Recently, Hu Feifei, Sun Hao, and others from China Telecom Research Institute have published articles pointing out that due to the explosive growth of artificial intelligence brought about by ChatGPT, the demand for intelligent computing power has significantly increased for large-scale artificial intelligence models with huge parameter quantities, and there may be a huge gap in future computing power. At the same time, the rapid growth in demand for intelligent computing power will also have a huge impact on future computing infrastructure such as data centers and cloud computing. At present, the demand for computing power in artificial intelligence is growing faster than the supply of computing power. According to OpenAI's analysis, since 2012, the amount of computation used in the largest AI training has increased Exponential growth with a doubling time of 3.4 months. In the future, artificial intelligence will further drive the explosive growth of computing power demand. According to OpenAI calculations, the parameter quantity of GTP-5 will be 100 times that of GTP-3, and the required computing amount will be 200-400 times that of GTP-3. Recently, global top cloud vendors and data center providers have released plans and budgets to accelerate the layout of data centers. Meta expects a capital expenditure of $39 billion in 2023, mainly investing in servers and network infrastructure. In 2023, NTT announced plans to invest $59 billion in data centers, artificial intelligence, and other "growth areas" over the next five years, of which at least $11 billion will be used to expand or upgrade its data centers. The design and construction mode of the data center needs to meet the needs of AI. The next generation data center needs to invest in AI dedicated hardware and adopt new data center design, which is mainly reflected in four aspects. First, the AI data center breaks through the computing bottleneck through Heterogeneous computing, including the integrated deployment of CPU, GPU and dedicated hardware such as ASICs and FPGAs in the data center. Secondly, artificial intelligence and machine learning may require three times the power density of traditional data processing, and cooling is a topic of widespread concern. Liquid cooling and immersion cooling may be development trends. Thirdly, high-speed storage access is crucial for artificial intelligence workloads, and data centers need to expand their storage capabilities to meet growing demands. Fourthly, the efficiency of artificial intelligence data centers is directly related to the performance of their networks. Intelligent lossless networks with zero packet loss, low latency, and high throughput will become the network solution for artificial intelligence data centers. (New News Agency)