AI infrastructure is shifting focus to applications, but this does not mean that the investment value of the computing power sector has disappeared; rather, it is a natural evolution of industry development. Currently, computing power remains the foundation of AI development, but market attention is shifting from "whether there is computing power" to "how efficiently computing power can be converted into actual value." The computing power sector still looks promising, but the logic has changed. In the past, it was a "arms race" style investment, pursuing the scale of computing power; now, it emphasizes effective computing power—that is, computing power that can be practically used to generate revenue. This means that the model of simply stacking computing power is unsustainable, and companies with domestic substitution capabilities, high utilization rates, low energy consumption, and deep integration with domestic ecosystems are more competitive. For example, domestic AI chips, liquid cooling technology, high-speed optical modules, and other areas, because they directly serve the construction of intelligent computing centers and improve energy efficiency, still have strong performance support. Meanwhile, the application and inference sectors are experiencing a burst of growth. As the cost of large model inference continues to decline and model capabilities become more practical, AI is moving from "usable" to "user-friendly." Enterprises are accelerating AI deployment in scenarios such as office work, marketing, industrial quality inspection, and financial risk control, while the penetration rate of AI smartphones, AI glasses, and other terminals in consumer markets is rapidly increasing. Subscription and paid models are gradually maturing. These applications not only directly generate revenue but also serve as the "engines" driving continuous consumption of computing power, forming a positive cycle of "application feeding back to computing power." Therefore, the current investment strategy should be: balance between computing power and applications, with a gradual shift of focus toward applications. Computing power is like "selling shovels"—demand is certain but growth may slow down; applications are like "prospectors"—though riskier, if a hit product emerges, the potential returns are huge. For investors, it is advisable to prioritize leading application companies that have already demonstrated performance, while also paying attention to core links in computing power with technological barriers and ecological advantages.