After reading about the expenses of the six major companies, what I want to say is that they are not burning money again. But rather, the market's judgment of AI is shifting from one sentence to another: The previous question was: Will there be too much capital expenditure on AI? ” The question now is: If you don't continue to invest, will your computing power be insufficient This is the recent storage HBM、 The core reason for the strong rebound of semiconductors. It's obvious from just a few changes. Google has raised its Capex guidance for 2026 from $180-190 billion to $195-205 billion. At the beginning of the year, Meta initially only offered $115-135 billion, but later continuously raised it, and now it has reached $130-145 billion. Intel continues to increase its advanced process, foundry, and packaging capacity. Tesla is also continuing to invest in Dojo computing power, robot production lines, and factory expansion. Although Microsoft has slightly lowered its annual Capex from 190 billion US dollars to 175 billion US dollars due to accounting standards, the key is not that it has stopped investing, but that the construction of AI hardware and computing power has not shrunk, and the overall investment scale is still much higher than last year. This indicates a problem: The AI capital expenditure cycle is still being forcefully pushed forward. And this round of investment is not just about buying a few GPUs. Behind an AI data center, what is needed is a complete industrial chain: GPU/ASIC chip HBM high bandwidth memory Server DRAM Enterprise grade SSD Advanced Packaging optical module electrical equipment Liquid cooling Data center land and power grid So why has the storage sector rebounded so strongly recently? Because the market suddenly reacted: If cloud providers continue to expand their AI data centers, it will be difficult for storage demand to quickly shut down. Micron, Hynix, Samsung and other companies are not just traditional DRAM cycle stocks. They are now more like 'capacity sellers' in the AI infrastructure chain. GPU is responsible for computing. HBM is responsible for feeding data. DRAM is responsible for operation. SSD is responsible for storage. The larger the data center, the more exaggerated the storage consumption. In the past, the storage industry looked at mobile phones PC、 Inventory cycle. It's different now. Now we need to see: Google Cloud Growth; Microsoft Azure growth; Meta data center expansion; Tesla robots and autonomous driving; AI inference quantity; HBM supply-demand gap; Enterprise SSD prices. That's also why once storage stocks rebound, they can be very aggressive. Because the market previously suppressed two expectations: Firstly, will AI Capex reach its peak; Secondly, will the storage price increase come to an end. Now with the release of financial reports and guidance from major companies, the market has found that: Money is still being spent, demand is still there, and expansion has not stopped yet. The valuation of the storage that was killed earlier will be quickly repaired. But brothers should also remain calm. A strong rebound does not mean there is no risk. The biggest problem with the AI hardware chain now is not that there is no story, but that expectations are too high. Once cloud providers slow down Capex or storage prices remain stagnant, the stock price will still be heavily affected. So when I look at this line, the key is not whether to chase or not. The key is to look at three things: Has Capex, a major manufacturer, continued to undergo repairs; Has the price of HBM/DRAM/SSD continued to improve; Can cloud business revenue prove that this money is not wasted. If all three are still present, the storage mainline has not been falsified. If one of them starts to weaken, the position needs to be reassessed. This round of storage rebound may seem like a stock price recovery, but behind it is actually the global AI capital expenditure cycle that has not yet ended. Short term prices depend on emotions. Mid term prices are expected to increase. The long-term price depends on whether AI data centers can continue to expand.