Despite a significant sell-off this week that hit global chip stocks due to concerns over overly crowded positions and excessive leverage in AI computing-related tech shares, South Korea-based SK Hynix Inc., one of the world's top three memory chip manufacturers and the leader in HBM market share, is seeking to raise 45.45 trillion won ($29.4 billion) through a listing on the US stock market.
This move aims to capitalize on robust global investor demand for shares of memory chip giants, which have seen their stock prices soar in recent years and remain in extremely short supply.
The plan for SK Hynix to list in the US, coupled with the exceptionally strong results and outlook just reported by US memory chip giant Micron Technology (MU), powerfully underscores that the memory chip super cycle is far from concluding.
The proposed issuance of American Depositary Receipts (ADRs) comes after SK Hynix's shares, traded on the Seoul market, have skyrocketed approximately 850% over the past 12 months, propelling the company's market capitalization above $1 trillion and briefly surpassing that of long-time South Korean leader Samsung Electronics.
Investor sentiment around semiconductor companies tied to the AI computing infrastructure boom has been volatile recently, with this year's overall rally punctuated by sharp two-way swings, as positions have grown increasingly crowded and leveraged strategies have expanded.
The US listing plan also marks a milestone in SK Hynix's extraordinary ascent. After becoming the preferred HBM supplier for NVIDIA (NVDA), the dominant force in AI chips, it has emerged as the global market share leader in High Bandwidth Memory (HBM) systems.
This has allowed it to surpass its long-time domestic rival Samsung Electronics in both market capitalization and overall DRAM share at times.
SK Hynix's plan to raise roughly $29.4 billion via a US listing, combined with Micron's stellar results and guidance, significantly reinforces a firm market conviction: the so-called "memory super cycle" is not over, and has evolved from a traditional PC/smartphone cycle into a structurally scarce super cycle driven by AI data centers.
Micron's revenue guidance for the next quarter remains well above Wall Street expectations, interpreted by the market as a sign that demand for HBM, high-capacity DRAM, and enterprise NAND memory chips continues to explode against the backdrop of fervent AI data center construction and persistently strong AI computing infrastructure needs.
Micron's latest earnings report for its fiscal third quarter ending May 31 showed revenue surged approximately 346% year-over-year to $41.46 billion, indicating the AI computing infrastructure boom continues to propel the memory chip industry into an upcycle, temporarily alleviating market concerns about AI-related memory demand.
The most compelling rebuttal to market doubts lies in the explosive data center and cloud business segments.
Core Data Center revenue for the third quarter was $11.52 billion, 7.5 times the year-ago figure and 1.7 times market expectations; Cloud Storage Chip revenue was $13.77 billion, roughly four times the year-ago quarter and nearly 30% above analyst forecasts.
Combined, these two segments accounted for approximately 61% of total company revenue at $25.29 billion.
Furthermore, Micron anticipates adjusted revenue for the fourth fiscal quarter to be between $49 billion and $51 billion, implying another record high this quarter, with the midpoint of the guidance range ($50 billion) nearly 20% above the analyst consensus estimate.
JPMorgan's bullish thesis for memory chip manufacturers is more aggressive than some peers, positing that this memory super cycle will be "higher and longer" because AI demand has spread from GPUs to CPUs, ASICs, inference servers, and agent computing systems.
While GPUs handle parallel matrix computations, agent AI requires CPUs for task orchestration, state management, tool invocation, and API execution, which significantly boosts demand for server DRAM, HBM, and enterprise SSDs.
Under this framework, the value share of memory within cloud service providers' hardware capital expenditure has jumped from the low teens percentage points in the early AI phase to potentially over 50% this year, and could approach 73% by 2030.
This signifies that memory is no longer just a supporting component for servers but a core resource determining the throughput, latency, energy efficiency, and scalability of AI systems.
Soon, SK Hynix, the HBM leader capitalizing on the AI frenzy to list in the US, will join other global tech giants raising record sums in the US market to fund AI computing infrastructure builds.
SpaceX completed the largest initial public offering in history earlier this month, while Alphabet Inc. recently announced plans to raise $85 billion to fund its ambitious AI application and computing goals.
Although investors have generally welcomed these heavily oversubscribed US IPOs, volatility in AI-related tech stocks persists.
The Philadelphia Semiconductor Index, a key global chip stock benchmark, fell sharply by 7.9% on Tuesday, marking the ninth trading day in the past month with a single-day move exceeding 5%.
