SK Hynix Just Raised $26.5B in the Biggest Foreign IPO in US History. The Bet Is That the RAM Shortage Never Ends.
TL;DR
On July 10, SK hynix sold 177.9 million American depositary receipts at $149 apiece and raised $26.5 billion, the largest US listing by a foreign company in history, beating the record Alibaba set in 2014. The offer was more than seven times oversubscribed, opened around $170, and closed its first session near $168, up roughly 13%. The company that just took the record is not a model lab or a GPU designer. It makes memory, the part of the AI boom nobody brings up at parties, and Wall Street just handed it a trillion-dollar valuation on the theory that the shortage feeding it is structural, not cyclical.
Day one, by the numbers
The mechanics: each ADR represents a tenth of a Seoul-listed share, priced at a 2.7% premium to the Seoul three-day average (efficient markets, give or take). Shares trade under the temporary ticker SKHYV until Monday, July 13, when they flip to the permanent SKHY. Demand ran seven times the available allocation, per the company's Form F-1 roadshow coverage and day-one reporting.
The Seoul-listed stock had already run 174% in six months on the AI memory squeeze, per Yahoo Finance, and the US listing let it shed the "Korea discount" that usually depresses Korean valuations. Why did American investors pay up anyway? Because SK hynix sells the one component every AI roadmap quietly assumes will be available, and right now it is not.
Why a RAM company got model-lab money
SK hynix is the dominant maker of high-bandwidth memory, the stacked DRAM that feeds data to AI accelerators. Per Counterpoint Research figures cited in SK hynix's own 2026 outlook, it ships 62% of the world's HBM and books 57% of HBM revenue, ahead of Samsung and Micron combined. Bank of America expects the HBM market to hit $54.6 billion this year, up 58% year over year.
HBM matters because modern accelerators are bandwidth-starved, not compute-starved: the GPU can do the math far faster than ordinary memory can deliver the operands. HBM fixes that by stacking DRAM dies and bonding them right next to the processor die. Think of it as moving the pantry into the kitchen: instead of the chef sprinting across the building for every ingredient, the shelves sit an arm's length from the stove. That physical proximity is why Nvidia signed a multiyear co-development partnership with SK hynix in June covering memory for its Vera Rubin AI supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robotics platforms. When the world's most valuable chip designer locks in your roadmap across four product lines, an IPO prospectus mostly writes itself.
Where the $26.5 billion goes, and why your RAM bill stays ugly
Per TechCrunch, the proceeds target three things: a new fabrication plant in South Korea, an advanced packaging facility, and EUV lithography scanners for next-generation nodes. None of that relieves the shortage soon. Fabs take years to build and qualify, which is exactly why analysts quoted by Yahoo Finance expect the memory shortage to persist into 2030.
The demand side is getting locked up too. Micron's chief business officer Sumit Sadana told Yahoo Finance that memory supply agreements now mostly run five years, where the old norm was one. Hyperscalers are signing half-decade commitments for output that does not exist yet, which puts everyone buying at spot prices, meaning you and your homelab, at the back of the line. The DRAM squeeze that already pushed Apple and Xbox to raise prices is the same squeeze this IPO is monetizing.
The straight-faced caveats
Memory is the most cyclical business in semiconductors, and the graveyard is full of people who called a supercycle at the top. Analyst Patrick Moorhead reminded Yahoo Finance that not long ago memory makers "were literally selling stuff below costs, and then they rapidly pulled back on capex." SK hynix's own 2026 outlook flags possible pricing corrections after 2026 as competitors add capacity. A seven-times-oversubscribed IPO at a trillion-dollar valuation prices in a long boom; it does not guarantee one.
There is also political pressure attached to the money. TechCrunch reports Commerce Secretary Howard Lutnick is pushing SK hynix and Samsung to build fabs on US soil, with Micron already committing $250 billion to US manufacturing. Where those fabs land will shape memory logistics, subsidies, and pricing for the rest of the decade.
What builders should actually watch
- DDR5 and used-GPU prices: a shortage priced as structural means consumer DRAM stays expensive. If you were waiting out the RAM market to expand a local-inference box, the market just voted against you.
- HBM allocation: analysts estimate SK hynix holds 60-70% of the HBM4 earmarked for Nvidia's Vera Rubin platform. Accelerator availability downstream tracks that allocation, not GPU die supply.
- Contract structure: five-year supply deals mean capacity is spoken for before it is built. Spot buyers absorb the volatility.
- The listing window: a record-setting, oversubscribed AI-infrastructure debut tells every AI company weighing a US listing that the window is wide open.
Key Takeaways
- SK hynix raised $26.5 billion on the Nasdaq on July 10, the largest US IPO by a foreign company ever, passing Alibaba's 2014 record of $25 billion.
- The offer was 7x oversubscribed and closed day one up about 13%, near $168 per ADR, valuing the company above $1 trillion.
- SK hynix ships 62% of global HBM, and Nvidia locked it into a multiyear memory partnership across the entire Vera Rubin generation in June.
- Proceeds fund a new Korean fab, advanced packaging, and EUV tools, but analysts still expect the memory shortage to run into 2030.
- Supply contracts have stretched from one year to five, so spot buyers, including homelab builders, stay at the back of the line while prices hold.
- Memory is brutally cyclical; SK hynix's own outlook flags pricing corrections after 2026 as new capacity arrives.
Sources: TechCrunch, Yahoo Finance, NVIDIA Newsroom, SK hynix Newsroom, SEC EDGAR (Form F-1), CNBC