I’ve been paying close attention to memory stocks lately, and I’ve noticed that many people still don’t really understand this sector. Today, I’d like to share some of my observations.

First, let’s talk about a phenomenon: why are memory stocks so much more volatile? The root cause is that this industry has a cycle you can’t escape: shortage → capacity expansion → oversupply → price collapse → production cuts → shortage again. Nomura’s latest forecast shows that in this year’s second quarter, DRAM and NAND prices rose quarter-over-quarter by 51% and 50%, respectively. This indicates that we are currently in an upcycle.

But memory stocks aren’t simply growth stocks. They are classic cyclical trading assets. The money is made from the timing of the business cycle—not from the companies themselves. Many people mix this up and trade memory stocks like they’re AI stocks, and the result is they get trapped badly. The distinction is simple: AI stocks are trend investments; memory stocks are cyclical trades.

When it comes to the supply chain, many people focus only on chip manufacturers, but the risks for companies in different links are completely different. Companies like Nanya Technology, Winbond, and Macronix, which directly produce chips, have the greatest earnings flexibility, but they are also the most affected by the cyclical swings—both up and down can be sharp. On the other hand, companies like Phison and ADATA, which control ICs and modules, have deeper moats because they hold software protocols and handle system integration, so their gross margin fluctuations are relatively smaller. If you want something steadier, module makers are a comparatively safer choice.

The global market is actually dominated by three giants. Samsung, SK Hynix, and Micron together hold more than 94% of the global DRAM market share. As for the hottest high-bandwidth memory (HBM) right now, the technology leaders are mainly Hynix and Micron. When these big manufacturers allocate their full capacity to producing HBM for AI needs, Taiwanese companies can capture the residual order-spillover benefits. That’s why Taiwanese memory stocks have been performing relatively well recently.

As for Micron, the only U.S. company with large-scale DRAM and NAND manufacturing capabilities, as HBM capacity continues to expand, its overall profitability is currently in a clearly visible recovery phase—so the upside potential is worth watching. Hynix’s HBM3e and HBM4 have already entered mass production first, directly benefiting from the explosive demand for high computing power driven by AI.

On the Taiwan market, Nanya Technology is the purest DRAM concept stock. Its customized AI memory products have already begun contributing to revenue. Winbond focuses on niche DRAM and NOR Flash, avoiding the price-cutting wars of general-purpose DRAM, which leads to relatively smaller earnings volatility. Phison is the highest-purity NAND Flash company. Currently, the NAND supply shortfall is still close to 20%, so it’s difficult to change the supply-and-demand imbalance in the short term.

So when is the best time to buy memory stocks? Usually, they start to rebound when prices stop falling and the market is still pessimistic. You should watch for three signals: whether DRAM prices have stopped declining and stabilized, whether leading companies have begun reducing production, and whether inventory days have fallen back from their peak. Currently, inventories at global memory manufacturers are at historic lows. For some major companies, inventory is only about 4 weeks left, which is a direct reason why prices are easier to rise than to fall.

To be honest, the trading logic for memory stocks is to gradually build positions at the bottom of the cycle, then gradually exit when prices surge and market sentiment becomes overheated. Right now, memory prices are still rising, and the supply shortage situation on the supply side is unlikely to ease in the short term. Holding stocks related to both the manufacturing and module ends still has upward momentum.

My final advice is to practice first using a simulated account. Over the next few weeks, observe the price trend of DRAM contracts, track the financial reports and capital expenditure directions of major manufacturers, and determine which stage the memory cycle is currently in. Once you have a clearer grasp of the cycle, then consider trading with a small amount of real money. Remember: you earn from the rhythm, not from the company itself.