According to a South Korean regulatory filing on Wednesday, SK Hynix expects its US ADRs to begin trading officially on July 10.
Following the ADR announcement and Micron's strong results, shares of US memory leaders like Western Digital, SanDisk, and Seagate showed strong gains in after-hours trading on Wednesday.
The offering is being led by Wall Street giants Bank of America, Citigroup, Goldman Sachs, and JPMorgan.
A filing with the US Securities and Exchange Commission indicates the company expects the ADRs to trade on the Nasdaq Global Select Market under the ticker symbol SKHY.
The core logic behind global capital embracing "silicon-based inflation" is clear: in the AI era, the greatest scarcity lies not in traditional labor, real estate, or general manufacturing capacity, but in "silicon-based means of production" like GPUs/ASICs, HBM/DRAM/NAND memory chips, data center CPU components, high-performance Ethernet infrastructure, advanced packaging capacity, cutting-edge semiconductor manufacturing equipment like EUV, data center power chains, and data center optical interconnects/communications.
The grand investment narrative this year of "seeking silicon-based inflation, weakening carbon-based" is essentially capital shifting from traditional "carbon-based assets" reliant on population, resources, and linear economic growth—like manufacturing, autos, consumer goods, real estate, and energy—towards high-end manufacturing chains centered on silicon wafers related to AI computing infrastructure.
This is not merely a narrative chase for tech stocks; it is global capital repricing the "core vehicle for future growth." Entities controlling AI computing infrastructure resources for AI training/inference command higher valuation premiums, as capital allocation weight shifts from old-economy assets dependent on population, oil & gas, real estate, and consumer cycles towards infrastructure assets capable of supporting AI training/inference and the expansion of automated physical AI productivity.
Growth elasticity in the carbon-based economy comes from people, cars, housing, oil, and linear consumption; growth elasticity in the silicon-based economy broadly stems from model parameters, computing infrastructure clusters, memory bandwidth, data centers, and AI agent automation.
Global capital is betting that inflation and profit pools for the next decade will depend not just on oil and labor, but increasingly on the scarcity repricing of silicon-based economy elements like computing power, memory/storage, power-related semiconductor capacity, and semiconductor manufacturing equipment.
Ranking Among Top Three Global IPOs
Based on the proposed fundraising and issuance size, which is subject to exchange rates, SK Hynix's US ADR offering would rank among the top three initial stock offerings in global market history.
According to compiled data, it would be comparable to Saudi Aramco's $29.4 billion mega-IPO in 2019, while the largest global IPO remains the unparalleled $75 billion-plus fundraising by Elon Musk's SpaceX.
This listing will also test investors' appetite for AI semiconductor exposure and the market's capacity to absorb additional new share supply.
It also precedes potential blockbuster IPOs from AI startups like Anthropic PBC and OpenAI, which could value them in the trillions and arrive as early as this year.
Year-to-date, memory-related tech companies constitute four of the top-performing stocks within the S&P 500 index benchmark.
Investors are collectively betting that robust AI computing infrastructure-related demand represents a more enduring growth story for these stocks, which have traditionally been viewed as cyclical, with demand and growth rising and falling with PC and smartphone cycles.
US NAND memory chip leader SanDisk's stock has skyrocketed 709% year-to-date, making it the S&P 500's best performer by a wide margin.
HDD storage leader Western Digital's stock has surged 274%, while another HDD leader, Seagate, has seen its stock rise 261% this year.
US-based DRAM/NAND memory chip giant Micron Technology has seen its stock price jump 267% year-to-date, with its valuation also breaching the $1 trillion mark.
However, these gains have been accompanied by intense volatility, especially recently. On Tuesday, the stock fell 13%, marking its fourth double-digit percentage move this month alone.
SK Hynix's status as a tech industry superstar is the result of one of the most dramatic success stories in the AI high-performance hardware sector.
Its early and aggressive embrace of HBM proved to be a major breakthrough. Pivoting to this technology earlier—coupled with Samsung's relatively slower response—delivered a massive windfall for SK Hynix and allowed the underdog to surpass its rival by certain metrics.
According to Counterpoint Research data, by Q4 2025, it is projected to control 57% of the global HBM market by revenue, far ahead of Samsung and Micron.
As shown in the chart, SK Hynix and Samsung trade at a significant discount relative to their US memory chip peers.
According to its regulatory filings, the company intends to use the proceeds from the fundraising to build additional production capacity and purchase cutting-edge semiconductor manufacturing equipment like extreme ultraviolet (EUV) lithography machines.
