Recently, I’ve been looking into the memory sector and realized why the fluctuations in semiconductor stocks can vary so much. The key lies in which part of the supply chain you’re investing in.

The memory industry is actually divided into three layers. The upstream includes companies that directly manufacture chips, like Nanya Technology, Winbond, and Macronix. These companies are most flexible when prices rise but are also the first to be hit when the market turns sour. The midstream consists of memory module manufacturers like Phison and Team Group, responsible for integrating chips into common products like SSDs, with relatively stable profit fluctuations. The top layer includes the three global giants—Micron, Samsung, and SK Hynix—controlling over 90% of the worldwide DRAM market.

Why is now a noteworthy time to pay attention? Because the HBM (High Bandwidth Memory) wave has completely changed the supply landscape due to the AI boom. Hynix and Micron have shifted their capacity to produce HBM, while Taiwanese manufacturers are benefiting from order redirection. According to the latest forecasts, by Q2 2026, DRAM and NAND prices are expected to rise by 51% and 50%, respectively, significantly higher than previous estimates of 6% and 20%.

Honestly, memory stocks are cyclical trading assets, not stable growth stocks. The entire industry is always cycling: shortage → capacity expansion → oversupply → price collapse → reduction → shortage again. This cycle repeats every few years, and your profit depends on your ability to judge this rhythm.

On the U.S. stock side, Micron Technology is the purest play, producing both DRAM and NAND. As HBM capacity expands, profits are clearly recovering. SK Hynix is the leader in HBM, with HBM3e and HBM4 already in mass production, directly benefiting from AI computing demands. Lanki Technology focuses on DDR5 and HBM memory buffer chips, holding a quasi-monopolistic position in this field.

In Taiwan stocks, Nanya Technology is the purest DRAM concept, with customized AI memory already contributing revenue. Winbond adopts a niche approach, specializing in NOR Flash and special DRAM, avoiding the price wars of general-purpose DRAM. Phison is the most specialized NAND Flash company; currently, NAND supply is still about 20% short, and the storage demand driven by AI inference is almost unlimited, making oversupply unlikely in the short term. Macronix focuses on NOR Flash and ROM, with technological advantages in automotive and industrial fields, suitable for balancing memory market fluctuations.

What are the key signals now? First, watch whether DRAM prices stop falling, as this is the most critical indicator of an industry turnaround. Second, pay attention to the capex of leading manufacturers—Samsung, Hynix, and Micron—although their performance is expected to surge in 2026, they are all pulling back to control capacity and prevent oversupply in 2027. Third, look at inventory days; currently, global memory manufacturers’ inventories are at historic lows, with some major companies holding only about four weeks’ worth, which explains why prices tend to rise easily but fall slowly.

If you want to trade cyclically, the volatility of memory stocks is both a risk and an opportunity. Consider gradually accumulating at the bottom of the cycle and gradually exiting when market sentiment overheats. But if your risk tolerance is low, wait until prices fall deeply before entering, because these stocks tend to be best bought at their lowest points.

In short, memory stocks profit from the rhythm, not the company itself. You need to judge where the cycle currently stands, rather than looking for stocks to hold forever. Due to the AI supply gap, memory stocks that plunged last cycle have become big dark horses. Now is a good opportunity—start by tracking DRAM contract prices in a demo account, study the financial reports of major manufacturers, and once you have a clearer grasp of the cycle, consider small real trades.