AI storage giant encounters ice and fire: SK Hynix takes advantage of HBM, but Samsung suffers from cyclical dependence
Storage giant SK Hynix has opened the door to the U.S. capital market with a huge fundraising plan of US$26.5 billion. On July 10, SK Hynix launched a pre-issuance in the form of American Depository Receipts (ADRs). It closed up 12.76% that day at $168.01 per share. After-hours trading rose again by 2.43% and is now at $172.09 per share. For this big test to go to the United States, SK hynix broke with industry practice and completely canceled the traditional "price ceiling" clause in the long-term supply contract.
Storage giant SK Hynix has opened the door to the U.S. capital market with a huge fundraising plan of US$26.5 billion. On July 10, SK Hynix launched a pre-issuance in the form of American Depository Receipts (ADRs). It closed up 12.76% that day at $168.01 per share. After-hours trading rose again by 2.43% and is now at $172.09 per share. For this big test to go to the United States, SK hynix broke with industry practice and completely canceled the traditional "price ceiling" clause in the long-term supply contract. This extremely offensive move is not only to demonstrate its absolute voice in the AI storage industry chain, but also to prove to global capital its unlimited possibilities in the future. Just as SK Hynix is making a high-profile sprint to the U.S. stock market, Samsung Electronics is stuck in the quagmire of "exhausting all the benefits" in the local market. Although Samsung Electronics previously disclosed an "explosive" performance forecast for the second quarter of 2026 - operating profit is expected to soar more than 18 times year-on-year, setting a new high in the company's history, the capital market has staged an extreme reverse trend. In the trading from July 8 to 10, Samsung Electronics not only failed to stabilize its stock price with its "strongest performance in history", but instead triggered a "stampede" selling of leveraged funds in the Korean stock market, which seriously dragged down the overall trend of KOSPI (Korea Composite Stock Price Index). On one side, SK Hynix went to the United States to compete for global capital with its "monopoly pricing power". On the other side, Samsung Electronics reached its peak performance but was ruthlessly abandoned by capital. Behind the extreme divergence of the two giants, Times Finance talked to investment bank researchers, investors, academics and storage industry traders, trying to find the real logic behind this AI storage super cycle from the undercurrent of the industrial structure and market sentiment... Source of changes in the Korean stock market on July 8: Screenshots from Times Finance reporters Capital logic: forced correction from “buying expectations” to “selling facts” Regarding Samsung Electronics' "new high performance and sharp drop in stock price", a researcher from a leading investment bank in Hong Kong told Times Finance that this is not caused by a single factor, but the result of the dual factors of combined risk control and profit forecast reduction. The first is the hard constraint on the portfolio risk budget. The researcher analyzed that most fund managers have strict limits on individual stock holdings, and generally a single stock does not exceed 8% to 10% of the portfolio's net value. In the past year, storage stocks such as SK Hynix and Samsung Electronics have surged, causing their positions to passively exceed the standard. At this time, once a negative signal appears (even if it is just a loosening of the narrative), the risk control system will forcefully trigger the reduction of positions. “This is reactive compliance, not proactive timing.” Another reason is the collective reduction of forward-looking EPS (earnings per share), which is the core trigger of this round of sector correction, and "Meta news" is the key to puncturing the bubble. The logical chain seen by institutional traders is: Meta rents out idle AI servers, and the industry's overall computing power supply increases. As a result, the marginal benefits of cloud manufacturers' capital expenditures begin to diminish, causing the price increase slope of HBM (high-bandwidth memory) to slow down. Researchers from multiple brokerages immediately revised down their 2026 HBM unit price forecasts. In addition, the capital market has already overdrawn the benefits in advance. In the past 18 months, Samsung Electronics' stock price has soared by 158%, benefiting from the increase in memory chip prices driven by the explosion of AI computing power. Gary Tan, portfolio manager of Allspring Global Investments (a well-known global asset management company), once publicly stated: "In a strong memory upcycle, when data exceeds expectations, most of the good news has been reflected in positions and expectations. The performance forecast may only confirm investors' existing expectations, triggering profit-taking rather than further rises." Storage giants are heading towards differentiation: SK Hynix steps on HBM’s wind, Samsung Electronics suffers from storage dependence