The US listing will expose the company to a new investor base and may help SK Hynix narrow its valuation gap relative to competitors.
Asian issuers already listed in the US include Taiwan-based chipmaking titan Taiwan Semiconductor Manufacturing Company (TSMC).
TSMC's US listing has enabled it to attract foreign investor capital flows, solidifying its position as a市值 leader and an AI darling—especially during the AI-driven rally.
Supply Shortages May Persist Until 2028
Whether it's Google's massive TPU AI computing clusters or NVIDIA's immense AI GPU computing clusters, all require fully integrated HBM memory systems paired with AI chips, alongside the massive procurement of server-grade DDR5 memory and enterprise high-performance SSDs/HDDs by tech giants accelerating the construction or expansion of AI data centers.
Samsung Electronics, SK Hynix, and Micron Technology are precisely positioned across these three core memory segments: HBM, server high-performance DRAM (including DDR5/LPDDR5X), and high-end data center SSDs, making them direct beneficiaries within the "AI memory + storage stack" and prime recipients of the "super红利" from the AI infrastructure wave.
GPUs generate intelligence, HBM/DRAM feeds data at high speed, enterprise NAND/eSSD handles hot data and caching, while HDDs are for the long-term retention of massive cold/warm data.
Therefore, Wall Street giants like Goldman Sachs believe the AI computing arms race led by cloud giants is transforming memory chips from cyclical commodities into scarce strategic assets, with price increases for DRAM/NAND in 2026 not being the end but potentially the initial phase of a super cycle.
A recent report from Jefferies indicates that AI computing-related demand continues to push the global memory chip market size higher, with the memory chip price uptrend expected to persist into the second half of 2026 and potentially throughout 2027.
The report estimates global memory chip prices could rise approximately 40%-50% quarter-over-quarter in Q3 2026, with a possible further 30%-40% increase in Q4, marking two consecutive quarters of significant price jumps.
Regarding an inflection point, the Jefferies study suggests signs of moderation may not appear until 2028, when, with increased capacity and higher market supply, average selling prices for memory chips could decline by 15% to 20%.
A recent study from semiconductor industry research firm SemiAnalysis shows that although China's DRAM memory giant CXMT is expected to maintain one of the world's fastest capacity expansion rates in the coming years, SemiAnalysis does not believe this will alter the overarching trend of tight DRAM supply.
Conversely, the firm believes AI servers, high-performance computing, and cloud computing infrastructure construction continue to drive memory demand higher, while the pace of new capacity coming online still struggles to keep up with demand growth.
According to SemiAnalysis calculations, even factoring in the expansion plans of major players like CXMT, Samsung, SK Hynix, and Micron, the global DRAM market in 2026 will still face a high-single-digit percentage supply gap.
By 2027, the supply-demand gap could even widen further to low-to-mid double-digit levels.
Based on this assessment, the research firm expects the tight DRAM supply situation could persist until around 2028.
In other words, even with CXMT's rapid rise and continued expansion, it is unlikely to fundamentally reverse the industry's supply shortage, which is a key rationale for SemiAnalysis's continued bullish stance on DRAM prices and the overall HBM/DRAM/NAND memory industry upcycle.
As revealed in a recent interview by Jeremy Werner, Senior Vice President and General Manager of Micron's Data Center Business Unit, the underlying driver of this cycle from a data flow processing engineering logic perspective in AI data centers is not simply "AI needs more compute chips."
Rather, the era of AI inference dominated by AI agents like Claude Cowork and OpenClaw is pushing memory/storage from a supporting component to a system bottleneck.
AI training engineering relies more on massive parallel computing, while inference—especially with long context, multi-turn conversations, and Agentic AI workflows—requires persistent saving of KV Cache, context states, and intermediate results.
When memory/storage space is insufficient, models are forced to recompute historical states, leading to decreased GPU utilization and increased token generation costs.
Therefore, HBM, DDR5, LPDDR, enterprise SSDs, and even HDDs/data lakes are forming an "AI memory chain" from GPU-proximate to distant storage, determining an AI system's throughput, latency, concurrency, and per-token economics.
This explains the synchronized surge in shares of memory and data storage companies like Micron, Samsung, SK Hynix, SanDisk, and Western Digital: demand is not concentrated solely in HBM but is spilling over across the entire chain of DRAM, NAND, SSD, and HDD along the AI server architecture.
Comments