SK hynix has decided to build new semiconductor plants simultaneously in Yongin and Cheongju. At a board meeting on the 7th, the company approved investments of 35.2 trillion won for the Yongin “Y2” fab and 19.1 trillion won for the Cheongju “M17” fab, bringing the total to about 54 trillion won.
This is the first major execution phase of the company’s mid- to long-term investment strategy announced in June, which calls for 600 trillion won for the Yongin semiconductor cluster and 100 trillion won for the Cheongju production base.
Y2 is the second of four fabs to be built sequentially in the Yongin cluster. It will be a DRAM production hub covering 341,000 pyeong, producing next-generation DRAM including high-bandwidth memory (HBM). Construction is scheduled to begin in July next year, with the first clean room set to open in June 2029.
The investment will be executed through October 2031. The company has set a goal of completing all four fabs by 2033, 12 years earlier than the original completion target of 2045, and Y2 is the second stage supporting that timeline.
M17 in Cheongju is a NAND-only fab. Covering 206,000 pyeong, construction will begin in February next year, and the first clean room is scheduled to become operational in December 2028. The Cheongju campus already produces NAND at M11, M12, and M15, making integration with existing fabs easier, and much of the land, power, and water infrastructure is already secured. The company says it chose the site that can be built the fastest.
This decision shows that competition in the memory industry has expanded beyond technology to production capacity. SK hynix itself said that in the AI era, technological competitiveness alone is not enough, and that the ability to supply the amount customers need at the time they need it is what constitutes competitiveness.
Having secured an edge in HBM technology, the company now aims to solidify its lead in volume supply as well.
Demand forecasts are the basis for the investment. Market research firm Omdia expects demand for DRAM and NAND to grow at an annual average rate of 19% in both markets from last year through 2030. The World Semiconductor Trade Statistics organization forecasts that the global semiconductor market will exceed $1.5 trillion this year, with the memory market expanding significantly year on year.
The assumption behind the 54 trillion won investment is that the memory industry, which has typically swung between booms and busts every one to two years, is entering a structural growth phase driven by AI infrastructure investment.
NAND’s status is also changing. Until now, the AI boom has largely benefited DRAM and HBM. But as AI services spread, enterprise SSD demand is rising quickly, and with added demand for KV cache storage during AI inference, NAND is emerging as a new growth pillar. The 19 trillion won investment in M17 signals that SK hynix sees NAND as the second major battleground in the AI era.
Opinions remain divided on demand. SK Group Chairman Chey Tae-won said in an interview with a U.S. media outlet that even if production capacity were doubled within the next five years, customers would still say it is not enough, suggesting that the supply shortage will continue.
Micron’s stronger-than-expected earnings and its announcement that most of next year’s HBM4 supply volume has already been contracted also support the optimistic view.
However, counterarguments are equally strong. Some global investment banks have raised concerns that AI infrastructure investment and HBM demand could have already peaked, and as a result Samsung Electronics and SK hynix shares have seen corrections.
The fact that capacity expansion is happening simultaneously at multiple companies is an even bigger variable. Samsung Electronics is making large-scale investments in Pyeongtaek P5 and in Yongin’s national industrial complex, bringing forward the start of its first Yongin fab to 2029, while Micron is also expanding its production capacity.
That is why there are warnings that if new supply comes flooding in all at once in 2028 to 2029, oversupply could return. The challenge from China’s CXMT, whose second-quarter revenue rose more than sevenfold from a year earlier, is also seen as a potential threat to the commodity DRAM market.
The center of gravity in the competition is shifting from the speed of building fabs to the ability to fill the fabs once built. In the industry, five-year long-term supply contracts are increasingly viewed as the new standard for memory transactions.
Micron has signed 16 multi-year contracts that set price floors and ceilings in advance, and if all of them are reflected, more than half of its revenue is expected to come from contract-based sales. The gap between companies that build capacity on top of secured demand and those that expand based on market conditions will become clear in the next cycle.
From a domestic perspective, it is important that semiconductor investment in Korea is now split between Yongin and Cheongju. Since fab construction and startup will continue in both regions through the late 2020s, power, water, and labor supply have become national infrastructure issues rather than matters for individual companies alone.
First, investment execution must be supported by contractual flexibility. SK hynix plans to proceed with building the facilities and clean rooms on schedule while adjusting equipment installation according to demand. For this strategy to work, demand visibility must be secured.
Expanding the preorder-based contract model proven in HBM to NAND products such as enterprise SSDs, and increasing contract structures that limit price fluctuations, would be a practical safeguard against oversupply by locking in a significant portion of M17’s output before startup.
The task for the government and local authorities is to build infrastructure in advance. The first-stage power and water progress rate needed for Y2 operations has already reached 99%, but to meet the 2033 completion target, the second-stage infrastructure needed for the third and fourth fabs must be secured before those fabs begin construction.
Unless approval times for transmission network construction are shortened, power grid reinforcement in the Cheongju area is advanced, and water supply plans are finalized early, the 12-year-accelerated schedule will be difficult to keep.
Labor supply must be prepared on the same timeline. When the two fabs begin operation in 2028 and 2029, demand for process and equipment specialists will surge.
The size of semiconductor department programs and local university training should be expanded in line with the startup schedule, and housing, transportation, and educational conditions in Cheoin-gu, Yongin, and around Cheongju should be improved so that workers can stay in the area. Building a supporting urban base is essential.
If materials, parts, and equipment suppliers are encouraged to invest in the two regions as well and used as testing grounds for the domestic supply chain, the 54 trillion won investment could become a catalyst not just for one company’s expansion but for strengthening the entire Korean semiconductor ecosystem.