I understand the demand for electricity from AI in China and the United States In my personal opinion, the core demand of AI for electricity in the United States is large-scale, stable, continuous, and accessible power supply. The power consumption intensity of AI data centers is much higher than that of traditional data centers. In the past, data centers mainly ran cloud computing, storage, web pages, and app services, which already consumed a lot of electricity. Now, AI data centers run GPU clusters, training models, inference services, video generation, and intelligent agent applications, all of which keep servers under high load for a long time. The more GPUs there are, the higher the power, the more difficult it is to dissipate heat, and the higher the power requirements for the entire computer room. The power consumption of a large AI data center can reach several hundred megawatts, and in the future, top projects may even approach the GW level. This concept is important because it is close to the electricity consumption of a small city. AI companies expanding their computing power may appear to be buying GPUs, building data centers, and deploying servers, but at the bottom, they are actually consuming more electricity. So AI has very special requirements for electricity. It requires stable electricity, continuous electricity, and electricity that can be locked in for a long time. Data centers need to operate 24 hours a day, not just during the day or when electricity prices are cheap. Training models requires stable computing power, and inference services must respond to user requests at all times. Once the power supply is unstable, it will affect server operation, customer service, cloud business, and overall delivery capabilities. From the perspective of energy types, nuclear power is the most suitable for the long-term needs of AI data centers. Nuclear power is stable, low-carbon, and has a high capacity factor. It can provide continuous power supply and is also suitable for large technology companies to sign long-term power purchase agreements. Companies such as Microsoft, Amazon, and Meta are willing to sign long-term contracts for nuclear power resources, essentially locking in a stable computing power base for the future in advance. But the problem with nuclear power is also very practical, as the construction period is too long. It takes many years for a new nuclear power plant to be approved, constructed, and connected to the grid. So in the short term, it is difficult for the US AI power gap to be completely solved by new nuclear power generation. The most likely power source that can be replenished faster in the coming years is natural gas. The construction period of natural gas power plants is relatively shorter, with stronger peak shaving capabilities, which can meet the power consumption growth of data centers faster. So, the structure of AI electricity consumption in the United States is likely to be nuclear power providing long-term stable base load, natural gas undertaking short - to medium-term power replenishment and peak shaving, wind power, solar energy, and energy storage as supplements, helping to reduce comprehensive electricity prices and meet green energy goals. The real power shortage in the United States is still on the grid side. Many regions do not have complete power generation capacity, and the problem lies in the inability to deliver electricity to data centers. There are not enough transmission lines, transformers, substations, grid connected capacity, and long approval and construction periods. AI companies can quickly purchase GPUs, build data centers, and deploy servers, but power grid expansion cannot be completed quarterly. Therefore, my personal judgment is that in the next few years, the demand for AI powered electricity in the United States will drive the development of three types of resources: The first category is stable power sources, including nuclear power, natural gas, and some hydropower. AI data centers require long-term stable power supply, and the value of stable power supply will continue to increase. The second category is power grid infrastructure, including transmission lines, transformers, switchgear, substations, and power engineering. Whether electricity can be delivered to the data center will directly determine whether the data center can be deployed. The third category is the internal power system of data centers, including America, power distribution systems, liquid cooling, heat dissipation, and energy management. The higher the GPU power, the greater the power supply and heat dissipation pressure inside the data center, and the importance of these infrastructure will continue to increase. Therefore, I believe that in the coming years, the AI power mainline in the United States will continue to drive up the value of stable power sources, grid equipment, power engineering, and data center infrastructure. The more intense the competition in AI, the higher the strategic value of power resources. Bitget is here, VIP! Crypto、 US shares CFD， Global Advantage One Stop Layout