SK hynix ADR Listing and HBM Strategy Propel AI Memory Leadership

In July 2026, SK hynix made headlines by raising a record $26.5 billion through the listing of its American Depositary Receipts (ADRs) on NASDAQ, marking the second-largest offering in US history. This move signifies a major step for the company, long recognized within Korea, onto the global stage. SK hynix's journey from a struggling semiconductor firm to an AI memory powerhouse is a testament to resilience and strategic foresight.

The Genesis of Turmoil: Post-1997 Asian Financial Crisis

SK hynix's tumultuous history began in the aftermath of the 1997 Asian financial crisis. Korea initiated a restructuring program called the "Big Deal" to stabilize highly leveraged conglomerates, encouraging consolidation of overlapping businesses. The semiconductor industry was a key target. In 1998, Hyundai and LG, then Korea's second and third-largest semiconductor companies, discussed merging their operations.

The proposed merger offered a compelling opportunity to achieve a DRAM market share comparable to Samsung and gain the necessary scale to compete with Micron and major Japanese manufacturers. In 1999, Hyundai acquired LG's semiconductor business, creating one of the world's largest memory manufacturers. However, this transaction also brought substantial financial obligations.

Hyundai Electronics, with approximately $13–14 billion in debt, faced severe challenges when the Dotcom and telecom bubbles burst in late 2000, leading to an industry-wide downturn. Analysts described it as the worst DRAM market since 1985, with prices falling below manufacturing costs. Semiconductors accounted for 80% of Hyundai Electronics' revenue. In 2000, despite an operating profit of $1.5 billion, the company recorded a net loss of $2 billion due to debt-related financial costs and valuation losses. Losses worsened in 2001 as DRAM prices plummeted further.

Hynix's Struggle for Survival

Facing $3.1 billion in debt due in 2001, the company scrambled for funds and officially separated from the Hyundai Group, rebranding as Hynix in March 2001. The name, a portmanteau of "high" and "electronics," aimed to convey advanced technology, but it did little to alleviate the company's struggles, which intensified after 9/11.

In October 2001, Hynix's creditors, led by Korea Exchange Bank, initiated a comprehensive restructuring. US firm Micron Technology then proposed acquiring Hynix's memory chip fabs. After negotiations, a non-binding Memorandum of Understanding was signed to sell Hynix's memory business for 108 million Micron shares. However, Hynix's revenue was predominantly from DRAM, and carving out this business would leave the remaining debt-heavy units unsustainable. Critics also believed Micron's intent was to eliminate a competitor by taking capacity offline rather than growing the Korean fabs.

Despite pressure from creditor banks, Hynix's board unanimously rejected Micron's offer in April 2002, citing undervaluation, unacceptable restructuring conditions, and a recovering memory market. Public and political opposition to selling a Korean company to an American rival also played a significant role.

The "Comeback Darling" and Subsequent Setbacks

Hynix resolved to overcome the crisis independently. The following years saw the company divest assets to reduce debt. In 2003, its TFT-LCD panel display business was sold to China's BOE Technology. In 2004, its system IC business was spun off to become MagnaChip. Hynix also partnered with STMicroelectronics to open a new fab in Wuxi, China. To restore its technical reputation, Hynix launched successful branding and quality initiatives like the Blue Chip, Prime Chip, and Golden Chip grading systems.

After a challenging 2002 and 2003, Hynix rebounded in 2004 and 2005, achieving record profits and a significant stock price increase. In September 2005, The New York Times lauded Hynix as a "comeback darling," calling its turnaround a "miracle."

However, this success was short-lived. In 2007, Hynix, under new CEO Kim Jong-gap, announced an aggressive 300-millimeter expansion plan for DRAM and NAND, aiming to become the world's largest memory maker by 2017. This coincided with falling DDR2 DRAM prices and the realization that Windows Vista would not trigger a super-cycle. The Global Financial Crisis further exacerbated the situation, leading to Hynix's worst quarterly loss in eight years and dangerously low cash flow. Competitors like Micron also suffered, and Infineon/Qimonda collapsed. By late 2008, Hynix, still creditor-owned, was again on life support.

