This Korean semiconductor analyst's research report (article) is quite interesting: Why are memory stocks transitioning from "cyclical stocks" to "growth stocks" in a Supercycle? one ️⃣ Why were memory stocks valued using PB in the past? The profit fluctuation is too large. Memory factories (such as Samsung, Micron, and Hynix) build factories first, start production first, and then wait for customers to place orders. The result is that once there is a downward cycle, there will be oversupply and price collapse. So when you make a profit, you make a huge profit; when you lose, you make a huge loss. It's like riding a roller coaster, a very typical cyclical stock. So analysts use PB price to book ratio to see: Because 'book assets' are relatively stable, while' profits' fluctuate too dramatically, PB can better reflect the company's bottom value. This is the valuation logic of 'cyclical stocks'. Here is a comparison of TSMC (TSMC): TSMC's model is "customers place orders first, and then build production capacity", so it is not easy for supply to exceed demand. Stable profits can be estimated using PE. So in the manufacturing industry, TSMC is a 'stable growth stock', while memory factories used to be a 'cyclical stock'. two ️⃣ Why PB is malfunctioning now: AI has changed everything The extensive use of HBM in AI training (NVIDIA, OpenAI, etc.) has changed the logic of Hynix. SK Hynix HBM is almost exclusively supplied, with stable and skyrocketing profits. As a result, its stock price "broke through the historical PB limit", indicating that the market no longer views it according to old standards. The demand for AI will not decline with the macroeconomic downturn, even if the economy is in a downturn, AI servers will still be built, and data center CapEx is skyrocketing. >This is questionable, but at least at this point (2025 Q4) it seems relatively healthy. The financial situation of the 7 sisters is still much better than that of the telecom companies in the Internet foam. So the prosperity of AI memory is no longer fluctuating with GDP. This is a qualitative change. On the other hand, production capacity cannot be easily expanded, and memory factories have been very conservative in recent years (mainly due to fear from previous cycles). At the same time, advanced processes and packaging equipment are very expensive. Expanding production is not easy either. Supply reduction+long-term demand growth=price stability. three ️⃣ Memory stocks should now be valued using PE Because profits have become more stable. It's no longer about 'expanding production first, placing orders later', but about 'signing orders first, expanding production later'. Having a long-term supply contract (2-3 years). This means that both cash flow and profit can be predicted, and can be valued like TSMC. SK Securities said: Now is the AI super cycle. >The author emphasizes that SK Securities and SK Hynix are not the same company, haha HBM is just the beginning, and AI's demand for server DRAM, SSD, GDDR7, etc. is also exploding. Data center capital expenditures are expected to increase by 40% over the next five years, and the entire memory market will be upgraded to 'part of AI infrastructure'. four ️⃣ Chives love to fantasize New target price and valuation multiple: Samsung Electronics - Calculated at 15 times PE, the target price is 170000 Korean won. SK Hynix → Calculated at 11 times PER, the target price is 1 million Korean won. They are all 50-100% higher than before. SK Securities believes that the memory industry is undergoing structural upgrading. It's no longer about looking at the economic cycle, but at the pace of AI construction. The industry characteristics have changed from 'volatile' to 'growth oriented'. So we should use PE estimation just like looking at TSMC or NVIDIA. This means that the valuation multiple will increase. The rise in stock prices is not just a short-term trend, but reflects a structural reassessment.