A video interview with SK Securities analyst Han Dong-hee was uploaded today on the 채국장 channel.
Those who are invested in Samsung Electronics/Hynix may have their own views, but please refer to it once.
The following are my personal thoughts;
Recently, there has been a market view that the aggressive expansion plans announced by the three memory semiconductor companies (from the perspective of traditional cyclical industries) could lead to oversupply. This concern is believed to be reflected in recent stock price movements.
But is that really the case?
1) In the AI era, the nature of memory changes - while it used to be a general-purpose item purchased only when needed, platform companies (mainly hyperscalers we know) seeking to generate revenue from AI will have their profitability directly tied to how much memory they possess. In other words, if a competitor has more memory than them, those companies that do not will be eliminated from the market.
2) Demand for high value-added custom memories such as HBM and SOCAMM is unlikely to decrease as technology advances - in the past, the era of simply mass-producing standard specifications like DRAM and NAND meant that increasing wafer yield per wafer and mass production were key competitive advantages. However, as high-performance GPUs, NPUs, CPUs, and LOGIC chips requiring HBMs and SOCAMMs continue to be introduced, demand for these memories will never decrease. For token sellers who need these chips, high-performance chips are directly linked to price competitiveness.
3) Interestingly, there was a recent report that Hynix's second-quarter sales, which has a high proportion of HBM, would fall short of consensus. The reason is that it takes four times as many general DRAM wafers to make one HBM, resulting in a shortage of DRAM supply and driving up DRAM prices higher than HBM. As a result, Hynix, which has a higher proportion of HBM than Samsung Electronics or Micron, is expected to have seen its operating profit margin eroded. Even if this report comes out, it is a very short-term phenomenon. Interestingly, the current market situation is that the wafer total quantity for HBM, SOCAMM, DRAM, and NAND is intertwined with a very complex sampling. From a company's perspective, maximizing profits with these limited resources will lead to allocation adjustments. This (based on a 2-3 year outlook) means that even if the wafer production volume of the three memory companies doubles, it will not change easily. The key here is HBM. As NVIDIA's roadmap shows with products like Vera Rubin, Rubin Ultra, and Feynman, chips that continuously consume HBMs with faster speeds and larger capacities are likely to make wafers a scarce resource. Companies will allocate more wafers to the product with higher margins, which will drive up prices for other products.
4) Why Hynix is better than Samsung Electronics or Micron: If DRAM is the only focus, it's not easy to adjust the volume around HBM. In other words, there can be greater volatility in operating profit margins depending on DRAM prices. However, since Hynix has a high proportion of HBM, when DRAM margins are higher as in the last quarter, they will gradually allocate more wafers to DRAM to maximize profits. As a result, the supply of HBM will decrease, naturally leading to an increase in HBM margins. Of course, there are contracts with customers who want to offset this price volatility through LTAs, so it won't be reflected immediately. However, when new HBM and customer chips are released, Hynix will have the pricing power.
5) I agree with Han Dong-hee analyst's view that Hynix will ultimately break away from traditional cyclical industries and adopt a growth structure similar to NVIDIA's.
If you have your own logic for why you invested and your conviction doesn't change, this is the time when you need a canceling function that can effectively filter out all the noise in the market.
