SK Hynix commits to $720 billion investment to expand AI memory (HBM) production capacity significantly.

In the heart of South Korea, SK Hynix is undertaking its most ambitious expansion yet. The company is investing a staggering $720 billion to triple its memory production capacity by 2034, a monumental bet fueled by insatiable demand for High Bandwidth Memory (HBM) chips essential for powering the AI revolution. CNBC was granted unprecedented access inside SK Hynix's clean rooms, offering rare insight into the manufacturing of these critical components.

**The AI Memory Arms Race**

The demand for HBM—specialized memory that enables rapid data access for AI processors—has created a global scarcity. Leading tech giants like Nvidia and Apple are now locked in a scramble to secure long-term contracts with memory manufacturers like SK Hynix. What was once considered a low-margin commodity has transformed into a high-stakes battleground, with memory chips becoming a bottleneck for AI computing power.

SK Group Chairman Chey Tae-won highlighted the urgency: "Everybody wants the memory chips. Without that, they cannot produce their AI computing and AI chips." This sentiment underscores the critical role memory plays in the current AI landscape.

**From Textiles to Semiconductors**

SK Hynix's journey to the forefront of AI memory reflects strategic, albeit risky, bets. The company began as Sunkyong Textile Company and diversified over decades, navigating near-bankruptcy during the 1997 Asian financial crisis. Under Chairman Chey Tae-won, the group pivoted toward semiconductors. A pivotal moment came in 2012 when SK Group acquired Hynix Semiconductor for $3 billion, a product of Hyundai and LG's chip divisions. The following year, SK Hynix unveiled the world's first High Bandwidth Memory.

Myeong-jae Park, who now leads HBM design at SK Hynix, recalled the early days: "At that moment, there was no application or devices that needed that. But we had strong conviction." This conviction paid off. By 2015, SK Hynix sold the first commercial HBM on AMD's Radeon graphics cards. The true inflection point arrived with the AI boom sparked by ChatGPT in 2022, when SK Hynix was first to market with HBM3, prompting Nvidia to switch its primary supplier.

**The Yongin Cluster**

The scale of SK Hynix's current expansion is unprecedented. The company is constructing the Yongin cluster, a sprawling complex encompassing 45 million square feet—roughly equivalent to 600 soccer fields. This facility will house four new fabs, including the first with six clean rooms, and is scheduled to begin operations in February 2025. The stacked design was necessitated by South Korea's mountainous terrain. The Korean government is supporting this massive undertaking with a chip support package valued at up to $22 billion, assisting with infrastructure needs like land flattening. This initiative aligns with South Korea's goal to double the country's memory production capacity within five years. The estimated cost of the project is $400 billion.

**The Cyclical Trap**

Despite the current demand, the memory industry is notorious for its cyclical nature. MS Hwang, Research Director at Counterpoint Research, cautioned: "We cannot avoid a cycle. It's coming, especially against China." The memory market has historically experienced dramatic swings, and a downturn in 2023 saw SK Hynix incur nearly $6 billion in operating losses due to oversupply.

However, SK Hynix believes the AI era marks a structural change, potentially altering the traditional cycle. The company is accelerating construction timelines, targeting completion of its fourth fab 12 years ahead of schedule. The massive investments are underpinned by long-term contracts, some extending up to five years, with significant prepayments, indicating shared commitment and risk-taking among major tech players to secure future supply.

**Proprietary Advantage**

At the heart of SK Hynix's dominance is its proprietary packaging technology, MR-MUF (Mass Reflow Molded Underfill), developed around 2019 to regain HBM leadership through a low-pressure, quick manufacturing process.