Recently, while researching the memory industry, I discovered a quite interesting phenomenon. Why is it that, although they are both semiconductors, memory stocks fluctuate so wildly? I started digging deeper and realized that the logic behind this is far more complex than I initially thought.

The memory industry, to put it simply, is an eternal cycle game. Shortages, expansion, oversupply, price drops, reduction in production, and then shortages again. This cycle repeats roughly every few years, and each time, the game is played very aggressively. According to the latest forecasts, in Q2 2026, DRAM and NAND prices are expected to increase by 51% and 50% quarter-over-quarter, respectively, much higher than the previous estimates of 6% and 20%, indicating that demand for memory is truly rebounding.

I noticed that many people investing in memory stocks actually have no clear idea what they are buying. Some buy chip manufacturers, some buy module makers, some focus on overseas giants, resulting in completely different risk tolerances. Chip manufacturers like Nanya Technology, Winbond, and Macronix directly produce chips, which have the highest profit flexibility when the market is good, but also face severe declines. In contrast, companies like Phison and Adata, which are IC control and module manufacturers, because they control software protocols and system integration, have relatively deeper moats, and their gross margin fluctuations are smaller.

Regarding the global ranking of memory companies, Samsung leads with a market cap of $897 billion, followed closely by SK Hynix and Micron. These three companies together dominate over 94% of the global DRAM market, with complete pricing power. Kioxia, as a major NAND Flash manufacturer, has seen its market value jump from 43rd place to 10th within half a year, reflecting how crazy the demand for high-end storage in AI data centers is.

On the US stock side, Micron (MU) is the stock I pay the most attention to. It controls both DRAM and NAND manufacturing. As HBM capacity continues to expand, overall profitability is clearly recovering, and the upside potential is worth looking forward to. SK Hynix is the absolute leader in HBM, with HBM3e and HBM4 already in mass production, directly benefiting from the explosive growth of AI computing power, with the strongest long-term moat. Although Lianqi Technology (MONT) is small, it focuses on DDR5 and HBM memory buffer chips, and as DDR5 penetration rapidly increases, this company almost has a quasi-monopoly position.

In Taiwan stocks, Nanya Technology is the most pure DRAM concept stock, with customized AI memory already contributing to revenue. Winbond adopts a niche strategy, avoiding price wars in general-purpose DRAM, and ranks eighth among global memory companies, which shows its competitiveness. Phison is the company with the highest purity in NAND Flash, and currently, the NAND supply gap is close to 20%, making it difficult to change the supply-demand imbalance in the short term.

A detail I especially want to emphasize is that memory stocks and AI stocks are fundamentally two different things. Memory stocks are cyclical trading; you profit from the rhythm of the economic cycle. AI stocks are trend investments; you profit from structural growth dividends. Many people confuse the two, resulting in completely wrong strategies.

If you want to trade memory stocks, swing trading is the most suitable approach. Currently, memory prices are still rising, and supply shortages are hard to alleviate in the short term. Holding manufacturing and module stocks now still has upward momentum. But the key is to grasp the rhythm—gradually positioning at the bottom of the cycle and gradually exiting when market sentiment overheats.

When judging entry points, I usually look for three signals. First, the spot price of DRAM stabilizes after falling, indicating demand is starting to absorb excess inventory. Second, leading manufacturers begin to cut production; although Samsung, SK Hynix, and Micron are experiencing booming earnings, they are all slowing capex, which shows they are guarding against supply excess around 2027. Third, inventory days decline from high levels; global memory manufacturers' inventories are already at historic lows, with some major companies holding only about four weeks of stock, which is a direct reason prices are easy to rise but hard to fall.

Ultimately, memory stocks are not stable growth stocks but cyclical trading assets. What you need to judge is which stage of the economic cycle we are in, rather than looking for companies to hold forever. The memory stocks that plunged deeply in the last cycle have become big dark horses this time because of the AI supply gap. The essence of memory stocks is that you profit from the rhythm, not the company itself.