Mar 06, 2024
Memory Semiconductors: Why SK Hynix Is the Answer
Ryunsu Sung
Investing in semiconductors is simpler than most people think. The clear answer is to invest in semiconductor companies whose business cycles are relatively less volatile, whose demand is growing, and whose demand is highly likely to keep growing in the future. NVIDIA, which is up more than 70% just this year, is one example.
So before explaining why SK Hynix is the right answer for a memory semiconductor play, I want to first tell investors who insist on investing in the memory segment to look for alternatives. Among cyclical semiconductor industries, memory is the most commoditized segment with the largest boom-and-bust swings, and the same is highly likely to become true even for HBM, which is being marketed as AI-only memory.
However, at this point in time, it seems clear that the memory cycle—especially DRAM—will rebound on the back of strong demand for HBM (High Bandwidth Memory) used in AI chips. I will first cover points 1 and 2, and then explain the logic for why rising HBM demand leads to a rebound in DRAM as a whole.
- Samsung Electronics: excluded. Putting aside the debate over whether Samsung misjudged its HBM strategy and whether its leadership in memory technology is eroding, the company cannot be considered a pure memory semiconductor play, so it is excluded. Of its 2023 revenue of 258.94 trillion won, memory semiconductors accounted for 44.13 trillion won, only about one-sixth of the total, which makes Samsung Electronics an unsuitable vehicle for a pure memory bet.
It is also awkward to classify Samsung Electronics as a semiconductor IDM (Integrated Device Manufacturer) like Intel, because of the sheer size of its MX (mobile experience) division. Therefore, Samsung Electronics is not an appropriate company if your goal is to invest in anticipation of a memory upcycle. For this reason, I personally believe Samsung Electronics should be split into three separate businesses—mobile, memory, and foundry—but since this is unlikely to align with Lee Jae-yong’s position, I will leave that discussion aside. - Micron: the only alternative.
Micron is a pure-play US memory semiconductor company that produces DRAM and NAND. In effect, it is the only real substitute for SK Hynix. However, in HBM, which is arguably the highest value-added segment within memory, its market share is only about 10%. SK Hynix is around 50%. Once you move to HBM3 and above, the gap widens even further (I have not found official statistics).
In addition, Micron still has no production experience using EUV equipment. It is investing around 5 trillion won to introduce EUV-based processes at its plants in Japan and Taiwan, but given that Samsung Electronics and SK Hynix adopted EUV ahead of it, there is a high likelihood that Micron will be at a slight cost disadvantage when EUV is finally applied.
Among all types of semiconductors, memory is the closest to a commodity. Despite ongoing consolidation and the resulting oligopoly of three major players, memory still goes through boom-and-bust cycles because it requires heavy capital expenditure and incumbents can always spend more to ramp up output. Like raw materials, memory supply can be increased by investing more in capacity, and if supply is increased too quickly, it leads to oversupply. And just like commodities, there are multiple types of memory.
Now it is time to understand the context. Memory makers are just emerging from an unprecedented downturn and have only recently escaped negative cash flow, so their total capex budgets are constrained. They will convert most of the cash they generate into HBM capacity and keep the supply growth of DRAM and NAND low. This is the best-case scenario for memory manufacturers.
HBM on its own will bring an enormous boom to memory semiconductor companies, but because HBM requires heavy capital investment, capex will be diverted away from DRAM and NAND, leaving bit growth in memory semiconductors short of demand growth. As a result, the industry is likely to hit a new cyclical peak within a few years, and in that scenario, I believe it is rational to bet on the leader, SK Hynix. Although it is headquartered in Korea, it trades at a discount to Micron in terms of the P/E multiple.
If you have decided to invest in memory semiconductors, the answer is SK Hynix.
Newsletter
Be the first to get news about original content, newsletters, and special events.