The Bank of Korea supports the chip super cycle. Why is SK Hynix sold off?
The Bank of Korea ended up singing a multi-chip super cycle, but it failed to prevent the stock prices of South Korea's "storage duo" from falling. On July 13, the Bank of Korea issued a report stating that the global semiconductor market is still in short supply and the current super cycle driven by artificial intelligence is expected to continue for some time, trying to dispel investors' concerns that the chip cycle has peaked. The Bank of Korea stated in the report that investment in AI infrastructure has significantly boosted semiconductor demand, but the pace of supply expansion is still slow, so the semiconductor industry cycle has not yet shown signs of slowing down. Moreover, this chip cycle is different from the past. Major companies are competing to invest, and these investments are based on the expectation that the popularization of AI will trigger fundamental changes in the industrial ecosystem. Although the Bank of Korea has just given reassurance to the market, investors' concerns have not dissipated, including companies borrowing money to promote AI infrastructure, industry valuations are already at high levels, and the inflection point of memory chip surplus seems to be around the corner. The concentrated outbreak of worries has put global technology stocks under pressure recently. On July 13, the South Korea Composite Index closed at 6806.93 points, down 8.95%. The share prices of South Korea's two major storage giants, SK Hynix and Samsung Electronics, also plummeted. Among them, SK Hynix closed down 15.37%, setting a record for the largest decline. The chip market is still in short supply. Why are chip stocks bucking the trend and diving? Although SK Hynix hopes to tell the memory story into a "customized solution" through HBM, the agency warned that memory is a cyclical commodity after all. A core question emerges from this: In the AI era, can memory chips really break free of their cyclical nature? In early July, it was revealed that Metaverse Corporation (Meta) planned to sell excess AI computing power to external customers. This news quickly triggered a re-evaluation of the supply and demand relationship in AI infrastructure. At a time when the outside world is generally worried about excess computing power and questioning whether the investment logic of AI hardware is broken, the Bank of Korea tried to dispel market concerns through a report. The Bank of Korea pointed out in the report that the core driving force of this round of chip cycle is not the traditional fluctuation of supply and demand, but the increase in investment by major companies in order to seize the opportunity, just because the industry judges that the popularization of AI will rewrite the industrial ecological pattern. More importantly, the current market is dominated by customized chip products such as high-bandwidth memory (HBM), and there are more constraints on capacity expansion than before. The growth and expansion trend of the global semiconductor market is expected to continue for a long time. In its report, the Bank of Korea did not avoid the multiple uncertainties faced by the development of the AI industry, including the pace of technology popularization, the breadth of application coverage, and the prospects for commercialization profitability. Despite this, the bank still maintains the same judgment as JPMorgan Chase, Goldman Sachs and other major investment banks that the high prosperity of the global semiconductor market will continue at least until next year. Coincidentally, on the first day of listing on Nasdaq, SK Hynix CEO Guo Luzheng made a heavy prediction: the global storage industry is likely to face the most severe supply gap in history in 2027. He said that although the company is actively expanding production capacity, he expects that memory chip demand will continue to exceed the company's production capacity and continue beyond 2030. SK Hynix’s prospectus for listing in the United States has a clear expansion plan. The prospectus document mentioned that the company will expand the new wafer factory "M15X" dust-free workshop in Cheongju City, South Korea in October 2025. The M15X factory has started wafer feeding in the first quarter of 2026. The company also plans to build an industrial park in Yongin City, South Korea, and build an advanced packaging factory "P&T7" in Cheongju City. In December 2024, the company also built a new advanced packaging factory in Indiana, USA, which is expected to be put into production in the second half of 2028. Memory chip giant Micron Technology is also accelerating its wafer fab and technology investment plans in the United States, and has increased its estimated expenditure by 2035 from the previously planned US$200 billion to more than US$250 billion, including steadily advancing the construction of two wafer fabs in Idaho, USA, and launching mass production of DDR4 storage processes at the Manassas wafer fab in Virginia.