To put it simply, HBM and ordinary storage used in mobile phones (such as LPDDR) are essentially different "snacks" made from the same "flour" (wafer). Both require wafers to manufacture, but since HBM mainly supplies AI servers and has extremely high profits, storage manufacturers are naturally more willing to use limited wafers to produce HBM, which results in fewer raw materials left for mobile phone memory. Moreover, it is very difficult to manufacture HBM. It is necessary to stack multiple layers of chips like a building. To produce a 1GB HBM, the wafer production capacity consumed is 3 to 4 times that of ordinary mobile phone memory chips. Therefore, every time a manufacturer produces more HBM, it will seriously squeeze the production capacity of storage chips, eventually leading to a shortage of mobile phone storage chips and rising prices. In the context of the explosive demand for AI computing power, this "crowding out effect" of storage capacity caused by AI has caused obvious differentiation in the situation of major manufacturers. On the one hand, the bargaining power of upstream suppliers has been significantly enhanced. According to multiple industry media reports, SK Hynix has completely canceled the traditional price ceiling clause in the long-term supply agreement (LTA) recently signed with customers, becoming the only major memory manufacturer in the market that is not subject to such constraints. Mr. Wang, a senior researcher at the Investment and Financing Committee of Jiangxi University of Finance and Economics, pointed out that SK Hynix’s confidence in this move stems from its leading technological advantages in the field of HBM (high-bandwidth memory) and its deep binding with core customers, which enables it to deliver more flexible profit expectations to the capital market. Compared with SK Hynix and other manufacturers, which can travel lightly and concentrate their resources on the high-profit HBM track, although Samsung Electronics occupies an important share of the overall storage market, its business structure is more diversified, covering mobile terminals, wafer foundry and other sectors. The pricing strategy needs to take into account the synergy and balance of different business lines. Although the memory chip department contributes more than 90% of Samsung's operating profits, businesses such as mobile phones and wafer foundry continue to be under pressure. Affected by the storage shortage caused by the AI boom, Samsung Electronics' mobile business unit even faces the risk of a full-year loss for the first time in history. Reliance on a single business has made Samsung Electronics a "storage cycle vassal". Once the storage cycle goes down, the company's overall performance will fall rapidly. At the same time, the demand logic of downstream giants is also undergoing subtle changes. Senior researcher Mr. Wang analyzed that the core reason why giants such as Meta choose to rent out idle computing power is that their self-developed model capabilities have not yet ranked among the top echelons in the world, and they are trying to share costs by revitalizing their existing computing power. This strategic shift is a "double-edged sword" for Samsung Electronics: On the one hand, in order to reduce computing power costs, Meta and others are accelerating the transfer of customized AI chip foundry orders to Samsung Electronics; but on the other hand, downstream giants have shifted from "grabbing hardware regardless of cost" to "refined operations," which means that the marginal benefits of AI infrastructure investment are diminishing, which directly shakes the underlying logic of Samsung Electronics' continued surge in memory chips. Judging from market performance, although the average price of DRAM memory (Dynamic Random Access Memory) at Samsung Electronics increased by more than 40% month-on-month in the second quarter, and the price of NAND (non-volatile flash memory) flash memory increased by more than 50%, the market has reached a consensus that the most rapid period of price increases has ended. Mr. Zhang, an investor at a state-owned venture capital institution in Shenzhen, said that although storage, as a core cost item of AI infrastructure, will continue to grow in the future, the secondary market is highly susceptible to sentiment. On the one hand, Apple and other terminal products have announced price increases, which has triggered concerns in the industry that excessive storage profits will push up upstream costs and lead to reduced demand; on the other hand, high profit growth means that the base of the next financial report will become higher and the growth rate will inevitably slow down, which will suppress stock price performance to a certain extent. Market fragmentation: the "temperature difference" between the surge in spot prices and the ebb of capital Panic is spreading in the capital market, but the experience in the real industry is completely different, showing a severe "temperature difference." Liu Chao (pseudonym), a storage trader in Huaqiangbei, Shenzhen, told Times Finance that the price increase in the spot market is staggering. Taking DDR5 64G 5600 memory sticks as an example, the price in trade channels has increased from 1,620 yuan in March last year to 16,200 yuan today, a 10-fold increase in more than a year.
