Summary of Key Points
The current memory market is in the most complex phase of this cycle: AI demand (especially from cloud service providers' capital expenditures) is still supporting high market activity, but price growth has significantly slowed down, and there are potential issues with inventory structures. The expansion of Yangtze Memory Technologies (YMTC) becomes a key factor affecting NAND supply. The overall cycle will not collapse suddenly; it is more likely to peak in the fourth quarter of 2026 with a "plateau at high levels"— prices will remain high, but the growth rate will decline, and the market has already reflected concerns about this turning point through stock price adjustments.
1. AI Demand Is Still Supporting the Market, But It Depends on Whether Investments Can Be Profitable
AI is the core reason for the extended memory cycle. Cloud providers like Microsoft and Google have raised their capital expenditure forecasts for 2027 by approximately 55% this year. They are investing in GPUs, HBM (high-end memory), server memory, and large-capacity SSDs to build AI data centers and train large models.
The crucial question is not how much they invest, but whether these investments can generate profits. For example, Google Cloud's revenue increased by 63% in the first quarter of 2026, and its operating profit tripled, indicating that AI computing power has been converted into actual revenue, which will motivate cloud providers to continue purchasing memory. Conversely, if AI investments do not yield profits, they will reduce their purchases, leading to a decline in memory demand. The second quarter of 2026 is an important period for observation to determine whether AI computing power has turned into revenue or become excess capacity that cannot be utilized.
2. Prices Are Still Rising, but More Slowly, with a Possible Peak in the Fourth Quarter
Memory prices are still increasing, but the pace of growth has slowed significantly. For instance, contract prices for DRAM (ordinary memory) once approached 700% year-on-year but have since dropped to 100%-150%—meaning the price increase has shifted from a rapid rise to a more gradual one.
The market has already reacted to this change: memory-related stocks (such as Micron and Samsung Electronics) have fallen by 30%-40%. Why? Investors believe that the current high profits are cyclical and not permanent. Although the memory cycle used to last about two years, this cycle has been extended due to AI, the price peak is likely to occur in the fourth quarter of 2026, after which the growth rate will continue to decline.
3. Channels Are Starting to Stockpile, Which Could Be a Potential Problem
There have been changes in the memory industry's inventory structure: chip manufacturers (such as Samsung and SK Hynix) have very low inventories (only 2-3 weeks' worth of stock), while module manufacturers (which convert chips into memory modules) have increased their inventories to over 12 weeks.
There are two possible explanations for this: either module manufacturers are stocking up in anticipation of future price increases or shortages, or the actual demand from end-users (such as server and PC manufacturers) is not as strong, resulting in unsold inventory. If the latter is the case, module manufacturers will reduce their orders to chip manufacturers, which could weaken their bargaining power and lead to lower prices. Although inventory levels are not yet out of control, this "shortage at the upstream level and stocking up at the middle level" indicates that the negative effects of price increases are beginning to emerge.
4. Yangtze Memory Technologies' Expansion Will Determine NAND Supply
The future supply of NAND (flash memory, used in SSDs) depends largely on YMTC's expansion plans. YMTC is building two new factories, each capable of producing 100,000 wafers per month. If all five factories are used for NAND production, its global market share could reach 24%.
Whether there will be an oversupply depends on two factors: the growth in demand for AI SSDs (if it increases by 60% annually, demand will be high); and YMTC's production capacity (if it produces 310,000 wafers per month, supply will be relatively tight; if it produces 470,000 wafers per month, there may be a 9% oversupply). In short, as long as YMTC's expansion is not too rapid and AI demand continues to grow, there will not be an oversupply of NAND. Otherwise, supply will exceed demand.
Conclusion: The Cycle Will Not Collapse, but Expect a "Plateau at High Levels"
The memory cycle will not collapse suddenly; it is more likely to experience a gradual slowdown at high levels—prices will remain high, but the growth rate will gradually slow down. For investors, the peak in profits may not coincide with the peak in stock prices, as the market will anticipate the start of the next cycle. Ordinary consumers may not notice significant changes immediately, but corporate purchasing and industry investment patterns are already adjusting quietly.
(Note: All data in this analysis comes from third-party forecasts and is for reference only; it does not constitute investment advice.)