目录
Interview Overview
I. Interview Participants
II. Memory Is Critical Infrastructure for the AI Era
III. The 2012 Hynix Acquisition: One of the Most Successful High-Risk Investments of His Life
IV. Memory Capacity May Need to Reach 5 Times Today’s Level Over the Next 10 Years
V. Nvidia, TSMC, and the AI Supply Chain
VI. Nvidia Is Currently the Most Important Customer
VII. New Capacity Takes at Least Four to Five Years
VIII. “Chip Inflation” Is Spreading to Consumer Electronics and Autos
IX. Technological Evolution and Reducing AI Token Costs
X. Capacity Expansion Must Be Backed by Real Demand and Long-Term Agreements
11. HBM Requires More Than Four Times the Wafer Area of Conventional DRAM
12. Customers Are Proactively Seeking Long-Term Supply Agreements
13. Next Year May Mark the Most Severe Point of This Memory Shortage
14. How AI Differs from Previous Industry Cycles
15. A New Memory Fab Requires a Complete Industry Ecosystem
16. A Full-Scale US Memory Fab Remains Under Review
17. Competitive Advantages over Samsung and Micron
本内容基于公开资料和研报数据整理，不构成任何投资建议，不代表任何个人观点，仅供学习参考，请理性阅读
Interview Overview
The interview centers on Chey Tae-won’s assessment of memory demand in the AI era, SK hynix’s responsibility to expand capacity, and the ripple effects of supply shortages across the global technology industry. Chey compares today’s AI to a 4-year-old child: as models acquire more experience, knowledge, and capabilities, they must retain more information. Memory is therefore not an ordinary component but essential infrastructure for AI’s continued evolution. SK Group’s energy, telecommunications, and data-center businesses complement its memory operations, although SK hynix has become its most important subsidiary. Chey also revisits the 2012 acquisition of Hynix. While he did not anticipate today’s AI boom, he believed the digital economy would generate enormous volumes of data. In hindsight, the deal was one of the largest and most successful high-risk investments of his life. As SK hynix embarks on another major expansion, Chey expects AI-agent adoption to accelerate rapidly. The capacity required to meet global memory demand over the next 10 years could reach 5 times today’s level, while new capacity typically takes 4 to 5 years to build. Supply therefore cannot catch up quickly, and he believes next year could mark the most severe point of the current memory shortage.
The interview also highlights the interdependence of the AI supply chain. Without TSMC’s advanced manufacturing and base dies, GPU output cannot scale; without sufficient HBM, additional GPU capacity cannot be converted into complete AI systems. Nvidia is currently SK hynix’s most important customer, but hyperscalers such as Google and Microsoft are also developing custom chips. Chey acknowledges that near-term demand is surging, with customers asking for nearly double their existing supply allocations. Rapidly rising memory prices have created what he calls “chip inflation,” which is spreading to personal computers, smartphones, automobiles, and consumer electronics. Capacity expansion is unavoidable, but HBM consumes far more wafer area than conventional DRAM, further intensifying capacity constraints. To avoid repeating past cycles of overinvestment and collapse, SK hynix will make expansion contingent on verified demand and long-term agreements, while exploring risk-sharing structures such as fab joint ventures and “memory as a service.” Chey offers no firm commitment to building a fully integrated memory fab in the United States, emphasizing that site selection depends on more than land, water, and electricity: it must also support an ecosystem of 600 to 700 materials, chemicals, and service suppliers.
I. Interview Participants
Chey Tae-won: Chairman of SK Group
Interviewer: CNBC reporter
II. Memory Is Critical Infrastructure for the AI Era
Interviewer: SK hynix is a major Korean company, but you oversee more than this one business: SK Group has over 100 subsidiaries. How important is memory within the broader group?
Chey Tae-won: We are now in the AI era. Humans need a great deal of memory, and so does AI.
I see today’s AI as a 4-year-old child. As it grows, it will acquire more capabilities, experience, and knowledge, all of which must be stored somewhere—just as human memories are.
A child’s brain is still small and holds relatively few memories. By adulthood, however, a person has accumulated vast experience, memories, and wisdom. I believe AI will develop in much the same way. It will therefore require enormous amounts of memory and memory chips.
SK hynix is now the largest of SK Group’s subsidiaries, which is unsurprising. But AI also requires energy and networks. From that perspective, SK Group’s portfolio includes energy and network businesses—what we now call telecommunications companies—all of which are working to build AI data centers.
Our portfolio is therefore complementary to some extent. But you are right: SK hynix is currently the most important part of the group.
III. The 2012 Hynix Acquisition: One of the Most Successful High-Risk Investments of His Life
Interviewer: SK Group was not originally a semiconductor company; it began in textiles. I visited the company museum yesterday. When and how did SK decide to enter semiconductors? Was that your decision?
Chey Tae-won: Yes. That was in 2012.
At the time, I believed SK needed to add new business pillars. We were already active in energy, chemicals, and telecommunications, but I wanted the group to have more globally oriented businesses.
The digital era had already begun. Although I did not foresee today’s AI wave, I believed the digital economy would require massive amounts of memory. Vast quantities of knowledge and data would have to be stored somewhere, making memory chips indispensable.
I therefore began looking at companies in the sector and concluded that this could be the future. It was a high-risk bet. Judged by the outcome today, it has been one of the largest and most successful investments of my life.
IV. Memory Capacity May Need to Reach 5 Times Today’s Level Over the Next 10 Years
Interviewer: You are now making another enormous investment, with plans to spend $720 billion on capacity expansion. How do you demonstrate that the business is sustainable enough over the long term to justify an investment of this magnitude?
Chey Tae-won: Demand forecasts suggest that people are likely to use many more AI agents in the future. AI agents will become an important part of the technology landscape.
Compared with 2025, I believe AI-agent usage could grow 10-fold over the next five years and spread worldwide. That will also require a great deal of memory.
