Over the past year or so, the "memory buying war" triggered by generative AI and large model training has continued in the semiconductor industry. The contract prices of non-volatile NAND flash memory and DRAM memory have experienced a rare surge, benefiting upstream manufacturers and investors, but putting heavy cost pressure on downstream machine manufacturers and end consumers. According to Bloomberg citing industry data, NAND contract prices have increased by more than 600% since September 2025, and DRAM contract prices have also increased by nearly 400%. The market generally expects that this AI-driven price increase will be difficult to alleviate in the short term.
Many research institutions believe that this round of "memory shortage" is greater than expected at the beginning of the year and will last longer. Michael Brown, senior research strategist at Pepperstone Group Ltd., pointed out that there are current signs that supply and demand tensions are intensifying, and some sources close to the industry chain even judge that shortages may continue all the way to 2030 or even beyond. Against the background of the continuous expansion of high-performance computing and training clusters, a large number of AI startups and cloud service providers are "hoarding" memory chips, which has absorbed almost all the production capacity that manufacturers can release. Traditional consumer electronics manufacturers can only compete with each other in the remaining supply.
On the profit side, memory manufacturers, investors and employees have become direct beneficiaries. Driven by both price and demand, storage companies' profits hit record highs, and the stock prices of many companies strengthened, driving the outstanding performance of related assets. Some companies even distribute the dividends brought by the AI wave directly to employees by issuing high bonuses. For example, Samsung and SK Hynix have both issued considerable rewards to employees due to surges in performance. JP Morgan strategists pointed out in the latest research report that as long as the demand story led by AI remains unchanged, there is still room for storage prices to continue to rise.
In sharp contrast are downstream machine manufacturers that passively bear cost pressure. Smartphones, PCs, game consoles, and various consumer electronics products are highly dependent on storage devices such as NAND and DRAM. However, they are now faced with the dilemma of continued tight inventory and sharp rise in component prices. In order to cope with the soaring costs, some manufacturers have to increase the selling price of the whole machine, compress profit margins, or make compromises on new product specifications, such as reducing memory and storage capacity, in exchange for more controllable material costs. There are also companies that find it difficult to maintain operations in the cracks and have no choice but to withdraw from the market or close business lines.
For ordinary consumers, this round of storage price increases is ultimately reflected in higher terminal prices for electronic products and longer replacement cycles. When the prices of consoles, game consoles, laptops and other devices have increased by hundreds of dollars compared to the initial launch period, more and more users are choosing to delay upgrading or simply give up on purchasing the latest devices and games. For example, game console manufacturers have previously announced increases in console prices in some regions due to cost pressures. Such price increases may spread further in the context of a new round of soaring storage prices.
In the view of industry observers, the current situation is an "overrebound" after the last round of cyclical downturns in the storage industry: In the early days of the AI outbreak, a large number of manufacturers reduced production and scaled back capital expenditures due to weak demand. Now, however, it is difficult to fully release new production capacity in a short period of time to match the surge in demand for AI clusters. Coupled with the structural shortage of specialized storage products such as high-end HBM and GDDR, general-purpose NAND and DRAM have also been squeezed into a more tight supply pattern. In the absence of new production lines and the need for time for technology conversion, rising prices have become the "new normal" that the entire industry chain has passively accepted.
In the medium to long term, the industry generally believes that AI's demand for storage will continue to run at a high level, compressing the cyclical downside space, but it may also bring higher volatility. If large-scale new production capacity is put into operation in the next few years, or the pace of AI investment changes, there is also a risk of a correction in the current "skyrocketing" prices. Until then, chip manufacturers and capital markets may still enjoy high price dividends, while equipment manufacturers and end consumers will need to make a more difficult trade-off between cost and performance.