Storage giants are increasing their production capacity. Can next year's "core shortage" be resolved? Li Huihui, a professor of management practice at Lyon Business School in France, told the 21st Century Business Herald reporter that production expansion cannot be judged solely by the amount of investment, and it is not appropriate to equate the completion of the factory with salable production capacity. From the entry of EUV equipment, process introduction, first wafer off the production line, to yield ramping, packaging testing and customer certification, the entire process is connected in series. It usually takes two to three years for a greenfield factory to achieve mass production. At present, the relatively rapid increase mainly comes from tapping the production capacity potential of existing factories, such as the SK Hynix M15X project. As for Micron, its Singapore HBM packaging project and ID1 factory in Idaho are expected to contribute output one after another from the first half to the middle of 2027; SK Hynix's new Yongin factory will have to wait until the second half of 2027 to complete construction, and Samsung's Pyeongtaek P5 factory will be put into production even later, expected after 2028. Li Huihui said: "From this point of view, significant relief on the memory chip supply side may have to wait until 2028, and the excess risk window may appear from the second half of 2028 to 2029." Tanya, deputy director of the BRICS Research Center at the University of International Business and Economics, told the 21st Century Business Herald reporter that the production cycle of AI memory chips is about two years, and 2027 is the expected release node. If additional investment is made this year, production capacity will be further expanded in 2028. Although the expansion of AI infrastructure has indeed generated strong demand for memory chips, its overall trend is affected by the supply side. Currently, various economies are accelerating the construction of independent supply capabilities for their own AI industry chains. This variable is also worthy of attention. Some interviewed experts believe that the subsequent supply shortage in the chip market may only exist for specific memory chip products. Li Huihui believes that the so-called "core shortage" should not be simply understood as a general shortage of all chips. The most scarce areas next year may be concentrated in HBM, server DDR5, high-capacity RDIMM and qualified advanced packaging production capacity. Xiang Ligang, chairman of the Information Consumption Alliance, also told the 21st Century Business Herald reporter that the Bank of Korea and semiconductor companies believe that the chip market will be in short supply, mostly to maintain the stock prices of memory chip companies. "Except for the high-speed chip HBM and other advanced packaging production capacity required for artificial intelligence that are in short supply to a certain extent, there may not be a substantial supply shortage for most traditional chips." On July 13, the collective plunge in the share prices of Samsung Electronics and SK Hynix in the Korean market attracted market attention. On that day, Samsung Electronics closed down 10.7%, while SK Hynix closed down 15.37%, setting a record for the largest decline, with its total market value falling below US$900 billion. Last week, SK Hynix had just completed its debut on the Nasdaq in the United States with a 12.8% increase, and then suffered a plunge in the Korean stock market. Market analysis attributed the reason to investors taking profits and SK Hynix's operating profit forecast for the second quarter of this year to be slightly lower than market expectations. "This is not a sudden reversal of the logic of memory chip shortage, but a misalignment between industrial prosperity and stock price expectations." Li Huihui analyzed that in the year before SK Hynix went public in the United States, SK Hynix Korea's stock price had risen by about 630 yuan. %, and then the stock price fell, which was not only due to the realization of positive factors such as HBM shortage and listing in the United States, but also related to the failure of HBM4 shipments in the second quarter as expected, and the company's high exposure to HBM, which less benefited from the increase in general DRAM prices. It is worth noting that the second quarter performance forecast report of SK Hynix released by Korea Investment Securities (KIS), a local securities firm in South Korea, is regarded as the trigger for the sharp decline in Hynix’s stock price. The bank predicts that SK Hynix's second-quarter revenue will be 80.9 trillion won and operating profit will be 60.4 trillion won, a month-on-month increase of 54% and 61% respectively, and a year-on-year increase of 264% and 556%. Although KIS's forecast for SK Hynix's second-quarter performance is quite eye-catching, the bank's operating profit forecast is about 8% lower than the market consensus forecast of 65 trillion won, triggering investor concerns.
