In the past two years, my own layout on AI has been focused on electricity. Although the returns on electricity related assets are not as exaggerated as storage, chips, and servers, I always believe that electricity is the most fundamental, underlying, and irreplaceable demand for AI, because the core of all computing power ultimately returns to electricity. The United States is precisely one of the countries with the fastest development of AI and the most obvious demand for electricity. Data centers are not for ordinary residential use, but require high load electricity that can operate stably 24 hours a day. They require fast grid connection, substations, large transformers, gas turbines, nuclear power, energy storage, backup power sources, and cooling systems. The problem is that AI data centers can be built in 18 to 36 months, but the construction period for transmission lines, substations, large transformers, gas turbines, and dispatchable power sources is often 4 to 8 years, or even longer. That is to say, the capital expenditure speed of AI has exceeded the physical expansion speed of the US power system. (This data has been included in the BOA data shared in the past) If AI data centers become the most certain increase in electricity consumption in the next decade, then companies with power generation assets, transmission assets, substation equipment, grid connection capabilities, on-site power supply capabilities, and electrical engineering capabilities will no longer be just traditional utilities, but a part of AI infrastructure. AI is not just buying GPUs, AI is rebuilding an energy system centered around computing power. A 100MW AI data center is not actually a computer room, but a small power system. We need upstream power generation, midstream power transmission, downstream substations, and park distribution, America，PDU， Busbars, backup generators, liquid cooling, and extensive electrical construction. It can even be further imagined that global power shortages may become a consensus in the future, as AI will not only occur in the United States. Europe is developing sovereign AI, the Middle East is developing AI data centers, China is also developing computing infrastructure, and Southeast Asia is undertaking data center migration. More importantly, training the model is only a temporary requirement, while the inference requirement is the long-term load. If every search, video generation, autonomous driving judgment, robot execution, and enterprise AI agent call in the future correspond to real computing power consumption, then AI's demand for electricity will not be a one-time construction demand, but a long-term, sustained, and rigid energy demand. So I believe that the end of computing power is electricity. In the past two years, the main focus of AI trading has been GPU, HBM, servers, and optical modules. In the next few years, the main focus of AI trading is likely to gradually shift to electricity. Power generation, transmission, transformation, distribution, energy storage, liquid cooling, backup power sources, grid connected projects, and on-site energy systems. Bitget is here, VIP! Crypto、 US shares CFD， Global Advantage One Stop Layout