SK Group's Strategic Acquisition

In 2011, Hynix's banks put a controlling stake in the company up for sale. SK Group, founded in 1953 as Sunkyoung Textiles, had diversified into petrochemicals and refining, and later into telecommunications with the acquisition of Korea Mobile Telecom in 1994, which became SK Telecom. Chairman Chey Tae-won, nephew of the founder, saw an opportunity to re-enter the semiconductor business, an industry his father had briefly ventured into.

Despite internal resistance from SK executives who saw no synergy with Hynix, Chey was convinced the market had changed. Industry consolidation, with the collapse of Qimonda and the precarious position of Elpida, suggested that surviving companies would be better positioned. He also recognized the burgeoning mobile market, driven by the iPhone and Android, as a counter-balance to the volatile PC market.

Chey conducted extensive due diligence, personally studying semiconductors and consulting industry experts, including TSMC founder Morris Chang. Chang advised him on the importance of customer satisfaction, emphasizing connection during downturns and humility during upturns. Convinced, Chey personally led the acquisition process, overcoming internal resistance. In February 2012, SK completed the acquisition of a 21% controlling stake for $3 billion, and Hynix became SK hynix.

The HBM Revolution: A Strategic Pivot

After the acquisition, Chey met with SK hynix executives to understand the company's challenges and future direction. His strategy was to transform SK hynix from a commodity memory producer into a mainstream semiconductor company with indispensable products. The lesson from the previous decade was clear: relying solely on commodity DRAM was unsustainable. SK hynix needed unique offerings and closer customer collaboration from the design stage.

The concept of High Bandwidth Memory (HBM) dates back over two decades. Bryan Black, known as the "Godfather of the Chiplet," published a paper in 2006 discussing the benefits of stacking DRAM dies. The core idea was to improve compute performance by increasing data bandwidth from memory to logic cores while managing power consumption. With growing demands from graphics and deep learning, existing GDDR5 memory standards were insufficient.

This led AMD to collaborate with SK hynix to co-produce a new industry standard architecture: HBM. HBM stacks multiple DRAM memory dies on a logic "base die," connected by Through Silicon Vias (TSVs) and microbumps. This design allows for many slower channels, providing significantly higher overall bandwidth compared to fewer, faster channels in GDDR5. The shorter TSVs also improve resistance and power consumption.

HBM debuted with AMD's Radeon R9 Fury X GPU, a complex chip integrating 22 different dies. SK hynix and AMD submitted the HBM specification to JEDEC, which adopted the standard in 2013, fostering a broad product ecosystem. Although the Fury X was a pioneer and the fastest single-chip GPU at the time, it did not sell in high volumes due to production challenges and yield issues.

Despite the initial financial uncertainty, Chairman Chey continued to support HBM development, recognizing its strategic importance and AMD's commitment. This decision proved prescient, especially as Samsung Electronics announced its own second-generation HBM (HBM2) just six months after the Fury's launch, intensifying the race.

HBM2 Challenges and the MUF Breakthrough

SK hynix showcased its first HBM2 wafers in March 2015, aiming for a 2016 release. However, the development team faced significant challenges. HBM2 aimed to double performance by increasing per-pin transfer speed and bandwidth. To achieve better performance at lower cost, SK hynix gambled on skipping the 2y nanometer node and going straight from a high 20-nanometer process (2x) to a low 20-nanometer process (2z).

In late 2015, both Samsung and SK hynix were reportedly in the final stages of HBM2 development. However, a last-minute change in product specifications forced SK hynix to pivot its design, leading to severe issues. TSV speeds plummeted to 500 megabits per second, a quarter of the target, due to high capacitance. The higher-end 2 gigabit product was scrapped, and the rushed design led to packaging problems, including cracked dies and incorrect cuts, resulting in abysmal yields of around 50%. The product failed customer qualification, and its release was delayed until 2018.

Further misfortune struck with the crypto winter of 2018, which crashed demand for mining GPUs and HBM. SK hynix's new $400 million packaging fab, P&T4, built for HBM2 production, became underutilized.

