In the morning, I saw information about BlackRock's acquisition of American power company AES. Of course, it is still in the negotiation stage and the acquisition has not been completed yet. However, this incident itself has already shown that in the stage of rapid development of AI, what is scarce is not only storage, but also electricity. Moreover, "electricity" is the foundation of all computing power. So the reasons for BlackRock's acquisition of AES or BlackRock's interest in "electricity" are probably the following: The "essential needs" of the AI era are shifting from chips to power and grid connection Although GPUs and memory are scarce products, they can be achieved through expanding production capacity and other means. However, due to insufficient electricity, queuing for grid connection, and insufficient substation capacity, computing power is a joke. Moreover, obtaining electricity is much more difficult and costly compared to GPUs and memory. 2. Electricity is equivalent to a "toll station" of computing power Chips are physical architectures, models are upper level applications, but all paths ultimately have to fall back to 'watts'. When data center expansion becomes a long-term trend, power assets are like highway toll stations, with more rigid demand, more persistent bargaining, and more infrastructure investment required for expansion. Private investment is even more difficult. 3. Electricity is the 'cash flow' of infrastructure The form of electricity business is not due to the increase in electricity prices, but rather because it is suitable as an asset with "long-term contracts and predictable cash flow". As long as there is a demand for computing power, electricity is a must-have. For example, AES recently signed a 20-year power supply agreement with Google. What AI needs is not just 'electricity', but 'stable electricity' AI data centers require a stable load of 7x24 and may expand at any time. So what is needed is not just electricity itself, but also deliverable electricity, preferably guaranteed electricity. After "electricity" becomes the bottom layer, the next scarce layer is the "power chain" (aside from the topic) The 'chain' here is not a blockchain, but a link. When many people talk about power bottlenecks, their first reaction is power plants, grid connections, and substations. However, the electricity that AI really needs is not just connected and usable, but processed all the way to become computing power. For example, when traveling from the power grid to the computer room, a large electrical energy conversion is required, with high voltage coming in and low voltage going out. From the computer room to the cabinet, it is necessary to solve the problems of power distribution architecture, redundancy America、PDU、 The busbar, from the cabinet to the server, needs to convert AC to DC, and convert 12V or 48V into electricity that GPU and HBM can truly use. Further upstream to the equipment end of the wafer fab, etching and deposition processes essentially use extremely fine and controllable electricity to drive the plasma. Electrical instability is not due to slower machines, but rather a decrease in yield and an increase in cost. That's also why companies like Advanced Energy AEIS (which is different from AES) will become more critical in the AI era. Electricity is important, but how to "make electricity usable" is itself a supporting facility for electricity. In human terms, when the market begins to realize that "electricity" is the foundation of computing power, people will definitely pay attention to the source of "electricity", but how "electricity" is turned into computing power is a hidden and crucial layer. That's also why I think these "power chain" companies are worth re examining. @bitget VIP， Lower rates and more generous benefits