The market has actually overreacted to the issue of Meta renting computing power. Many people's first reaction when they see "rental computing power" is: Is the demand for AI no longer sufficient? Did you buy too many GPUs? Is the data center built too much? But from the perspective of Wall Street, this matter is not that simple. UBS's view is more direct: Meta may rent out some idle computing power or AI model services, which may bring new sources of income in the short term and even improve EPS. Morgan Stanley has also calculated that if Meta leases 250MW of computing power for one year at a rate of $40 per watt, EPS may increase by approximately $2.97 by 2028, which is roughly equivalent to an 8% increase. To put it simply, this may not necessarily be a bad thing. If the computing power is not used up in the short term, renting it out to make money is essentially improving asset utilization. What really needs to be distinguished is: Is this a 'collapse of demand' or a 'phased realization'? I am more inclined towards the latter. Because Meta rents out some computing power, it does not mean that the global trend of AI data center expansion has changed. As mentioned in the picture, Meta currently has a total global data center capacity of about 20GW, and plans to add about 14GW in the future. There are different uses, different generations of GPUs, and different leasing resources, which cannot be simply understood as all of them being available for rent. This is the key. The market's concern this time is not about Meta earning extra rent, but about AI infrastructure entering the stage of "calculating ROI" from "crazy expansion". Previously, the market only asked: Will AI companies still buy GPUs? Now the market is starting to ask: When will I make money after buying so many GPUs? So it's not surprising that storage, semiconductor, and AI hardware will be hit in the short term. Because it rose too quickly earlier, there were too many optimistic expectations embedded in the valuation. As long as there is a hint of 'possible surplus computing power', funds will run first. But in the long run, I don't think this news can directly overturn the trend of AI infrastructure. AI training, inference, and data center expansion still require GPU, HBM, memory, SSD, and power. What really matters is not how much computing power Meta rents out. But rather the following questions: Has cloud vendor CapEx continued to upgrade? Has the HBM price loosened? Has there been a demand for AI inference? Is the utilization rate of computing power in large factories really decreasing? If these haven't changed, then this time it's more like emotions killing valuation. Renting computing power from Meta does not necessarily indicate a collapse in AI demand. It is more like reminding the market that the AI industry cannot just tell stories, but also start making profits when connected.