I’ve recently been studying memory and found that some things are being seriously underestimated by the market.

First, the conclusion: memory stocks are basically not stable growth stocks; they’re cyclical trading assets. What you profit from is not the company itself, but the timing of the business cycle.

Why do I say that? Because the memory industry has a cycle you can’t escape—shortage → expansion → oversupply → price collapse → cutbacks → shortage again. This cycle runs through about once every few years. Nomura’s latest forecast says that in Q2 2026, the prices of DRAM and NAND Flash will rise quarter-over-quarter by 51% and 50%, respectively—much higher than the previous estimates of 6% and 20%. That’s why people who have been positioning now are starting to make money.

The global memory market is essentially dominated by just a few players. Samsung, SK Hynix, and Micron together monopolize more than 94% of the DRAM market. NAND Flash is similar: the combined market share of the top five manufacturers exceeds 80%. Supply is highly concentrated, and decisions by a small number of companies can determine the price cycle. That’s why memory stocks are so volatile.

When it comes to NAND Flash concept stocks, Phison is the purest one among Taiwan stocks. Their revenue and controller chip shipment volumes are hitting record highs at the same time. Even now, the NAND supply gap is still close to 20%. AI inference brings an almost unlimited demand for data storage, and even if capacity keeps expanding, it’s hard in the short term to change the supply-and-demand situation of shortages and insufficient supply.

In Taiwan stocks, there’s also Nanya Technology, focused on DRAM manufacturing. AI applications have already become a main growth driver, and their customized AI memory products have begun contributing to revenue. With the upgrade to DDR5 and the rebound in DRAM pricing, the overall fundamentals have clearly strengthened. Winbond is also decent: it focuses on niche DRAM and NOR Flash, avoiding the price-cut competition typical of general-purpose DRAM, so profit volatility is relatively smaller.

In the US market, Micron is the largest memory chip manufacturer in the United States. It produces both DRAM and NAND, making it the most resilient “storage” exposure in the US. As HBM capacity continues to expand, memory prices are entering an upward cycle, and overall profits are in a clearly visible recovery stage. SK Hynix is the global leader in DRAM and HBM shipments. HBM3e and HBM4 have already entered mass production, directly benefiting from the explosive demand for high-compute AI.

Phison, originally Toshiba Memory, is a major supplier of NAND Flash worldwide. Its market cap has recently jumped to the 10th place globally, rising sharply from 43rd at the end of last year. The NAND cycle lags behind DRAM—prices rise more slowly—making it suitable for a late-stage rotation layout.

The most important thing is to judge where the current cycle is. I’ll watch three things: first, whether the trend of DRAM contract prices has stopped falling; second, when supply-chain inventory days will turn from their peak and start trending downward; and third, whether major manufacturers show any signs of cutting capital expenditures. Currently, inventories of memory original-equipment manufacturers globally are at historic lows. Some big manufacturers have only about 4 weeks of total inventory left—this is why prices are easy to push higher but hard to pull down.

The volatility of memory stocks creates opportunities for both going long and going short. If you want to trade related swings or short-term moves flexibly, you can consider trading related stocks via CFDs. That way, you don’t need to actually hold the shares—you only need to predict the price direction.

To be honest, this round of memory stocks has become a big dark horse because of the AI supply gap, but memory stocks aren’t suitable for long-term holding. You need to judge where the business cycle is right now, build positions gradually near the bottom of the cycle, and step out gradually when prices surge and market sentiment becomes overheated. That’s the correct way to play memory stocks.