfall is real cash and real value, but it is not the memory business. By mix, DRAM was ~73% of revenue and NAND ~27%, with enterprise SSD more than half of NAND. The balance sheet inflected hard: cash of ~KRW 88T (~$59B), net cash of ~KRW 69.4T (~$47B) (up ~KRW 33.6T sequentially) as debt fell to ~KRW 18.6T (~$13B).
The rest of this dive is for paid subscribers.
The headline numbers are above. The 7 sections below carry the mechanism, the peer read, the valuation work, and the dated tests that decide it.
- What's actually driving it
- The guide
- How peers read this
- Valuation reality check
- What we'd watch from here
- Listed-market read-through
- Bottom line
Already a subscriber? Sign in.