AI's Memory Crunch: Why Storage Chips Are Back in the Spotlight
Every AI model, whether it answers questions or generates images, spends much of its life waiting for data. The chips that hold and feed that data — memory and storage semiconductors — are suddenly the center of attention in the global technology supply chain, and nowhere more so than in China’s stock market.
On September 29, storage chip shares on China’s A-share market rebounded strongly after a sharp sell-off the previous session. According to data from financial terminal iFind, the sector attracted 1.93 billion yuan (roughly $270 million) in net inflows from large institutional buyers in a single trading day, while major Chinese indices closed higher across the board.
The reason is straightforward: artificial intelligence has turned memory into a performance bottleneck. Training and running large models requires moving enormous volumes of data between processors and memory at high speed. If the memory cannot keep up, expensive AI accelerators sit idle. That has pushed demand for advanced technologies such as DDR5, the newest mainstream generation of dynamic random-access memory; HBM, or high-bandwidth memory, which stacks chips vertically to feed AI processors; and 3D NAND, a technique for packing more storage into the same physical space.
Industry analysts argue that this is not a short-lived spike. As data centers expand and AI computing demand keeps being released, memory is becoming a decisive factor in how efficiently computing power is used. The technology upgrades themselves also widen the market: each generation of HBM, for instance, carries a far higher price tag than the standard memory it replaces.
Supply, meanwhile, cannot respond quickly. Chipmakers have been shifting capacity toward higher-value products, and building new fabrication plants takes years rather than months. That combination — rising demand, limited new supply, and a product mix tilting toward premium chips — is why institutional investors believe storage chip prices will stay supported and suppliers will gain more bargaining power.
The pattern echoes earlier cycles in the semiconductor industry, when memory prices swung wildly between boom and bust. What is different this time, analysts say, is the scale of AI infrastructure spending by cloud providers and the fact that high-bandwidth memory is genuinely difficult to manufacture, with only a handful of companies worldwide able to produce it at volume.
For China, the stakes are high. The country is the world’s largest buyer of memory chips but still depends heavily on imports for the most advanced products. Domestic manufacturers have been investing in NAND and DRAM production for years, and the current boom gives them both a market opening and a reminder of how far they still have to go in the most demanding segments.
Investors are watching two things closely: whether AI server demand keeps growing at its current pace, and whether new capacity arrives faster than expected. For now, the mood in the market is optimistic. Memory, once seen as a commodity business prone to brutal price wars, is being repriced as a strategic component of the AI era — and the companies that make it are enjoying renewed attention from both customers and shareholders.