On Friday, July 10, a South Korean company that most consumers have never knowingly bought anything from rang the opening bell at the Nasdaq MarketSite in Times Square. By the closing bell it had completed the largest U.S. share sale ever done by a foreign company. The chipmaker is SK hynix, the world's second-largest memory manufacturer, and the money it raised is a fairly blunt statement about where the AI buildout has run into a wall: not compute, but memory.
What Happened
SK hynix priced and floated American depositary receipts on the Nasdaq, raising roughly $26.5 billion. That figure edges past the $25 billion Alibaba raised in its 2014 New York debut, which had held the record for a foreign issuer for more than a decade. The offering comprised about 177.9 million ADRs, with every ten ADRs representing one underlying common share.
The receipts were priced at $149. They opened at $170, drifted back through the session, and settled at $168.01 — a gain of roughly 13% against the offer price on day one. The book was reported as more than seven times oversubscribed. Friday's session ran under the provisional ticker SKHYV, with the permanent designation SKHY taking over when trading resumed the following week.
One detail matters more than it sounds: this is not a spin-off or a relocation. SK hynix remains listed in Seoul, where it has traded for decades and where it sits among the heaviest weights on the KOSPI. The Nasdaq ADR is a secondary listing — a second tap into a much deeper pool of dollars, layered on top of the existing one. Nothing about the company's ownership or its Korean fabs changed on Friday. What changed is who can buy it, and in what currency it can fund itself. CNBC's pre-debut explainer laid out the mechanics ahead of the open.
Why It Matters
For most of the last thirty years, DRAM was the least glamorous thing in a computer. It was a commodity, priced like one, and it moved in brutal boom-and-bust cycles that periodically wiped out whoever had built capacity at the wrong moment. Memory makers were the people you called when you needed a part number, not a partner.
That relationship has inverted. High-bandwidth memory — HBM, the stacked DRAM that sits next to an AI accelerator and feeds it data — has become the component that decides how many AI systems the world can actually build in a given quarter. SK hynix holds roughly 56% of the HBM market by IDC's first-quarter 2026 count, and around 32% of global DRAM overall. That is not a commodity supplier's position. That is a chokepoint.
The financials reflect it. SK hynix reported first-quarter 2026 revenue of about 52.6 trillion won with operating profit near 37.6 trillion won — margins that memory companies historically did not get to have. When a business with that shape decides to raise $26.5 billion in equity rather than borrow, the market reads it as a company converting an unusually rich valuation into physical capacity while the window is open.
Where the Money Actually Goes
SK hynix has been specific about the use of proceeds, which is refreshingly unusual for a raise of this size. The money is earmarked for the first wafer fab of the Yongin Semiconductor Cluster, the Cheongju P&T7 advanced packaging plant, and equipment purchases — most conspicuously EUV lithography systems, which come from exactly one supplier on earth.
The Yongin phase-one fab, designated Y1, carries a commitment of roughly 31 trillion won — about $21.5 billion — and is slated for structural completion around February 2027, with tool installation following in the quarter after. Cheongju's packaging line matters just as much as the fab, because HBM is as much an advanced-packaging problem as a lithography one: the value is in stacking and bonding the dies, not only in printing them.
Here is the part that should temper any narrative of the raise "fixing" anything. None of these projects produce a single sellable chip in time to relieve the memory shortage currently pushing prices up across the industry. Y1 is a 2027–2028 story at best. The capital being raised today is for the demand of two or three years from now — which means the entire raise is, structurally, a bet that AI memory demand is still there when the concrete cures.
The Demand Behind the Numbers
That bet is not being made in a vacuum. On June 7, in Seoul, SK hynix and Nvidia announced a multiyear technology partnership explicitly aimed at what Nvidia now calls AI factories. Under it, SK hynix will co-develop memory for Nvidia's Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-class PCs, and the Jetson Thor robotics platform.
The agreement goes further than a supply contract. The two companies also said they would apply AI to semiconductor design and manufacturing itself, and that SK hynix would build factory digital twins using Nvidia's Omniverse tooling with the eventual goal of largely autonomous fab operation. Read plainly: the memory supplier and the accelerator designer are now co-engineering the roadmap, which is a very different arrangement from a chipmaker bidding on a purchase order. Nvidia's own announcement of the partnership and SK hynix's newsroom post describe the same scope from both sides.
SK Group chairman Chey Tae-won has been saying for a while that AI demand will keep memory supply tight until around 2030. Whether or not that turns out to be right, it is clearly the assumption the capital plan is built on. And with HBM allocations effectively spoken for well in advance, the pricing dynamic has shifted from spot-market commodity toward long-dated contract — which is precisely what makes a fab commitment of this scale financeable in the first place.
