Who has the real bottleneck? Who can turn bottlenecks into profits, who will gain control points in the next generation architecture migration, and whose valuations have already been overdrawn? These are the four most important issues in the new stage of investment logic for the AI computing power industry chain. The core logic is that after all links in an industry chain have been hyped up, we cannot just scratch the surface. Instead, we need to truly distinguish which companies are truly irreplaceable and have strong core competitiveness in the industry chain. The market's expectations for them have already been supported by a large number of orders and unexpected deliveries. The essence is from sprinkling pepper to carefully screening. In other words, in the past two years, one can buy 'AI data center relevance'; Going forward, we need to buy scarce cash flow, structural control, and verifiable order continuity. Two key issues Question 1: Is it a 'real bottleneck' or 'theme related'? The characteristics of the real bottleneck are: extended delivery time, customers' willingness to make advance payments or sign long-term agreements, rising ASP, expansion of gross profit margin, and slow expansion of production. The characteristics related to the theme are: increased revenue but no increase in gross profit, highly concentrated customers, many capex but unclear ROIC, or only "entering the supply chain" but with a small share. The current bottlenecks that are closer to reality are HBM, high-end server DRAM/eSSD, advanced packaging, transformers/switchgear/grid equipment, and some high-end optical interconnects. Question 2: Will its value increase during architecture upgrades or be replaced by architecture upgrades? This is the most critical issue for optical interconnection, 800V DC, and CPU. CPO will compress the value of some traditional plug-in modules, but will increase the value of silicon optics, optical engines, lasers, connectors, packaging testing, and switch ASICs. 800V DC will lift the high-voltage power supply chain SiC/GaN、 The value of protective devices and rack power architecture, but may compress some traditional low-voltage power supply components. The most dangerous investment is: Just because a certain company is involved in AI data centers, buy it; Just because it mentions CPO/800V/liquid cooling, give high multiples; Only focus on revenue growth without considering gross profit, market share, customer concentration, and structural substitution risks. The most attractive investment is: The bottleneck asset that has already seen an increase in performance, but the market underestimates its sustainability; Or control point assets that have not yet fully realized their performance but have been locked in by the next-generation architecture platform. Simply put, the market has recognized the bottleneck and bottleneck of the AI computing power production chain, so investment needs to go further: it's not just about standing at the bottleneck, not just at the bottleneck link, you also need to see if this enterprise is unique, irreplaceable, technically correct at this bottleneck and bottleneck node, not only delivering a large number of orders but also continuously exceeding expectations. At the same time, it also depends on the pace of technological evolution and implementation? This article is sponsored by @ bitget_zh, titled 'Bitget Buying US Stocks: Instant Entry, Smooth Trading'