SK Hynix Launches U.S. IPO Roadshow: In the AI Era, HBM — Not GPU — Is the Scarcest Resource

SK Hynix officially launched its U.S. IPO roadshow this week. The South Korean chipmaker aims to capitalize on strong investor demand for the booming memory chip sector to complete this offering.

According to filings with the U.S. Securities and Exchange Commission (SEC), SK Hynix plans to issue approximately 17.79 million shares of ordinary stock in the form of American Depositary Receipts (ADRs) under the ticker symbol "SKHY." Based on the closing price of its Seoul-listed shares last Friday, the offering could raise around $28 billion. Prior to this ADR issuance, SK Hynix's Seoul-listed stock has surged approximately 260% year-to-date, pushing the company's market capitalization past the $1 trillion mark.

This transaction ranks as the largest foreign company IPO in U.S. market history and is the third-largest IPO globally, trailing only SpaceX and Saudi Aramco.

From Cyclical Component to Core AI Infrastructure

According to SK Hynix's U.S. IPO roadshow materials, this IPO represents not just a massive capital raise but a significant shift in pricing power within AI infrastructure. Historically, memory manufacturers were seen as "highly cyclical" companies vulnerable to supply-demand fluctuations. However, High Bandwidth Memory (HBM) has fundamentally changed this narrative. In the AI compute race, while GPUs handle peak computation, HBM plays a critical role in overcoming the "memory wall" bottleneck by continuously feeding data into computing units.

The roadshow highlights SK Hynix's dominant position in the global HBM market, with a 60% market share. Its technological lead is widening, as it plans to mass-produce HBM4 in February 2026—ahead of competitors—and supply products to NVIDIA's (NVDA-US) next-generation Rubin AI platform. NVIDIA CEO Jensen Huang famously wrote "Please Make More" directly on an SK Hynix wafer, underscoring the strategic importance of its production capacity.

Financial Explosion and Valuation Reassessment

The roadshow materials indicate that SK Hynix's financial performance has exploded due to structural shifts in AI demand. Revenue in Q1 2026 grew 198% year-over-year, operating profit surged over 400%, and profit margins reached 72%. The company's market cap has surpassed $1 trillion, making it one of the few Asian companies to achieve this milestone.

Despite this strong performance, SK Hynix has long faced a "Korea discount" while listed in Seoul. Its forward P/E ratio stands at only 6–8x, significantly lower than U.S. rival Micron's 7–9x. By listing on Nasdaq, SK Hynix aims to attract U.S. institutional capital that cannot easily invest in overseas equities and secure a valuation that reflects its identity as an "AI infrastructure enabler."

Use of $28 Billion in Proceeds

The $28.21 billion raised will be entirely invested in advanced manufacturing capacity in South Korea, not for general corporate purposes. The specific allocation is as follows:

- 31 trillion KRW: Construction of the first wafer fab in the Yongin Semiconductor Cluster - 19 trillion KRW: Building the Cheongju P&T7 advanced packaging facility, dedicated to AI memory packaging - 12 trillion KRW: Procurement of ASML's EUV (Extreme Ultraviolet) lithography machines, essential for producing advanced process nodes

Additionally, the company plans to build an advanced packaging plant in Indiana, with operations expected to begin in 2028, further deepening collaboration with U.S. customers.

Structural Challenges and Potential Risks

However, this investment boom is not without risks. First is legal risk: In June 2026, SK Hynix, along with Samsung and Micron, was named in a U.S. antitrust class-action lawsuit alleging collusion to restrict DRAM supply and inflate prices.

Second is cyclical risk: Although HBM is currently in short supply, the memory industry has historically experienced multiple crashes due to overcapacity. Some analysts warn investors not to ignore the industry's historical inertia.

Finally, the stock price is highly sensitive to AI demand. It once plunged 12% in a single day following rumors that NVIDIA might cut production.