"Storage products are all produced in South Korea, and most of the quotas are given to large American manufacturers. Not much of the quota is given to China. After being distributed through layers of Hong Kong, Taiwan, and mainland China, the goods will be less and the price will be higher." Liu Chao said that some cargo owners adopt the strategy of the highest bidder. As soon as the spot goods are sold, the price will increase immediately on the same day. Liu Chao believes that performance and stock price are affected by short-term sentiment, which is a behavior of the capital market and has little impact on supply and demand prices in channels and retail markets. Can this boom in the spot market continue? JPMorgan Chase’s latest research report points out that the proportion of AI storage in cloud service providers’ capital expenditures is rising rapidly, and is expected to reach 52% this year, and is expected to exceed 70% next year. Is this extreme spending structure really sustainable? If the commercialization speed of AI services cannot keep up with the expansion of hardware, this performance carnival driven by computing power may face correction at any time. It is worth mentioning that Samsung and SK Hynix recently jointly committed to investing up to 3,755 trillion won to expand chip production capacity. The memory chip industry itself has a cyclical development pattern. It takes 2.5 to 3 years for a new factory to reach full production. The current concentrated capital expenditure will release massive new supply from 2028 to 2029. Once the growth rate of capital expenditures of AI cloud manufacturers slows down, the supply and demand pattern will quickly reverse, and the downward pressure on forward prices cannot be ignored. This performance forecast from Samsung Electronics is not only the peak moment of the AI storage super cycle, but also an early warning signal that the cycle has peaked. When the surge in profits becomes a "star card", the capital market's attention has turned further afield. For the global semiconductor industry, saying goodbye to blind speculation in the cycle and turning to deterministic growth is the only way to get through the cycle. (
To put it simply, HBM and ordinary storage used in mobile phones (such as LPDDR) are essentially different "snacks" made from the same "flour" (wafer). Both require wafers to manufacture, but since HBM mainly supplies AI servers and has extremely high profits, storage manufacturers are naturally more willing to use limited wafers to produce HBM, which results in fewer raw materials left for mobile phone memory. Moreover, it is very difficult to manufacture HBM. It is necessary to stack multiple layers of chips like a building. To produce a 1GB HBM, the wafer production capacity consumed is 3 to 4 times that of ordinary mobile phone memory chips. Therefore, every time a manufacturer produces more HBM, it will seriously squeeze the production capacity of storage chips, eventually leading to a shortage of mobile phone storage chips and rising prices. In the context of the explosive demand for AI computing power, this "crowding out effect" of storage capacity caused by AI has caused obvious differentiation in the situation of major manufacturers. On the one hand, the bargaining power of upstream suppliers has been significantly enhanced. According to multiple industry media reports, SK Hynix has completely canceled the traditional price ceiling clause in the long-term supply agreement (LTA) recently signed with customers, becoming the only major memory manufacturer in the market that is not subject to such constraints. Mr. Wang, a senior researcher at the Investment and Financing Committee of Jiangxi University of Finance and Economics, pointed out that SK Hynix’s confidence in this move stems from its leading technological advantages in the field of HBM (high-bandwidth memory) and its deep binding with core customers, which enables it to deliver more flexible profit expectations to the capital market. Compared with SK Hynix and other manufacturers, which can travel lightly and concentrate their resources on the high-profit HBM track, although Samsung Electronics occupies an important share of the overall storage market, its business structure is more diversified, covering mobile terminals, wafer foundry and other sectors. The pricing strategy needs to take into account the synergy and balance of different business lines. Although the memory chip department contributes more than 90% of Samsung's operating profits, businesses such as mobile phones and wafer foundry continue to be under pressure. Affected by the storage shortage caused by the AI boom, Samsung Electronics' mobile business unit even faces the risk of a full-year loss for the first time in history. Reliance on a single business has made Samsung Electronics a "storage cycle