The HBM2 experience forced SK hynix to choose between cost and performance. For the third generation, HBM2E, launched in mid-2018, they prioritized performance. They doubled the number of TSVs and added thermal "dummy" bumps. The most significant change was in packaging.

Previously, SK hynix used Thermal Compression With Non-Conductive Film (TC-NCF) for bonding and underfill. This method involved laminating a non-conductive adhesive film, aligning dies, and applying high pressure and heat sequentially for each die. While offering good insulation and even layers, TC-NCF was time-consuming and prone to errors, especially with multiple dies and increased TSVs.

Recognizing the limitations of TC-NCF, SK hynix explored a new approach: mass reflow. This method involved stacking all DRAM dies first, then melting and bonding their microbumps in a single step. For underfill, they initially considered Capillary Underfill (CUF), but its slow process led them to Molded Underfill (MUF). MUF, previously used for small, single dies in mobile chip packaging, combines underfill and overmold encapsulation using a single Epoxy Molding Compound (EMC).

SK hynix embarked on developing MR-MUF (Mass Reflow-Molded Underfill), a technique that required co-developing equipment and materials from scratch with partners like Towa (molding equipment) and NAMICS (EMC material). Despite initial skepticism from customers, traumatized by the HBM2 debacle, SK hynix successfully scaled MR-MUF for HBM2E, announcing it in August 2019. Qualification tests showed exceptional bandwidth performance, leading customers to adopt SK hynix's HBM2E in their GPUs. Two years later, SK hynix announced HBM3, with 8-layer and 12-layer versions capable of 819 gigabytes per second.

The ChatGPT Boom and HBM's Vindication

Before ChatGPT, the HBM market was tiny, representing only 1.5-2% of the larger DRAM market. Many considered it an expensive, overkill product, especially after the crypto winter. Samsung, at one point, shifted its focus to HBM-PIM (Processing-in-Memory), an "edge inference" solution integrating an AI computing engine into memory, which was more suited for speech recognition or machine vision than Large Language Models.

On November 30, 2022, the day ChatGPT was released, SK hynix was the sole provider of the high-performance HBM3 critical for the suddenly in-demand Nvidia GPUs. This moment validated Chairman Chey's unwavering support for HBM development, despite its uncertain profitability. While the company faced market criticism for these choices, the ChatGPT boom proved the long-term customer-focused strategy paid off.

The semiconductor industry was in a post-COVID downturn in 2022, with falling DRAM and NAND prices. SK hynix reported a $1.5 billion operating loss, its first in a decade, followed by three more quarterly losses. However, by spring 2023, HBM demand surged due to ChatGPT's rapid growth.

SK hynix scrambled to increase HBM capacity. The P&T4 packaging fab, previously underutilized, began running at full capacity. With no time to build new fabs, SK hynix repurposed an empty cleanroom on the second floor of its Cheongju M15 fab, originally for NAND, to package HBMs within six months.

A more significant commitment was the M15X expansion. Announced in September 2022, before ChatGPT, the project was initially put on hold due to financial losses. However, with HBM demand exceeding projections by early 2024, SK hynix officially restarted the M15X project with a $4 billion investment, pivoting it to DRAM production. The first chips from this expansion are expected this year.

Global Ambitions and Future Investments

SK hynix's NASDAQ listing is a significant step towards becoming a globally impactful company. SK Group's strategic entries into energy and telecom under the previous chairman made it Korea's fifth-largest conglomerate. Chairman Chey's acquisition of SK hynix has propelled it to the second largest.

The company is already building a $4 billion advanced packaging fab in Indiana, USA, and Chairman Chey is actively participating in global AI conferences. With DRAM prices high and supply booked for the next two years, SK hynix is investing heavily. It announced a mid-to-long-term plan to invest 1.1 quadrillion won in its clusters in Cheongju and Yongin, plus a new one in Korea's southwestern region, with potential for a US cluster in the future.

SK hynix's journey exemplifies a company that has swung between crisis and success. After years of losses and being nearly written off, SK acquired it despite objections. The company persevered with HBM development, even when the market offered no immediate signs of payoff. Chairman Chey's commitment to building a company that could rival Japan's top-tier firms has been realized through the HBM revolution.