The Nasdaq got one of its biggest listings of the year on July 10, and it came from a Korean memory maker most American retail investors had never been able to buy directly. SK Hynix (SKHY.US) priced ADSs at USD 166 apiece and raised about USD 26.5B — reported as the largest ADR offering on record, ahead of Alibaba's USD 21.9B in 2014. Ten ADSs represent one share of the parent, which keeps trading in Seoul.

This matters because of what the company actually sells. SK Hynix is the world's biggest maker of high-bandwidth memory. Counterpoint put its HBM share at 58% in Q1 2026, with Micron and Samsung at 21% each; IDC has it as the largest DRAM supplier at 29.1% and second in NAND at 18.5%. It is Nvidia's primary HBM partner, holds long-term agreements with roughly ten customers, and has moved HBM4 into volume shipments.

To see why that lands harder than it sounds, remember what memory used to be. DRAM and NAND were the commodity corner of the chip business: you built capacity, caught the upcycle, watched competitors build the same capacity, then ate the downcycle together. The 2018 supercycle and the 2023 inventory glut were the same script with different dates on it.

HBM rewrites part of that script. It's co-designed with the customer, priced by contract, and committed years ahead. Which is why a single Nvidia accelerator launch can quietly set a memory supplier's revenue curve for two years — and why "memory company" stopped being an insult.

The financials are almost absurd. Q2 2026 revenue came in at KRW 79.32T, up 50.9% from the prior quarter per the July 29 filing, with press accounts citing a 257% year-over-year jump. Q1 2026 was KRW 52.58T, up 198% year over year, at a 72% operating margin with KRW 40.35T in net income. For scale: all of 2025 produced KRW 97.1T in revenue and KRW 42.9T in net income. One quarter now approaches a full year.

And yet the DRAM league table tells a less flattering story. Counterpoint's Q2 read puts Samsung first at 38%, SK Hynix down to 25%, and Micron at 24% — a one-point gap. Samsung, meanwhile, began shipping its first HBM4 chips to some customers back in February. SK Hynix leads where the margin is and gives ground where the volume is.

Pricing cuts both ways, too. Industry sources put HBM4 at roughly USD 4/Gb at the mid-to-high end, about triple HBM3E's rate, and DigiTimes reported in July that HBM prices are expected to double by 2027. Excellent if you're selling it. A warning if you're assuming the premium holds, and a direct cost problem if you're the one buying accelerators.

The capital is moving the way the story demands. On Oct. 4, Global Report detailed a USD 4.1B HBM4 fab in Indiana — part of the KRW 100T the company earmarked for new plants, including a NAND line. Reuters has also reported preliminary talks about leasing space at Intel's Ohio site, which SK Hynix confirms are exploratory and nothing more. Putting advanced memory on American soil is equal parts customer demand and policy reality.

The listing itself is part of the argument. In Seoul the stock trades at roughly 8 to 9 times expected earnings, against double digits for Micron in New York, and HSBC lifted its Korean-share target from KRW 2.9M to KRW 4.0M, implying about 20% more upside. Index inclusion would drag passive money along with it, and the company gains a currency for hiring and deals. Whether the discount closes is a governance and market-access question, not an earnings one.

Then there's the detail I keep circling back to. Chey Tae-won, chairman of SK Group, will start selling blocks of SK Inc. shares on Nov. 2 to fund a KRW 944B divorce settlement, taking his stake from 17.8% to about 15.5%, according to AASTOCKS and Chinese media. SK Inc. is SK Hynix's controlling shareholder. On one side: paper wealth created by an AI memory boom. On the other: a founder selling into that boom to settle a private bill. Both things can be true at once.

The sell side is unusually, openly split. Mizuho's Jordan Klein groups Micron (MU.US), SK Hynix, and Samsung as "dirt cheap," pointing to Micron's fiscal quarter ended May 28, 2026 — non-GAAP net income of USD 28.86B, up about 1,224% year over year, with a record 84.9% gross margin — and lifts Micron's target to USD 1,400. The same day, investor The Value Portfolio rates SK Hynix a sell, citing memory cyclicality, capacity additions that could tip into oversupply, and margin pressure. Morgan Stanley is watching the same clock: high-end capacity from several makers lands after 2028, and cloud capex is the variable.

The market, for now, buys first and asks later. SK Hynix rose through the week, beating the S&P 500 by roughly two points, and gained more than 3% intraday on Oct. 1 on 139,541 option contracts, 64% of them calls. Korea's September exports helped: up 83.5% to USD 120.94B, a monthly record, with semiconductor shipments up 262.8%. Gartner expects the total memory market to grow from USD 216B in 2025 to USD 633B in 2026, and HBM revenue from USD 33B in 2025 to USD 86B by 2027.

My view is this: the memory cycle hasn't been repealed. It's been repriced. HBM's contracts genuinely improve visibility, and SK Hynix's position inside Nvidia's supply chain won't be replicated on a short timeline. But capacity, once built, argues for itself — and every fab announced today is a supply decision for 2028. Q3 results land on Oct. 27. The interesting number won't be revenue. It'll be how much price those long-term agreements actually locked in.

The truth, as usual, is more complicated. A company can be the tightest supplier in the AI buildout and part of a family holding that has to sell stock to settle a private bill. There's no contradiction there. They're two faces of the same cycle.

This article does not constitute investment advice.