CCTV News: Since 2025, with the upgrading of domestic big models, the demand for computing power leasing has also ushered in explosive growth. Recently, many listed companies have issued announcements to plan for computing power leasing.

In the computer room of an intelligent computing center in Hefei, Anhui Province, the computing power of this eight-card computing power server can reach 2.4P, and the computing power of 1P is about 1000 trillion floating-point computing power per second. This machine is equivalent to the sum of computing power of 300 high-performance laptops per second. In this computer room, there are 256 such servers.

It is understood that as a large computing power cluster center, many intelligent computing centers have gradually been put into use in the past two years. The relevant person in charge of this smart computing center introduced that the smart computing center has been in use for more than half a year. At present, the computing power scale has reached 15,000 P.C. It is expected that the company's computing power leasing business will usher in a boom in 2025.

The company's head Rao Yougen introduced that the total revenue this year is expected to be around 1 billion yuan. Compared with 2024, the growth should be more than triple. Photovoltaics of Hefei Intelligent Computing Center was connected to the grid during this year's Spring Festival. They have made technological innovations and used a large number of high-speed copper cable access technology to shorten the connection between servers, which can improve transmission efficiency and reduce latency.

It is understood that in recent times, many companies are stepping up purchasing computing power servers, especially high-performance computing power servers, showing a grand occasion of "hard to find a machine". The person in charge of a computing power leasing company in Hangzhou said that due to the growth in downstream computing power leasing demand, they are promoting the procurement of computing power services for 200 high-performance servers, and this batch of servers will be invested in the computing power leasing market in the near future.

Recently, many listed companies have successively issued computing power server procurement announcements. Industry insiders predict that with the upgrading of domestic big models and the increase in application scenarios, the computing power leasing industry is expected to maintain a high prosperity. Reduced large-model access costs. Enterprises accelerate the upgrade of domestic large-models, which has led to a reduction in the cost of application. Many companies have also begun to try to access large models and apply them to all aspects of their company's business.

Chen Chao is the head of a technology service company in Shanghai. In 2025, he connected to a domestic big model on the company's web page. After users entered their needs, they could break down the user needs into a series of companies or patented technologies with related technologies, helping users achieve more accurate "point-to-point" matching. Chen Chao introduced that domestic big models are cheap and have good results, and a series of functions will be launched in the future.

Chen Chao, executive director of a technology service company in Shanghai, introduced that one of the main advantages of Deepseek is that the calculation cost is greatly reduced, and the semantic recognition and reasoning capabilities are very strong. Their matching rate has increased by about 20%. In terms of cost, it is reduced by more than 50%.

The reporter learned from several computing power leasing companies that at present, computing power leasing orders for small and micro enterprises are showing explosive growth. Public services, education, finance and other fields are all building models based on their own knowledge base, while the computing power leasing model is small in investment and low in cost, which is more popular than self-built computing power rooms. Many companies are also developing a new model of "computing power leasing + service" to embed large models into servers to reduce the debugging costs of enterprises.

It is reported that since the beginning of 2024, the price of computing power rental has also been gradually decreasing. Several computing power cards are common in the market, and the annual rent has generally been reduced by 10%-20%. With the emergence of domestic open source models, the cost of inference training has also been greatly reduced. The flexible billing model of computing power leasing has also led many small and micro enterprises to start accessing large models for customer service, content creation, data processing and other scenarios.