What Comes Next
The immediate mechanics are mundane: SKHY becomes the permanent ticker, index providers work out what a dual-listed Korean giant does to their weightings, and the usual ADR-versus-ordinary spread gets arbitraged into line. The more interesting question is whether the raise marks a top or a middle.
The bear case writes itself, and it is not a fringe view. Memory is the most reliably cyclical business in technology. Every previous supercycle ended when capacity that was commissioned at the peak arrived into softening demand, and there is a lot of capacity being commissioned right now — Samsung and Micron are both expanding HBM output, and Samsung is reported to have taken a meaningful share of Nvidia's next-generation HBM orders. A handful of AI buyers account for an outsized slice of the order book, which is customer concentration by another name.
The bull case is simply that this cycle is structurally different because the demand is coming from capital budgets at the largest companies on earth rather than from consumer PC refreshes, and because HBM's technical difficulty makes it far harder to flood with supply than commodity DRAM ever was. Both cases are coherent. The listing does not settle the argument; it just makes it a much larger one, conducted in public, in dollars. CNN's coverage of the debut framed the same tension.
Closing Thoughts
There is something quietly funny about a memory company breaking Alibaba's record. Alibaba's listing was a story about consumers, platforms, and a billion people shopping. This one is a story about a rectangle of silicon that exists to hold numbers still for a few nanoseconds while something else does the thinking.
But that is the shape of this era of the industry. The scarce thing is not the idea, and increasingly it is not even the model — it is the physical substrate that lets a model run at a price anyone can afford. For twenty years the memory business was where you went to lose money slowly. Now it is where the AI buildout goes to find out whether it can proceed at all. Friday's $26.5 billion was, in the end, the market pricing that fact.
한글 요약
SK하이닉스가 7월 10일(현지시간) 미국 나스닥에 주식예탁증서(ADR)를 상장하며 약 265억 달러를 조달했습니다. 이는 2014년 알리바바의 250억 달러를 넘어서는, 외국 기업의 미국 증시 사상 최대 규모 공모입니다. 총 1억 7,790만 ADR이 발행됐고(ADR 10주 = 보통주 1주), 공모가 149달러로 시작해 상장 첫날 170달러에 개장한 뒤 168.01달러에 마감하며 공모가 대비 약 13% 올랐습니다. 청약은 7배 넘게 몰렸습니다. 서울 상장은 그대로 유지되며, 이번 나스닥 상장은 달러 자금 조달 창구를 추가한 2차 상장입니다.
조달 자금의 용도는 명확합니다. 용인 반도체 클러스터 1기 팹(Y1, 1단계 약 31조 원, 2027년 2월경 준공 목표), 청주 P&T7 첨단 패키징 공장, 그리고 EUV 노광장비 등 설비 투자입니다. 다만 이 투자들은 지금의 메모리 공급 부족을 해소하지 못합니다. 실제 양산은 2027~2028년 이후이기 때문에, 이번 조달은 사실상 '2~3년 뒤에도 AI 메모리 수요가 남아 있을 것'이라는 베팅에 가깝습니다. 수요 측면에서는 6월 7일 서울에서 발표된 엔비디아와의 다년간 기술 협력(Vera Rubin AI 슈퍼컴퓨터·Vera CPU·RTX Spark PC·Jetson Thor용 메모리 공동 개발, 옴니버스 기반 팹 디지털 트윈)이 근거가 됩니다. SK하이닉스는 IDC 기준 2026년 1분기 HBM 시장의 약 56%, 전체 D램의 약 32%를 점유하고 있습니다.
물론 반론도 분명합니다. 메모리는 기술 산업에서 가장 주기성이 강한 사업이고, 과거의 호황은 늘 고점에 착공한 설비가 수요 둔화기에 가동되며 끝났습니다. 삼성전자와 마이크론도 HBM 증설에 나섰고, 주문의 상당 부분이 소수 AI 고객에 집중돼 있다는 점도 위험 요인입니다. 반대로 이번 사이클은 소비자 PC 교체 수요가 아니라 초대형 기업들의 설비 예산에서 나오고, HBM은 범용 D램처럼 쉽게 공급을 늘릴 수 없다는 점에서 다르다는 시각도 설득력이 있습니다. 이번 상장이 그 논쟁을 끝내지는 않았지만, 논쟁의 규모를 훨씬 키운 것만은 분명합니다.
참고: CNBC · CNN Business · NVIDIA Newsroom · SK hynix Newsroom