Recently, the memory chip circle is staging a crazy "rush to buy".

The demand for memory chips in AI is expanding wildly, and the expansion rate has far exceeded the supply capacity of the entire industry.

On Monday, storage bull stocks Micron Technology and SanDisk bucked the trend, and their stock prices hit record highs again. According to the current trend, the chips consumed by the AI business this year are expected to account for more than half of the total global memory chip market.

To make matters worse, industry experts generally believe that this serious shortage problem is very tricky and may be difficult to truly mitigate until 2028.

Memory has become a strategic asset for AI, and Micron's market value and performance have both exploded

Micron Technology's market capitalization is soaring. Last Friday, it broke through the $600 billion mark for the first time. You know, just a month and a half ago, its market value just stood at 500 billion; It only took 7 months to rise from 200 billion to 500 billion.

Micron Technology CEO Sanjay Mehrotra bluntly said in a recent interview that the AI industry is still only in the "first game" (that is, the initial stage). He explained that as AI applications become more popular, the demand for data computing (tokens) will continue to skyrocket, and this is supported by faster and larger memory.

In his words, "memory has become the most critical strategic asset in the hands of customers." The biggest problem facing the industry at present is not that no one buys or the price cannot be sold, but that the production capacity is too tight and cannot be expanded in the short term.

Micron's latest results (Q2 fiscal year 2026) also perfectly confirm this statement: the company's revenue, gross margin, earnings per share and free cash flow have all hit record highs, and it is widely predicted that the next quarter (Q3) will continue to set new records.

Supply gap: It is difficult to wait for an inflection point before 2028

Micron pointed out that at present, whether it is traditional servers or AI servers, the market demand is very strong, but everyone is stuck in the same problem - the supply of DRAM (memory) and NAND (flash memory) is too tight.

The Financial Times quoted analysts as saying that because new fabs require a long construction cycle, this serious shortage may be difficult to alleviate until 2028.

South Korean media reports give more specific predictions: the industry will face greater pressure as mass production of 12-layer HBM4 memory for Nvidia's next-generation AI graphics cards (such as Vera Rubin) begins. Conservatively estimated, by 2027, the entire market may only be able to meet about 60% of memory needs. This serious imbalance between supply and demand, although it gives memory manufacturers absolute pricing power, is also a double-edged sword - if the manufacturer's mass production progress cannot keep up, it will also face huge market pressure.

On the demand side, the thrust is still accelerating. Emerging AI workloads (Agentic AI) are driving server memory specifications to expand to 400GB, which is about 4 times the current level.

At the same time, both Nvidia and AMD's next-generation AI graphics cards will be equipped with HBM4 memory with larger capacity. In addition, with the advantage of low power consumption, LPDDR memory is gradually becoming the preferred solution for large-scale AI deployment.

HBM4 has been supplied, and HBM4E will be mass-produced next year

In the face of an extremely tight supply and demand environment, Micron is accelerating the next generation layout on multiple product lines.

In terms of high-end AI memory, Micron has already supplied 36GB of HBM4 memory to NVIDIA's next-generation platform (Vera Rubin), and the existing HBM3 process yield is also very mature. The more advanced next-generation HBM4E memory is scheduled to begin mass production next year.

In other areas, Micron has also made frequent actions:

Low-Power Memory (LPDDR): Recently, a new SOCAMM2 solution has been launched, with a single capacity of up to 256GB and a maximum capacity of 2TB.

DDR5 memory: Nvidia's Groq 3 LPX is being supplied, which can support up to an astonishing 12TB capacity on a single chip.

This tight supply situation has also been transmitted to our daily consumer market. Micron predicts that due to insufficient production capacity and rising prices, shipments of PC computers and smartphones may decline slightly this year (low double-digit decline). However, a new trend worth paying attention to is that in order to run local AI applications smoothly on computers, 32GB of memory is gradually becoming the "standard" of new computers.

AI storage demand has entered a "super cycle"

AI's consumption of memory chips has evolved from "short-term themes" to "long-term structural growth". As AI moves from training to inference, and the popularity of Agentic AI, the demand for large-capacity, high-bandwidth memory (such as HBM4, DDR5) will continue to grow exponentially.

Due to the long construction cycle of fabs and slow capacity expansion, the severe shortage of memory chips is expected to continue until 2028. This means that storage manufacturers will have strong pricing power and bargaining power in the next few years.

While high costs may dampen PC and mobile phone shipments in the short term, the demand for AI localization deployment is forcing hardware upgrades. 32GB of RAM has become standard on PCs, and the proportion of flagship phones equipped with more than 12GB of DRAM has soared, which will push up the average selling price and profit margins of the entire industry.

Opportunity 1: Leading memory chip manufacturers
In the context of a serious imbalance between supply and demand, memory chip giants (such as Micron, SK Hynix, Samsung Electronics) with advanced process capacity and stable supply of HBM4 and high-end DDR5 will directly benefit from the rise in volume and price, and their performance is expected to continue to explode

Opportunity 2: Upstream and downstream enterprises in
the storage industry chain With the surge in demand for memory chips, upstream semiconductor equipment and material suppliers, as well as downstream module packaging and testing companies, will also usher in huge market opportunities. At the same time, domestic storage manufacturers are also expected to accelerate the process of domestic substitution under the global supply gap