I think the investment direction for the next few years should not just focus on the word 'AI'. Many people only think of big models, chatbots, and application software when they see AI. But what can truly earn money in the long run is often not the busiest layer. But rather the infrastructure behind it. This picture actually makes it very clear: The next generation of trends may be included in these 7 types of ETFs. It's not about the names of ETFs, but about the industries behind them. The first line is the AI memory. DRAM、HBM、 High bandwidth memory will become increasingly important in the coming years. AI servers cannot do without GPUs, but GPUs also cannot do without memory. The larger the model, the more inference, and the more data, the stronger the demand for HBM and DRAM. So Samsung, SK Hynix, Micron are not just traditional storage companies. They are becoming key components of AI infrastructure. That's also why the storage sector fluctuates greatly, but the market can't escape it. The second line is electricity and nuclear power. Where did AI end up? Pinduoduo computing power. Where does the computing power end up? Electric splicing. A data center is not just about building a few servers, it relies on long-term, stable, and affordable electricity. So the return of nuclear power to the market's perspective is not accidental. Companies like Constellation, Vistra, and BWXT are essentially providing underlying energy for the AI era. The further AI moves forward, the more power will become a bottleneck. The third line is space economy and communication. This direction has not yet fully entered the public eye, but it has long been imaginative. Behind rockets, satellites, low orbit communications, and commercial aerospace are actually the next generation of global communication networks and data infrastructure. SpaceX, AST SpaceMobile, Rocket Lab and other companies represent the transition of commercial space from storytelling to industrialization. This direction will fluctuate greatly, but it is indeed a mainline for the next decade. The fourth line is a humanoid robot. Robots are not just about looking at the whole machine. It also depends on computing power, brain, sensors, and industrial automation. NVIDIA provides computing power, Tesla produces scenarios and complete machines, and ABB represents industrial automation. If AI is the brain, robots are the bodies of AI entering the real world. This line will definitely stir up concepts in the short term. But in the long run, it may be one of the most important landing scenarios for AI. The fifth line is advanced process equipment. EUV、 Photolithography and critical semiconductor equipment are the lifeblood of the entire advanced chip industry. Without ASML, there would be no state-of-the-art process. Without critical equipment, there can be no expansion of high-end chip production. So the equipment company may not look as sexy, but the moat is extremely deep. They are not selling stories, but tickets. As long as the advanced process continues to move forward, this line will be difficult to bypass. The sixth line is the core asset of semiconductors. If you don't want to bet on a single link, semiconductor ETFs like SMH are more like a basket of mainline representatives. Companies such as TSMC, Intel, and Qualcomm correspond to manufacturing, design, mobile computing, and the entire chip ecosystem. AI is not a single industry. It requires the cooperation of GPU, CPU, storage, packaging, OEM, and communication chips. So semiconductors will still be one of the most essential assets in the coming years. The seventh line is the power grid equipment. This is something that many people tend to overlook. AI data centers ultimately have to be connected to the power system. If the power grid is not upgraded, the power dispatch cannot keep up, and even with more GPUs, it cannot run. Companies such as NextEra, Eaton, and Schneider represent the upgrading of power infrastructure. The more AI develops, the more it will force the power grid, transmission and distribution, transformers, and energy storage to upgrade together. This is not the sexiest direction, but it is likely one of the most certain directions. So in the next few years, I think investment cannot just rely on who is shouting AI the loudest. More importantly, it depends on who is providing the real infrastructure for AI. Memory. Electricity. Semiconductor equipment. Chip core assets. robot Space communication. Power grid upgrade. These directions may seem scattered, but the bottom layer is actually a line: AI is moving from models to the real world. The next generation trend is not just about one AI application. But rather a reassessment of the entire AI infrastructure. What is truly worth watching is not who is the best at telling stories. But who is the most difficult to replace in the future industrial chain.