SK Hynix just pulled off the second-biggest IPO in history, and the number that should worry every AI startup founder isn't the $26.5 billion raised. It's what the company plans to spend it on.
SK Hynix priced its American depositary shares at $149 apiece, selling 177.9 million of them to raise about $26.5 billion ahead of its Nasdaq debut on July 10. Trading opened under the temporary ticker SKHYV before converting to the permanent symbol SKHY, and the stock closed its first session up roughly 6%, according to CNBC and stockanalysis.com. That makes it the largest US listing ever by a foreign company, surpassing Alibaba's $25 billion debut back in 2014, as Al Jazeera reported. Globally, only SpaceX's $85.7 billion Nasdaq listing last month raised more.
Demand outran supply by more than seven times. TipRanks and MarketScreener both reported the offering was oversubscribed sevenfold, with orders flooding in from global long-only funds and technology-focused investors who wanted exposure to the AI memory trade without waiting for a Seoul-listed stock to become easily tradable in New York.
Here's the part that actually matters for anyone building on top of AI infrastructure. According to reporting cited by KED Global, the bulk of the proceeds isn't going to a stock buyback or a war chest. It's earmarked for concrete, dated construction: roughly $21.5 billion toward the first fab at the new Yongin Semiconductor Cluster near Seoul, due to finish around February 2027, and another $12.9 billion for a packaging plant in Cheongju built specifically to assemble and test HBM chips, targeted for completion by late 2027. SK Hynix had already committed $7.9 billion to ASML in March for roughly 30 EUV lithography scanners to equip both sites. Those are hard, dated construction commitments - SK Hynix racing to build capacity it already knows it needs.
You can build all the data centers you want, but you cannot run them without high bandwidth memory, and right now there isn't enough of it. Micron has said its HBM capacity is sold out through 2027. Across SK Hynix, Samsung, and Micron, HBM output for 2026 is already spoken for. Micron's stock surged 19% in a single day on May 26 and pushed its market cap past $1 trillion, not because it shipped a breakthrough product but because Wall Street finally priced in what engineers have known for a year: memory, not compute, is the tightest link in the AI supply chain.
Samsung is responding the same way SK Hynix is, with concrete rather than press releases. The company is targeting roughly 50% more HBM production capacity in 2026 and used Nvidia's GTC conference this year to show off HBM4E modules rated at 16 gigabits per pin and 4.0 terabytes per second of bandwidth, built for Nvidia's Vera Rubin platform. SK Hynix still holds the technical lead in HBM specifically, even though Samsung ships more memory overall.
Frankly, the research brief's question about whether Samsung or Micron follow SK Hynix onto Wall Street with their own US listings gets ahead of the actual reporting. Neither company has filed anything, and nothing in the coverage around this IPO points to an imminent copycat listing. What the reporting does show is that both are already spending aggressively without needing a Nasdaq ticker to fund it, which tells you the capital is available through ordinary channels when the demand is this real.
What SK Hynix's listing actually signals to founders and investors is narrower and more useful than a prediction about who lists next. It's pricing information. Investors were willing to pay $149 a share, sevenfold oversubscribed, for a company whose growth story is entirely about whether it can pour concrete and install EUV scanners fast enough to keep up with orders it already has in hand. That's a scarcity premium, not a growth multiple on some speculative product. If you're building an AI company and your unit economics assume memory prices stay where they were two years ago, this IPO is the market telling you otherwise.
SK Hynix's own pricing materials put it plainly: the AI memory cycle is real, and the earnings behind it are real too. The fabs won't be finished until 2027. Until then, the chips stay scarce, and scarcity is exactly what investors just paid a record price to bet on.
Also read: Investors Are Rotating Out of Nvidia and Into Chinese Chip Stocks • Tencent Leads Buyback of AI Startup Manus From Meta at Original Price • Nexchip's $890 Million Hong Kong Debut Tests China's Chip Strategy