Misfortunes are where blessings depend, and blessings are where misfortunes lie. SK hynix's listing in the United States has achieved unparalleled success, but the other side of the spotlight is the market's more severe scrutiny. Li Huihui believes that operating profit is about 8% lower than market expectations does not mean that demand is weakening. The root cause is the lag in HBM contract prices. HBM usually adopts annual or three- to five-year long-term agreement (LTA) pricing and cannot keep pace with general memory chips. In the first quarter of this year, HBM’s revenue and profit per wafer were no longer comparable to ordinary memory modules such as 64GB DDR5 RDIMM. This creates an embarrassing situation: SK Hynix uses more of its scarce wafer capacity to produce HBM, but the money earned per wafer in the short term lags behind that of general-purpose DRAM. Li Huihui said that the future direction of SK Hynix’s stock price will also depend on two factors. The first is whether HBM4 in the third quarter of this year can increase its volume as scheduled; the second is the re-pricing of the long-term agreement (LTA) contract in 2027. If the return on wafers still cannot surpass DDR5 by then, SK Hynix’s financing in the US stock market will be difficult to increase its return on investment, and its US stock premium will also shrink. Historically, the storage industry has hardly been able to escape the boom cycle, which ultimately resulted in oversupply due to production expansion. Morgan Stanley and major investment banks have also warned that memory remains an inherently cyclical commodity and that the "service provider" premium only exists when supply is in short supply. Now, as more and more wafer fabs are built, will this round of boom driven by AI repeat the traditional cycle of "oversupply and falling prices"? Tanya believes that in the AI era, the cyclicality of memory chips will still exist, but its amplitude will be gentler than the traditional fluctuation cycle. The reason is that the demand side has locked in demand and price in advance through long-term agreements, which can smooth the violent fluctuations of the cycle to a certain extent. Li Huihui also said that the current market's enthusiastic pursuit of memory chips is not about buying the chips themselves, but about buying "the right to use priority in future production capacity." Therefore, general storage is still dominated by cycles, and a security premium has been added to the pricing logic of high-end DDR and HBM.
Storage giants are increasing their production capacity. Can next year's "core shortage" be resolved? Li Huihui, a professor of management practice at Lyon Business School in France, told the 21st Century Business Herald reporter that production expansion cannot be judged solely by the amount of investment, and it is not appropriate to equate the completion of the factory with salable production capacity. From the entry of EUV equipment, process introduction, first wafer off the production line, to yield ramping, packaging testing and customer certification, the entire process is connected in series. It usually takes two to three years for a greenfield factory to achieve mass production. At present, the relatively rapid increase mainly comes from tapping the production capacity potential of existing factories, such as the SK Hynix M15X project. As for Micron, its Singapore HBM packaging project and ID1 factory in Idaho are expected to contribute output one after another from the first half to the middle of 2027; SK Hynix's new Yongin factory will have to wait until the second half of 2027 to complete construction, and Samsung's Pyeongtaek P5 factory will be put into production even later, expected after 2028. Li Huihui said: "From this point of view, significant relief on the memory chip supply side may have to wait until 2028, and the excess risk window may appear from the second half of 2028 to 2029." Tanya, deputy director of the BRICS Research Center at the University of International Business and Economics, told the 21st Century Business Herald reporter that the production cycle of AI memory chips is about two years, and 2027 is the expected release node. If additional investment is made this year, production capacity will be further expanded in 2028. Although the expansion of AI infrastructure has indeed generated strong demand for memory chips, its overall trend is affected by the supply side. Currently, various economies are accelerating the construction of independent supply capabilities for their own AI industry chains. This variable is also worthy of attention. Some interviewed experts believe that the subsequent supply shortage in the chip market may only exist for specific memory chip products. Li Huihui believes that the so-called "core shortage" should not be simply understood as a general shortage of all chips. The most scarce areas next year may be concentrated in HBM, server DDR5, high-capacity RDIMM and qualified advanced packaging production capacity. Xiang