Phenomenon level financial report, profit surge, far exceeding expectations!! There was a big event in the storage circle early this morning, and Micron Technology released its second quarter financial report as of February this year. This data is really impressive - revenue surged by 197%, gross profit margin reached 74.4%, and core indicators exploded. Almost at the same time, Alibaba Cloud also took action, with AI computing power and storage products rising by up to 34%, for a straightforward reason: the number of token calls reached a historic high. It's not a coincidence that these two things happened on the same day. It actually points to a core judgment: AI is pushing the storage industry from the previous cyclical game of "ups and downs" to a new stage of "value reassessment". Let's take a look at Micron's financial report first. It is indeed a phenomenon driven by AI How to describe the performance in the second fiscal quarter? The word 'explosive' is not excessive: What's even more frightening is the forecast for next quarter - third quarter revenue is expected to be $33.5 billion, while analysts' expectations are only $24.3 billion; Earnings per share were $19.15, and gross profit margin continued to soar to around 81%. With such confidence, the board of directors directly announced a 30% increase in quarterly dividends. That's the question: Why is Micron so powerful? Just three words: AI takes flight. Firstly, HBM is in short supply. As AI chips sell more, high bandwidth memory naturally follows suit. The CEO of Micron has also made it clear that they are already mass producing HBM4 for Nvidia's next-generation Vera Rubin GPU, and this wave of AI dividends is directly coming in. Next is full flowering. Not only data centers, but also cloud storage business revenue of 7.75 billion US dollars, an increase of over 160%; Mobile and client business reached 7.71 billion, an increase of approximately 244%. This indicates that AI has not only ignited a single point, but also revitalized the entire storage system. There is also an implicit logic: structural supply constraints. The production capacity of large factories is squeezing towards HBM, while the supply of traditional DRAM and NAND is tight, naturally pushing up prices. Micron has become one of the biggest beneficiaries. However, there is also one thing worth noting: after the release of the financial report, the US stock market fell. Why? One reason is the disturbance of the situation in the Middle East, coupled with the fact that stock prices have just hit a new high, the market is prone to the mentality of "all the good news coming out"; Another core issue is that Micron's expansion plan has been released - capital expenditures for this fiscal year will be increased by 25% to $25 billion, and an additional $10 billion will be added in the next fiscal year, mainly for building factories in the United States. The market is a bit panicked, afraid that the expansion of production will be too strong and demand will not keep up, and it will return to the old path of "overcapacity price collapse". But in my personal opinion, this correction won't last long. After all, AI's demand for storage has not yet reached its ceiling. The price increase of Alibaba Cloud actually indirectly confirms this point. Behind the price increase of Alibaba Cloud is the real strength of the "Token Economy" The highest price increase this time is 34%, and the official reason is very direct: Token call volume has skyrocketed, and MaaS (Model as a Service) growth rate has reached a historic high. Many people may not have a concept of "token". Simply put, it is the smallest unit used by AI models to process text, such as asking a sentence, writing a summary, or generating a line of code, all of which consume tokens. Now AI is moving from the "training" stage to the "inference" and "application" explosion period, and token consumption is growing exponentially. For example, a simple conversation may have dozens of tokens, a complex AI task may have tens of thousands, and the consumption of multi-step, autonomous AI applications can even reach 20-30 times that of traditional conversations. I raised a viewpoint in December last year: no matter how fast the computing power grows, it cannot keep up with the speed of token consumption. At that time, a small company had tens of billions of tokens per day, and the daily consumption of large model tokens in the country exceeded 30 trillion. Now then? I checked the latest data and it has already reached 180 trillion yuan - a six fold increase in just a few months. The consumption of so many tokens is due to the massive amount of data that needs to be stored, read, and processed. Every time AI interacts, the storage chip must follow suit. This has two direct impacts: One is the explosive demand for inference computing power. Even if the growth rate of large-scale model training slows down in the future, the inference requirements brought by AI's landing in various industries will become a more sustainable growth engine for storage. Some institutions predict that by 2027, the proportion of China's reasoning and computing power will exceed 70%. The second is the redefinition of storage capacity. The data generated by AI is not only large, but also 'heavy', requiring larger capacity and higher performance storage to support it. This is not only beneficial for DRAM and NAND, but also for upstream and downstream enterprises such as SSDs and storage servers. The future of the storage industry has changed its logic Combining Micron data and Alibaba Cloud's actions, we can try drawing a few lines: The demand has changed: Previously, it was driven by the expansion of mobile phones and computers, but now the core driving force has completely shifted to AI. The memory requirement of an AI server is 8-10 times that of a regular server. Storage not only needs to be "large", but also "fast", and high-performance products such as HBM have become standard for AI chips. Whoever can capture the high-end segment holds the pricing power. Price transmission is occurring: the price of storage chips is rising and moving down the industry chain layer by layer. Upstream production capacity is concentrated in HBM, leading to tight supply and rising prices of traditional memory and flash memory; Cloud providers in China have started to adjust prices, and Alibaba Cloud's recent wave is an example, AWS、 Google Cloud is also taking action; Downstream mobile phone manufacturers are beginning to face pressure, with OPPO and Vivo raising prices and memory costs accounting for over 20% of the phone's BOM. To summarize briefly Looking at two things together, the conclusion is actually quite clear: the AI driven storage super cycle is on its way, but the endpoint is still far away. Short term (1-2 years): The HBM production capacity gap (expected to be 50% -60% in the industry) and the outbreak of tokens are combined, providing support for storage prices and continuing upward momentum in the industry chain. A-shares, Taiwan stocks, Japan and South Korea, as well as storage companies that have achieved performance, have generally gone against the trend and strengthened, and this trend is not over yet. In the medium to long term, the pressure of expanding production and slowing demand will come sooner or later, and the market will repeatedly tug between the "long logic of AI" and the "short game of cycles". The ultimate winners should be those companies that have technological barriers in high-end fields such as HBM and can control the pace of production capacity. Finally, let's talk about the judgment from a trading perspective: Upstream storage chip factories: Still the most directly benefiting direction from this round of AI dividends, with performance still in its explosive period. Leading technology companies such as Micron, Samsung, and Hynix are the trendsetters, and the supporting computing logic is also very smooth. But in two years, we have to keep an eye on the pace of expansion. Zhongyou Cloud Manufacturer: The price increase of Alibaba Cloud indicates that cloud manufacturers can pass down costs, but it also depends on whether they can retain customers. Downstream terminal manufacturers: Under the greatest pressure, the cost side is passive and the profit side is squeezed, so they can only rely on price increases to digest. The key is to see if the terminal can catch up. That's all, brothers figure it out for themselves.