Inside SK Hynix: We Went To Korea To See The World's Biggest AI Memory Buildout
SK Hynix, South Korea's memory chip giant, is undertaking a historic $720 billion expansion to triple HBM (high-bandwidth memory) capacity by 2034, driven by unprecedented AI demand. The company is building massive new facilities in Korea while establishing its first U.S. packaging fab in Indiana, though it faces traditional boom-bust cycle risks and competition from Samsung, Micron, and Chinese manufacturers.
Summary
SK Hynix, a South Korean semiconductor company, is in the midst of an unprecedented capital investment cycle driven by extreme global demand for high-bandwidth memory (HBM) chips essential for AI computing. The company's July 2024 Nasdaq debut at $26 billion was the largest foreign listing ever, reflecting massive investor appetite. HBM production has become so critical that major tech companies like Nvidia, Apple, Microsoft, and Meta are competing aggressively for supply and signing unprecedented five-year contracts worth billions of dollars, shifting memory from a commodity business to a premium, custom-designed product category.
SK Hynix's dominance in HBM stems from its pioneering role in creating the first HBM chip in 2013 and its proprietary MR-MUF packaging technology developed around 2019, which enables 40% thinner, higher-performance stacks. The company is executing its most ambitious buildout in history, including the massive Yongin cluster project covering 45 million square feet (600 soccer fields) with four new fabrication plants costing around $400 billion, with the first fab scheduled to open in February 2025. This expansion is accelerated by customer pressure, with the company targeting completion of the fourth fab by 2033, 12 years ahead of original schedules.
The shortage of HBM has created significant economic impacts: smartphone DRAM prices have surged over 80%, Apple raised iPad and MacBook prices due to chip costs, and Amazon increased its 2026 CapEx from $200 billion to $220 billion. The scarcity has also spurred alternative approaches from new entrants like Cerebras, Groq, and D-Matrix that reduce reliance on scarce DRAM memory, and Nvidia is testing lower-memory GPU versions. SK Hynix's $720 billion investment is more than 18 times the $39 billion the U.S. government allocated for all chip manufacturers through the Chips Act.
While company leadership argues the AI era represents a structural change that will sustain demand until at least 2030, the memory industry historically cycles through boom and bust periods. SK Hynix experienced an operating loss of nearly $6 billion during the 2023 downturn. The company is building facilities internationally: its first U.S. facility in Indiana ($4 billion investment for packaging, not front-end production) with incentives totaling over $1 billion, and three fabs in China, though U.S. export controls prevent advanced HBM sales to China, where emerging domestic manufacturers are rapidly catching up. SK Hynix faces competitive pressure from Samsung and U.S.-based Micron, which is investing $250 billion in massive memory fabs in Idaho and New York.
Key Insights
- SK Hynix argues that this AI cycle is structurally different from historical memory boom-bust cycles because AI applications require vastly more memory and the number of computers needed for AI is limitless, suggesting demand will outstrip supply until at least 2030.
- Memory contracts have fundamentally shifted from one-year agreements to five-year contracts with prepayments of $10 billion rather than $10,000, representing a massive commitment and risk-sharing shift that breaks from historical commodity pricing models.
- SK Hynix's proprietary MR-MUF packaging technology, developed around 2019 when the company lost HBM leadership to Samsung, enables 40% thinner stacks through simultaneous bonding of multiple DRAM layers with protective liquid injection, providing a competitive cost manufacturing advantage.
- The memory shortage has driven secondary effects across the tech industry: smartphone DRAM prices surged 80%, Apple raised iPad and MacBook prices, Amazon increased 2026 CapEx by $20 billion, and Nvidia is testing lower-memory GPU versions to reduce HBM reliance.
- Building HBM manufacturing in the U.S. costs approximately double that of Korea due to expensive land, electricity, water, and regulations, making U.S. front-end production economically challenging despite government incentives, which is why SK Hynix's U.S. facility focuses only on packaging rather than chip production.
Topics
Transcript
[0:03] Memory is in the midst of an unprecedented moment, and it's led us to Korea to see this, the largest memory build out in the world by SK Hynix, which makes the majority of the world's high-bandwidth memory chips for AI inside their clean rooms where no media has ever set foot until now. In this place, HBM is assembled. This secretive Korean memory king is also the hottest new entrant to the U.S. Market, following a blockbuster Nasdaq debut in July. It was the biggest foreign listing on the Nasdaq ever [0:35] at $26 billion, and the appetite was insane. Demand for these tiny stacks of memory to enable AI has created such scarcity that the world's most…
Full transcript available for MurmurCast members
Sign Up to AccessMore from CNBC
Inside China’s Race To Build The World’s Data Centers
China's prefabricated data center manufacturing offers a significant speed advantage over U.S. construction, potentially reducing build times from 2-3 years to 9 months, while the U.S. becomes increasingly dependent on Chinese supply chains despite geopolitical tensions and national security concerns.
How Williams-Sonoma Became One Of Retail’s Best-Performing Stocks
Williams-Sonoma has outperformed the S&P 500 and competitors by leveraging its omnichannel model, focusing on full-price selling to affluent customers, and deploying AI to enhance e-commerce and supply chain efficiency. Despite a sluggish housing market, the company has achieved margin expansion and is pursuing growth opportunities through smaller home furnishings, B2B expansion, and renewed store openings.
Why It’s So Hard To Build An AI Kill Switch
The video explores whether an AI kill switch is technically feasible and practically implementable. While experts acknowledge the need for safety mechanisms, they conclude that a single universal kill switch is impossible due to distributed computing infrastructure, and that comprehensive regulation similar to other industries would be more effective than a technological quick fix.
America’s AI Chip Talent Crisis — Why The U.S. Needs 157,000 More Workers
The U.S. faces a critical shortage of 157,000 semiconductor workers by the end of the decade as major chip manufacturers build advanced fabs domestically. Universities like Purdue and Arizona State are launching hands-on semiconductor programs, while companies are recruiting aggressively and investing heavily in workforce development to close the talent gap that currently exists compared to Asia.
Exclusive look inside SK Hynix's $720 billion AI memory expansion
SK Hynix, the world's leading maker of high-bandwidth memory (HBM) chips for AI, is investing $720 billion to triple production capacity by 2034 in response to unprecedented demand from major tech companies. The company is expanding aggressively across South Korea and into the US, but faces the classical risk of the memory industry's boom-and-bust cycles.