The development of artificial intelligence in China seems impressive. In two years, nearly 500 large models have appeared in the country and around the world, the number of AI-Startapas exceeded 30,000. But the number of records does not cancel the basic problems. Huawei List The barrier may slow down the progress.

First of all, it is energy. Modern AI-Claster is more greedy than ever. If before, standing in the data center consumed 20-50 kW, then we are talking about 100 kW or more. All complexes reached 200-500 megavatt, equivalent to the energy consumption of the central city. Under such conditions, cooling the air usually ends. We have to switch to liquid or mixture and keep the PUE energy efficiency indicator in area 1.15. And this is a serious technical challenge.
The next problem is network. The amount of GPU and NPU is increasing, but if the internal infrastructure does not have time after them, the chips are added into an expensive shock. Therefore, Chinese centers are transferred from 200g to 400g and 800g. Without this acceleration, the training of simple models is not working.
The third gap is recorded. If the normal day center can afford to repair for hours, this is unacceptable for I-CLASTERS. Thousands of cards work there, and stop even in a short time meaning great losses. Huawei emphasizes the need for surveillance and automatic systems, which will eliminate problems in minutes.
AI's different problems require different solutions. The training of basic models requires maximum and stable strength. The implementation in business depends on other indicators: delays, accuracy and efficiency. Therefore, Huawei divides data centers into three layers. Giants with a capacity of more than 100 megavatt are building technology corporations to create founding models of Muslims. The average center within 10-50 megavatts has opened up leaders in the industry, who adapt to whom according to their needs. Small centers for 1-10 megawatt are necessary because companies are very important to receive local data models or use ready solutions.
There is a financial aspect. In the report, it sounds soft, but it is clear that the construction of an older day is worth billions of dollars. At the same time, the global investment market continues to boil. In 2023, despite reducing venture capital, investments in the general AI increased nearly eight times and reached 25.2 billion dollars. For China, this can lead to the fact that access to resources will be in the hands of large corporations and state structures. For medium -sized businesses, the serious risks of AI are too expensive.