SK Hynix commits $29 billion to scale AI-grade HBM memory production via major US listing and ADR expansion
Three months ago, SK Hynix was quietly discussing a potential U.S. listing valued at around $14 billion. When the company confirmed its plans this week, the valuation had grown to $29.65 billion. This surge reflects how dramatically AI infrastructure spending has accelerated since March and the central position SK Hynix now occupies in the global AI investment cycle.
The South Korean memory maker said Wednesday it plans to sell American depositary receipts on the Nasdaq Global Select Market, with trading expected to begin July 10, subject to regulatory approval. At the top of its range, the offering would surpass both Saudi Aramco's $25.6 billion IPO in 2019 and Alibaba Group's $21.8 billion New York debut in 2014, making it the largest ADR transaction in history. Bookbuilding is ongoing and final pricing has not been set; the $29.65 billion figure represents a ceiling, not a confirmed outcome.
The intended use of proceeds is unambiguous. Every dollar will fund physical manufacturing capacity: expansion of the Yongin semiconductor cluster on the Korean peninsula, construction of the Cheongju P&T7 advanced-packaging facility dedicated to high-bandwidth memory, and acquisition of extreme ultraviolet lithography scanners from ASML, the Dutch equipment maker whose technology defines the frontier of advanced chip production. A single EUV tool costs approximately $380 million. SK Hynix intends to acquire multiple units.
The urgency derives directly from Nvidia. SK Hynix supplies the HBM3E stacks integrated into the chips powering Nvidia's H200 and B200 AI accelerators, the processors at the core of data centers running Google Gemini, ChatGPT, and comparable systems at scale. The company controls roughly 60 percent of the global high-bandwidth memory market, and its entire HBM supply through 2026 is fully contracted. No spare inventory exists at any price. Nvidia needs more of these chips than currently exist, and SK Hynix must build the fabs to produce them.
This constraint drove a 12 percent surge in SK Hynix shares on the Seoul exchange when listing plans were disclosed—a market read not of current earnings but of the value of controlling the most critical component in the world's fastest-growing hardware category. The AI memory supercycle, which TrendForce projects will sustain HBM price premiums through at least 2027, has made SK Hynix's market position worth considerably more than its pre-AI valuation reflected.
The offering converts each common Korean share into 10 ADRs, giving U.S. institutional and retail investors direct dollar-denominated exposure without navigating the Korea Stock Exchange. The company said it expects the Nasdaq listing to expand its investor base and allow its "true corporate value to be properly evaluated"—a pointed remark, given that SK Hynix has historically traded at a discount to Micron Technology on a price-to-earnings basis despite holding a larger share of the AI memory market. Closing that valuation gap is a stated objective of the listing.
Samsung Electronics, which holds the second-largest position in high-bandwidth memory, has struggled to certify its HBM3E chips to Nvidia's specifications and has ceded market share as a result. Micron Technology holds the remaining portion of HBM supply, reporting last week that its production is also fully sold out through 2026 and projecting quarterly revenue of $50 billion. The memory market is a three-party oligopoly in which all suppliers are running at capacity, none can fulfill incremental demand, and the largest is now seeking $30 billion to build more.
This dynamic has cascaded into consumer electronics. As Apple and Microsoft demonstrated last week when raising prices on Mac computers, iPads, and Xbox consoles, AI infrastructure demand has crowded out the standard DRAM supply that consumer electronics companies depend on. SK Hynix's new fabs would theoretically ease some pressure, but the Yongin cluster is designed to produce HBM for AI customers, not DRAM for laptop makers. The consumer electronics industry's memory shortage will persist until AI demand growth slows or Samsung closes its HBM certification gap and frees up mainstream DRAM production.
South Korea's government has positioned semiconductors at the center of its industrial strategy, offering tax incentives and infrastructure support for the Yongin cluster that SK Hynix's listing proceeds will partially fund. The listing is more than a capital raise; it is a statement about where the company and its government believe the next decade of semiconductor manufacturing leadership will be decided: in the AI memory stack, at the frontier of EUV lithography, at a scale only such a listing can support.
What remains uncertain is the most critical variable: whether the HBM supercycle sustains. Memory chip markets carry cyclical histories that dwarf almost every other semiconductor category. DRAM and NAND have moved through boom-bust cycles measured in years. HBM has so far avoided these dynamics only because supply has been genuinely insufficient against surging AI demand. That calculus could shift if hyperscaler capital budgets contract, if Nvidia's next-generation chips prove more memory-efficient than the H200, or if Samsung closes the certification gap faster than markets assume. At those inflection points, the economics underlying a $29.65 billion raise would look considerably different, and whether SK Hynix has locked in its lead or borrowed against its peak would become the urgent question.
The market will begin pricing this question on July 10, when SK Hynix ADRs open on Nasdaq and the world's largest memory chipmaker by market position becomes available to every U.S. retail investor. The number that matters most that morning is not $29.65 billion. It is where the ADRs actually trade.