The market value of Micron, a major storage company, has exceeded $1 trillion, soaring eightfold in one year with the help of AI.

Another company has entered the trillion - dollar club thanks to AI.

On May 26, 2026, Micron's stock price soared by 19% in a single day, and its market value exceeded $1 trillion for the first time. This figure was almost unimaginable three years ago. At that time, Micron was a typical cyclical semiconductor company. As the prices of DRAM and NAND fluctuated, its stock price was like a roller - coaster. Analysts would come out every now and then to say that "the memory industry has hit the bottom."

But if you had held Micron's stock since the beginning of this year, you would have probably earned 150% by now. In the past 12 months, the company's stock price has risen more than 8 times.

What drives all this is not a technological revolution but a very simple logic: AI models are getting larger, and there are more and more GPUs. And next to each GPU, an HBM memory is needed; otherwise, the computing power cannot be fed in, and it's all a waste.

For NVIDIA's H100, H200, and B200, the chips that make Jensen Huang excited when he gives a speech on stage, there are multiple HBM chips stacked next to each one. Without HBM, a GPU is like a super - car with a small engine, unable to unleash its computing power.

It took the market a full two years to really understand this principle.

01

The "Pig Cycle" of HBM

On May 22, Micron CEO Sanjay Mehrotra said something that sent shivers down many people's spines: The company's factory in Virginia is expanding production, but the gap between production capacity and demand "cannot be filled in the short term."

Three days later, on May 23, at the TMC industry conference, Micron further provided a clearer timeline - the supply shortage in the memory market will last until after 2026 and is likely to extend into 2027 or even longer.

The entire production capacity of HBM chips in 2026 has been sold out, and new orders are already queuing up for 2027.

This is not a company boasting about itself but a real industrial dilemma. The manufacturing process of HBM is extremely complex. Multiple layers of DRAM chips need to be vertically stacked through "Through - Silicon Via" (TSV) technology, with extremely high yield requirements. The capacity expansion cycle often takes 18 to 24 months. If you decide to build a new HBM production line today, the products will not be available until the end of 2027 at the earliest.

There are only three companies in the world that can produce HBM: Samsung, SK Hynix, and Micron. And in the most cutting - edge HBM4 and HBM4E products in terms of technological iteration, the progress of the three companies is not synchronized. Thanks to its early bets, SK Hynix is still NVIDIA's largest HBM supplier at present, but Micron is catching up quickly and has production capacity in the United States, which is an increasingly important variable in the current geopolitically sensitive situation.

Micron's Q2 financial report confirms all this: Revenue increased by 196% year - on - year, and the gross profit margin was as high as 74.9%. Memory chips used to be a well - known "pig - cycle" commodity in the semiconductor industry, but now it has become a business with long - term contracts, scarcity premiums, and pricing power.

02

Local Dividends

In the past few years, the narrative of the chip competition has been almost monopolized by NVIDIA. Jensen Huang's leather jacket, the in - short - supply GPUs, and the rapidly rising computing power prices - these have formed the most prominent pictures of the AI infrastructure war.

But the real battlefield has never had only one bottleneck.

Computing power, memory, interconnection, and heat dissipation are the four basic pillars of an AI data center, and none of them can be missing. The shortage of GPUs has been discussed repeatedly, but the shortage of HBM is more hidden because it is not directly targeted at consumers, and most people can't even see it. It wasn't until one day when a large cloud provider began to complain externally that "the GPUs have arrived, but the supporting memory is still waiting" that this bottleneck began to surface.

UBS has directly raised Micron's target price from $535 to $1,625 today. This target price has become the highest among the 46 covering analysts, and the magnitude is so large that it has even surprised Wall Street. But if you understand the supply - demand logic of HBM, this figure may not be a pipe dream.

A more interesting perspective is geopolitics. The United States is constantly tightening its export controls on high - end GPUs, and Chinese chip companies are severely restricted in the field of logic chips. However, the situation in the memory chip field is also delicate. Yangtze Memory Technologies Co., Ltd. (CXMT) is quickly catching up in standard DRAM, but in the HBM track with extremely high technological thresholds, Chinese enterprises are almost at the starting line. This means that whoever controls the HBM production capacity controls the rhythm of the global AI infrastructure to some extent.

As the only HBM manufacturer headquartered in the United States, Micron is enjoying the double dividends brought by this logic: on the one hand, there is an explosive growth in AI demand; on the other hand, there is the policy favor of ally production and local priority. The expansion of the Virginia factory is both a business decision and a political stance.

03

Is Trillion Just a Starting Point?

Historically, it has been extremely rare for a chip company to reach a trillion - dollar market value. Before Micron, only NVIDIA, TSMC, and Broadcom had crossed this line one after another. Micron's arrival means to some extent that the market's definition of "AI infrastructure" is expanding - it is no longer just GPUs but every irreplaceable link in the entire supply chain.

Of course, there is an unavoidable risk. The "pig cycle" in the memory industry has not disappeared; it has just been temporarily covered up by AI demand. If the AI investment boom cools down even a little, if the marginal benefit of large - model training begins to decline, or if the growth rate of memory demand on the inference side slows down - Micron's valuation logic will be re - examined. After all, a 74.9% gross profit margin won't last forever.

But at least today, that little memory chip hidden behind the GPU is enjoying its moment in the spotlight.

The war of the AI arms race is far from over. And what really determines the outcome may never be the most dazzling bullet but the consumable that runs out first and no one notices.

This article is from the WeChat official account "GeekPark" (ID: geekpark), author: Hualin Wuwang. It is published by 36Kr with authorization.