Summary of Key Points
Recently, the stock prices of South Korean storage giants Samsung Electronics and SK Hynix have rebounded, driven by market confidence stemming from their collaboration with NVIDIA on HBM (High Bandwidth Memory) chips. The surge in AI demand has led to a tight supply and rising prices of HBM chips. However, there were concerns earlier that the peak in AI storage demand would cause stock price declines. Choi Tae-won, the chairman of the SK Group, stated that AI and overall semiconductor demand will increase significantly next year, but the widening supply gap requires accelerated production expansion. Market fears of potential overcapacity have been alleviated due to the long cycle for new production capacity to be deployed. Securities firms predict that storage prices will continue to rise in the second half of 2026, with structural shortages being the main factor, although weak demand from consumer electronics and expanded production may limit the increase. The high difficulty of producing HBM chips is driving up prices, and whether the storage supercycle has ended depends on the pace of AI model iteration.
1. Stock Price Rebound: Cooperation with NVIDIA as a Boost
On July 22, the stock prices of Samsung and SK Hynix rose by 4.63% and 5.94%, respectively, pushing the South Korean KOSPI index up by nearly 5%. The direct reason is the roundtable discussion scheduled for this Friday (July 24) between the CEOs of these two companies and NVIDIA's CEO Jensen Huang. Samsung and SK Hynix are supplying NVIDIA with the essential HBM chips for its AI applications, and NVIDIA's next-generation AI superchip, "Vera Rubin," will also use their HBM4 technology.
Earlier in July, stock prices fell sharply (SK Hynix' price dropped by up to 30%) due to concerns that AI storage demand had peaked. Now that the cooperation has been confirmed, market confidence has returned, leading to a rebound in stock prices.
2. Why are HBM Chips in Such High Demand?
HBM chips can be considered the "super memory" for AI applications, as AI operations require processing massive amounts of data, and conventional memory is not fast enough. The high demand for HBM chips is due to three main factors:
- Surging Demand: AI models and supercomputers have a tremendous need for HBM chips.
- High Production Difficulty: The production cycle for HBM chips takes 4-6 months, and initial yields are low (many chips are defective). Additionally, the wafers used are three times larger than those for conventional DDR5 memory, making it challenging to increase production quickly.
- Customer Capacity Lock-ups: Downstream companies (such as NVIDIA) have signed long-term agreements with Samsung and SK Hynix to secure their supply, further exacerbating the tight supply situation.
As a result, the price of HBM4 chips has been on the rise.
3. The Debate over Expansion: Expand to Meet Demand or Fear Excess Capacity?
Choi Tae-won believes that AI demand will increase by 60%-100% next year, and overall semiconductor demand will grow by 50%-60%, but there will be almost no new supply, leading to a larger gap. Therefore, expansion is necessary (both South Korea and the United States are building factories if conditions permit).
Some worry that expanded production could end the semiconductor "supercycle" prematurely (with prices plummeting). However, Choi Tae-won argues that chip prices are already abnormally high, and expanding production can still be profitable. Moreover, only by increasing supply and maintaining market scale can South Korea's semiconductor industry continue to develop. More importantly, experts point out that the 800 trillion won investment in storage facilities announced by South Korea will not result in excess capacity until at least 2033, so there is no immediate risk of overproduction.
4. Future Storage Prices: Can the Rise Continue?
South Korean securities firms and institutions predict that storage prices will continue to rise in the second half of 2026, driven by structural shortages caused by AI-driven demand for chips like HBM. However, two factors may limit the increase:
- Weak Consumer Electronics: Poor sales of smartphones, computers, and other consumer electronics are reducing overall memory demand.
- Manufacturer Expansion: Although new production capacity will be available later, expansion plans will raise market expectations of increased supply, potentially limiting price increases.
5. Has the Storage "Supercycle" Ended?
Experts believe that the storage market cannot be separated from the AI cycle. As long as AI models continue to evolve (for example, with GPT updates and the emergence of new large models), the storage supercycle will not end.
However, there are two risks to consider:
- If high chip prices are passed on to end products (resulting in higher prices for smartphones and computers), consumers may be reluctant to upgrade, which could suppress demand for conventional memory.
- If AI demand does not meet expectations, excess capacity created by expanded production in 2028 could lead to inventory buildup and price declines.
In summary, this news reflects the supply-demand dynamics in the storage chip industry driven by AI demand. In the short term, chip shortages are driving up prices, but in the long run, expansion is needed to avoid overcapacity. The pace of AI development will be a critical factor determining the outcome. For consumers, it means that companies like Samsung and SK Hynix will benefit from the growth of AI, but they also need to be cautious about potential excess capacity risks.