Innerhalb eines Jahres stieg der Kurs um das Achtfache. Mit Unterstützung von KI hat der Speicherriese Micron einen Marktwert von über einer Billion US-Dollar erreicht.

Another company has entered the club of companies with a market capitalization of over one billion US dollars thanks to AI.

On May 26, 2026, Micron's stock price jumped by 19% within a single day, and its market value exceeded one billion US dollars for the first time. This value would have been almost unimaginable three years ago – back then, Micron was still a typical cyclical semiconductor company. With the fluctuations in the prices of DRAM and NAND, the stock price fluctuated like a rollercoaster, and analysts constantly said: "The memory industry has reached its lowest point."

If you had held Micron stocks since the beginning of this year, you would probably have achieved a profit of 150% by now. In the last 12 months, the stock price of this company has increased by more than eight times.

The driving force behind all this is not a technological revolution, but a very simple logic: AI models are getting bigger and bigger, there are more and more GPUs, and next to each GPU, an HBM memory module is needed; otherwise, the computing power will not be fully utilized and everything will be wasted.

In the case of the Nvidia H100, H200, and B200, the chips that Huang Renxun is so enthusiastic about that he talks about them on stage, several HBM chips are arranged next to each chip. Without HBM, a GPU is like a super sports car with a small engine, and the computing power cannot be fully utilized.

It took the markets a full two years to really understand this situation.

01

The "Pig Cycle" of HBM

In an interview on May 22, Micron CEO Sanjay Mehrotra said something that sent shivers down many people's spines: Micron's factory in Virginia is expanding production, but the gap between capacity and demand "cannot be closed in the near future."

Three days later, on May 23, Micron provided a more detailed timeline at the TMC industry conference – the tight supply in the memory market will last until after 2026 and will probably extend until 2027 or even longer.

The entire production capacity of HBM chips for the year 2026 is already sold out, and new orders are being queued for 2027.

This is not a company's self - praise, but a real problem in the industry. The manufacturing process of HBM is extremely complex. It requires the vertical stacking of multiple DRAM chips using the "Through - Silicon Via" (TSV) technology. The yield requirements are very high, and the period for capacity expansion often takes 18 to 24 months. If you decide today to build a new HBM production line, the products can be delivered at the earliest by the end of 2027.

There are only three companies in the world that can produce HBM: Samsung, SK Hynix, and Micron. In the case of the state - of - the - art HBM4 and HBM4E products, the progress of the three companies is not synchronized. SK Hynix is still the largest HBM supplier to Nvidia due to its early investments, but Micron is quickly catching up and has production capacity in the United States, which is becoming an increasingly important variable given the increasingly sensitive geopolitical situation.

Micron's Q2 results confirm all this: Revenue increased by 196% compared to the previous year, and the profit margin was an impressive 74.9%. Memory chips were once known for their "pig cycle" in the semiconductor industry, but today they have become a business with long - term contracts, tight supply, and pricing power.

02

The Advantage of Domestic Production

In recent years, the narrative about the chip competition has been almost completely dominated by Nvidia. Huang Renxun's leather jacket, the scarce GPUs, and the rapidly rising prices of computing power – all these formed the most prominent images of the war for AI infrastructure.

But the real battlefield has never had only one bottleneck.

Computing power, memory, networking, and cooling are the four fundamental pillars of an AI data center and are by no means lacking. The shortage of GPUs is constantly being discussed, but the shortage of HBM is less visible because it doesn't go directly to consumers and most people don't even see it. Only when a large cloud provider began to complain: "The GPUs are here, but the suitable memory is still waiting," did this bottleneck become visible.

UBS has now directly raised the target price for Micron from $535 to $1,625. This target price is the highest among 46 analyzing analysts, and the magnitude of the increase has also somewhat surprised Wall Street. However, if you understand the supply - and - demand logic of HBM, this figure may not be so unbelievable.

An even more interesting perspective is geopolitics. The United States is constantly tightening export controls on high - end GPUs, and Chinese chip companies are severely hampered in the logic - chip industry. But the situation in the memory - chip industry is also subtle. Yangtze Memory Technologies Co., Ltd. (YMTC) is quickly falling behind in the production of standard DRAM, but in the HBM industry with its high technological threshold, Chinese companies are almost just starting out. This means that whoever controls the HBM production capacity controls to some extent the rhythm of the global AI infrastructure.

Micron, the only HBM manufacturer based in the United States, benefits from this logic and enjoys double advantages: On the one hand, there is a sudden growth in AI demand; on the other hand, there is political support for production in the United States. The capacity expansion of the factory in Virginia is both a business decision and a political stance.

03

Is One Billion US Dollars Just the Beginning?

In history, it was extremely rare for a chip company to reach a market capitalization of one billion US dollars. Before Micron, only Nvidia, TSMC, and Broadcom had successively crossed this threshold. Micron's reaching this threshold means to some extent that the market is expanding the definition of "AI infrastructure" – it's not just the GPU, but also every indispensable part of the entire supply chain.

Of course, there is an unavoidable risk here. The "pig cycle" of the memory industry has not disappeared, but is only temporarily masked by AI demand. If the AI investment boom cools down even a little, if the marginal benefits of large - model training start to decline, if the demand for memory on the inference side grows more slowly – then Micron's valuation logic will be re - examined. After all, the profit margin of 74.9% will not stay that way forever.

But at least today, the small memory chip behind the GPU is having its moment of glory.

The AI arms - race war is far from over. And what really decides the outcome may never be the most prominent bullet, but the consumable part that no one notices but that runs out first.

This article is from the WeChat account “GeekPark” (ID: geekpark), author: Hualin Wuwang. Published by 36Kr with permission.