Ligang, chairman of the Information Consumption Alliance, also told the 21st Century Business Herald reporter that the Bank of Korea and semiconductor companies believe that the chip market will be in short supply, mostly to maintain the stock prices of memory chip companies. "Except for the high-speed chip HBM and other advanced packaging production capacity required for artificial intelligence that are in short supply to a certain extent, there may not be a substantial supply shortage for most traditional chips." On July 13, the collective plunge in the share prices of Samsung Electronics and SK Hynix in the Korean market attracted market attention. On that day, Samsung Electronics closed down 10.7%, while SK Hynix closed down 15.37%, setting a record for the largest decline, with its total market value falling below US$900 billion. Last week, SK Hynix had just completed its debut on the Nasdaq in the United States with a 12.8% increase, and then suffered a plunge in the Korean stock market. Market analysis attributed the reason to investors taking profits and SK Hynix's operating profit forecast for the second quarter of this year to be slightly lower than market expectations. "This is not a sudden reversal of the logic of memory chip shortage, but a misalignment between industrial prosperity and stock price expectations." Li Huihui analyzed that in the year before SK Hynix went public in the United States, SK Hynix Korea's stock price had risen by about 630 yuan. %, and then the stock price fell, which was not only due to the realization of positive factors such as HBM shortage and listing in the United States, but also related to the failure of HBM4 shipments in the second quarter as expected, and the company's high exposure to HBM, which less benefited from the increase in general DRAM prices. It is worth noting that the second quarter performance forecast report of SK Hynix released by Korea Investment Securities (KIS), a local securities firm in South Korea, is regarded as the trigger for the sharp decline in Hynix’s stock price. The bank predicts that SK Hynix's second-quarter revenue will be 80.9 trillion won and operating profit will be 60.4 trillion won, a month-on-month increase of 54% and 61% respectively, and a year-on-year increase of 264% and 556%. Although KIS's forecast for SK Hynix's second-quarter performance is quite eye-catching, the bank's operating profit forecast is about 8% lower than the market consensus forecast of 65 trillion won, triggering investor concerns.
Misfortunes are where blessings depend, and blessings are where misfortunes lie. SK hynix's listing in the United States has achieved unparalleled success, but the other side of the spotlight is the market's more severe scrutiny. Li Huihui believes that operating profit is about 8% lower than market expectations does not mean that demand is weakening. The root cause is the lag in HBM contract prices. HBM usually adopts annual or three- to five-year long-term agreement (LTA) pricing and cannot keep pace with general memory chips. In the first quarter of this year, HBM’s revenue and profit per wafer were no longer comparable to ordinary memory modules such as 64GB DDR5 RDIMM. This creates an embarrassing situation: SK Hynix uses more of its scarce wafer capacity to produce HBM, but the money earned per wafer in the short term lags behind that of general-purpose DRAM. Li Huihui said that the future direction of SK Hynix’s stock price will also depend on two factors. The first is whether HBM4 in the third quarter of this year can increase its volume as scheduled; the second is the re-pricing of the long-term agreement (LTA) contract in 2027. If the return on wafers still cannot surpass DDR5 by then, SK Hynix’s financing in the US stock market will be difficult to increase its return on investment, and its US stock premium will also shrink. Historically, the storage industry has hardly been able to escape the boom cycle, which ultimately resulted in oversupply due to production expansion. Morgan Stanley and major investment banks have also warned that memory remains an inherently cyclical commodity and that the "service provider" premium only exists when supply is in short supply. Now, as more and more wafer fabs are built, will this round of boom driven by AI repeat the traditional cycle of "oversupply and falling prices"? Tanya believes that in the AI era, the cyclicality of memory chips will still exist, but its amplitude will be gentler than the traditional fluctuation cycle. The reason is that the demand side has locked in demand and price in advance through long-term agreements, which can smooth the violent fluctuations of the cycle to a certain extent. Li Huihui also said that the current market's enthusiastic pursuit of memory chips is not about buying the chips themselves, but about buying "the right to use priority in future production capacity." Therefore, general storage is still dominated by cycles, and a security premium has been added to the pricing logic of high-end DDR and HBM.