SK Hynix Raises $26.5B in Largest Foreign IPO in US History, Powered by AI Memory Boom

On July 10, 2026, a South Korean company rewrote the record books on Wall Street. SK Hynix, the world’s dominant maker of high-bandwidth memory chips, raised $26.5 billion in its Nasdaq debut — the largest initial public offering ever by a foreign company on a US exchange, eclipsing Alibaba’s 2014 blockbuster by more than a billion dollars.

The numbers were staggering. The company sold 177.9 million American depositary receipts at $149 each. The offering was more than seven times oversubscribed. By the end of the first trading session, shares had climbed over 13% above the IPO price. SK Hynix’s chairman, Lee Kyu-hwan, appeared on CNBC mid-session to declare simply: “Demand is enormous.”

The Memory Chip Behind Every AI Breakthrough

To understand why investors stampeded into SK Hynix’s offering, one must understand the chip that makes modern AI run. High-bandwidth memory — HBM — is the specialized semiconductor that sits directly on top of AI accelerators like Nvidia’s H100 and the company’s next-generation Vera Rubin platform. Unlike conventional DRAM, HBM stacks memory dies vertically using through-silicon vias, delivering data to the GPU at speeds that standard memory cannot approach.

Without HBM, large language models cannot train or inference at scale. And right now, SK Hynix makes roughly 60% of the world’s supply.

The company’s relationship with Nvidia is not just commercial — it is architectural. SK Hynix engineers co-develop HBM generations in close coordination with Nvidia’s chip design teams, a level of technical lock-in that makes switching suppliers costly and slow. When Nvidia’s Vera Rubin GPU ships later this year, it will depend on SK Hynix’s HBM4 — a product that has already been under joint development for two years.

Escaping the Korea Discount

For decades, South Korean companies suffered a persistent “Korea Discount” — a valuation markdown applied by global investors who cited concerns about governance structures, cross-shareholding complexity, and the geopolitical risks of operating near North Korea. Samsung Electronics, LG, and SK Group companies have long traded at meaningful discounts to their Taiwanese and US peers with similar fundamentals.

SK Hynix’s Nasdaq listing was partly designed to escape that discount. By listing in the US, creating a liquid ADR that American institutional investors can buy and sell in their home time zone, and submitting to SEC disclosure requirements, the company essentially signaled that it was willing to play by Western market rules to access Western market valuations.

The strategy worked. Seven-times oversubscription suggests that at $149 per ADR, the market felt the shares were meaningfully underpriced even accounting for geopolitical risk premiums.

Capital Deployment: A New Korean Fab and Next-Gen Equipment

SK Hynix plans to deploy the $26.5 billion raised across three areas. The primary use of proceeds is construction of a new memory fabrication facility in South Korea specifically designed to alleviate the global shortage of AI-grade memory. The company will also invest in advanced packaging infrastructure — the manufacturing step where HBM dies are stacked and bonded — and in EUV (extreme ultraviolet) scanner equipment necessary for producing the next generation of memory at 3nm-class geometries.

The fab investment comes at a geopolitically sensitive moment. US Commerce Secretary Howard Lutnick has been publicly and privately lobbying both SK Hynix and Samsung to build manufacturing facilities inside the United States, framing AI memory as a national security matter on par with leading-edge logic chips. The Chips Act has already committed tens of billions to bringing TSMC and Intel back to competitive fab status; memory chips are seen as the next frontier.

SK Hynix has not announced US fab plans, though the pressure from Washington is mounting. Rival Micron, which announced a $250 billion US investment plan earlier this year, has positioned itself as the American alternative to Korean memory dominance — a narrative that may accelerate Washington’s pressure on Seoul.

What Wall Street Is Actually Buying

Investors in SK Hynix’s Nasdaq offering were not simply buying a memory chip company. They were buying exposure to what analysts increasingly describe as a structural constraint in the AI supply chain.

GPU compute — once the assumed bottleneck — has scaled remarkably in capacity since 2023. Nvidia builds chips by the millions; its partners deploy them across hyperscale data centers at a pace that has surprised even internal forecasters. But HBM production has not kept pace. The stacking process is technically demanding, yield-sensitive, and dependent on a narrow supply chain of bonding equipment dominated by companies like BE Semiconductor Industries.

This asymmetry — abundant GPU, constrained HBM — has made SK Hynix’s position increasingly rare in a commodifying semiconductor landscape. The company isn’t just selling memory; it’s selling a capability that the rest of the AI industry cannot easily replicate.

A Record With Geopolitical Undertones

The timing of the listing was not accidental. The US-Korea semiconductor relationship has grown strategically vital as Washington seeks to reduce dependence on Chinese chip supply chains and reinforce allied semiconductor ecosystems. South Korea is a central pillar of that strategy, alongside Taiwan and Japan.

SK Hynix’s $26.5 billion US listing deepens the financial ties between the two countries’ semiconductor industries in a visible and durable way. American pension funds, mutual funds, and retail investors now have direct exposure to South Korea’s most critical chipmaker. That creates a constituency for continued US-Korea cooperation that goes beyond diplomatic statements.

Whether or not Lutnick succeeds in persuading SK Hynix to build US fabs, the Nasdaq listing has already accomplished something significant: it has made one of the AI era’s most essential suppliers a permanent fixture of the American financial landscape.

For a company that a decade ago was seen as a commoditized memory maker competing on cost, that is a remarkable transformation — and a testament to what controlling the memory layer of the AI stack is worth in 2026.