vassal". Once the storage cycle goes down, the company's overall performance will fall rapidly. At the same time, the demand logic of downstream giants is also undergoing subtle changes. Senior researcher Mr. Wang analyzed that the core reason why giants such as Meta choose to rent out idle computing power is that their self-developed model capabilities have not yet ranked among the top echelons in the world, and they are trying to share costs by revitalizing their existing computing power. This strategic shift is a "double-edged sword" for Samsung Electronics: On the one hand, in order to reduce computing power costs, Meta and others are accelerating the transfer of customized AI chip foundry orders to Samsung Electronics; but on the other hand, downstream giants have shifted from "grabbing hardware regardless of cost" to "refined operations," which means that the marginal benefits of AI infrastructure investment are diminishing, which directly shakes the underlying logic of Samsung Electronics' continued surge in memory chips. Judging from market performance, although the average price of DRAM memory (Dynamic Random Access Memory) at Samsung Electronics increased by more than 40% month-on-month in the second quarter, and the price of NAND (non-volatile flash memory) flash memory increased by more than 50%, the market has reached a consensus that the most rapid period of price increases has ended. Mr. Zhang, an investor at a state-owned venture capital institution in Shenzhen, said that although storage, as a core cost item of AI infrastructure, will continue to grow in the future, the secondary market is highly susceptible to sentiment. On the one hand, Apple and other terminal products have announced price increases, which has triggered concerns in the industry that excessive storage profits will push up upstream costs and lead to reduced demand; on the other hand, high profit growth means that the base of the next financial report will become higher and the growth rate will inevitably slow down, which will suppress stock price performance to a certain extent. Market fragmentation: the "temperature difference" between the surge in spot prices and the ebb of capital Panic is spreading in the capital market, but the experience in the real industry is completely different, showing a severe "temperature difference." Liu Chao (pseudonym), a storage trader in Huaqiangbei, Shenzhen, told Times Finance that the price increase in the spot market is staggering. Taking DDR5 64G 5600 memory sticks as an example, the price in trade channels has increased from 1,620 yuan in March last year to 16,200 yuan today, a 10-fold increase in more than a year.
"Storage products are all produced in South Korea, and most of the quotas are given to large American manufacturers. Not much of the quota is given to China. After being distributed through layers of Hong Kong, Taiwan, and mainland China, the goods will be less and the price will be higher." Liu Chao said that some cargo owners adopt the strategy of the highest bidder. As soon as the spot goods are sold, the price will increase immediately on the same day. Liu Chao believes that performance and stock price are affected by short-term sentiment, which is a behavior of the capital market and has little impact on supply and demand prices in channels and retail markets. Can this boom in the spot market continue? JPMorgan Chase’s latest research report points out that the proportion of AI storage in cloud service providers’ capital expenditures is rising rapidly, and is expected to reach 52% this year, and is expected to exceed 70% next year. Is this extreme spending structure really sustainable? If the commercialization speed of AI services cannot keep up with the expansion of hardware, this performance carnival driven by computing power may face correction at any time. It is worth mentioning that Samsung and SK Hynix recently jointly committed to investing up to 3,755 trillion won to expand chip production capacity. The memory chip industry itself has a cyclical development pattern. It takes 2.5 to 3 years for a new factory to reach full production. The current concentrated capital expenditure will release massive new supply from 2028 to 2029. Once the growth rate of capital expenditures of AI cloud manufacturers slows down, the supply and demand pattern will quickly reverse, and the downward pressure on forward prices cannot be ignored. This performance forecast from Samsung Electronics is not only the peak moment of the AI storage super cycle, but also an early warning signal that the cycle has peaked. When the surge in profits becomes a "star card", the capital market's attention has turned further afield. For the global semiconductor industry, saying goodbye to blind speculation in the cycle and turning to deterministic growth is the only way to get through the cycle. (