the wave of artificial intelligence is in the ascendant. as ai infrastructure, intelligent computing centers are regarded as "new infrastructure" like water and electricity in the era of artificial intelligence, and have been put into operation all over the country. especially since the emergence of chatgpt at the end of 2022, the construction of intelligent computing centers in various places has accelerated, and a large number of intelligent computing center plans have been intensively introduced. intelligent computing centers in many regions have been built and entered the operation stage. so far, how many intelligent computing centers have been built in the country? who is the main construction body behind it? when the intelligent computing center turns from the in full swing and lively construction tide to the "second half" of operation, how can the intelligent computing center operate in the long term and achieve profit goals? with these questions, the reporter conducted an interview and investigation.

more than 250 intelligent computing centers have emerged

according to the different types of computing power, data centers are generally divided into three categories: general computing centers that focus on general computing tasks, intelligent computing centers that focus on artificial intelligence computing tasks, and supercomputing centers that focus on supercomputing tasks. these three categories have different application fields. general computing centers can provide computing power support for daily internet services, enterprise digitalization, etc.; intelligent computing centers support ai large model training and reasoning, which can be applied to fields such as autonomous driving and smart cities; supercomputing centers provide ultra-high performance computing power for extremely complex problems such as astronomical detection and resource exploration.

in the past, due to the small scale of the artificial intelligence industry and the narrow scope of supercomputing applications, data centers were mainly general computing centers. with the advent of generative artificial intelligence models such as chatgpt and sora, the wave of artificial intelligence has swept the world, and cultivating and expanding the artificial intelligence industry has become a consensus among all regions to seize the future opportunities. the development of the artificial intelligence industry is highly dependent on intelligent computing power, so at the same time, a wave of planning and building intelligent computing centers has also been set off across the country.

according to a securities times reporter's analysis of public data, more than 20 cities have built intelligent computing centers, including first-tier cities such as beijing, shanghai, and shenzhen, as well as regional central cities such as zhengzhou, wuhan, changsha, nanjing, xi'an, chengdu, and harbin, as well as cities in western regions such as hohhot in inner mongolia, zhongwei in ningxia, and karamay in xinjiang. in addition, intelligent computing centers are also accelerating their expansion to county towns, including the hubei lichuan intelligent computing center, the laiyuan hebei intelligent computing center, and the chongqing mobile zhongxian big data intelligent computing center... in the past two years, a number of county-level intelligent computing centers have sprung up like mushrooms after rain.

"as of the first half of 2024, there are more than 250 intelligent computing centers that have been built or are under construction in china," said bai runxuan, senior analyst at the artificial intelligence and big data research center of ccid consulting, a research institute under the ministry of industry and information technology, in an interview with a securities times reporter. it is worth noting that, in terms of time, there were 791 bidding-related events for intelligent computing centers in the first half of 2024, a year-on-year increase of 407.1%. "this shows that the construction of intelligent computing centers has received widespread attention and support across the country. starting from 2023, local governments have increased their investment in intelligent computing centers, promoted the development of infrastructure, and provided a solid technical foundation for the artificial intelligence industry." bai runxuan said.

government and operator-led cross-border players are actively deploying

behind the "blossoming" intelligent computing centers, there are many different construction entities, among which the most important participants are local governments and the three major operators. from the perspective of bidding alone, bai runxuan told reporters that among the 292 bidding events for intelligent computing centers in the first half of this year, operators had the largest number of biddings, reaching 163, accounting for 55.82%, accounting for half of the total. the next largest participants were governments (33.90%), enterprises (9.25%) and scientific research institutions (1.03%).

combing through the 2024 semi-annual reports, the smoke of the three major operators competing for the "computing power map" is clearly discernible. china mobile said that the company has accelerated the development of intelligent computing power, and the scale of intelligent computing power has reached 19.6eflops. the first batch of 13 intelligent computing center nodes in beijing-tianjin-hebei, yangtze river delta, guangdong-hong kong-macao greater bay area, chengdu-chongqing and other places have been put into production; china unicom's financial report stated that the company has built wanka intelligent computing centers in shanghai lingang and hohhot, and the intelligent computing power of the entire network has reached 10eflops; china telecom stated that the company has built the country's leading all-liquid-cooled wanka pool in beijing-tianjin-hebei and the yangtze river delta, and deployed inference pools in 31 provinces to meet the inference needs of various ai applications. the intelligent computing power has increased by 10eflops, reaching a total of 21eflops.

the market structure of the three major operators in the mobile communications era is also playing out in the era of artificial intelligence. at the same time, facing the booming demand for intelligent computing power, established data center companies such as gds and 21vianet, as well as internet cloud vendors such as baidu, alibaba, and tencent, have also become important forces in building intelligent computing centers by relying on their own advantages in big data, cloud computing, and artificial intelligence.

in addition, there are cross-border players who are also trying to "get a piece of the pie". not long ago, an announcement from hainan expressway attracted market attention. hainan expressway announced that in order to seize the development opportunities of the digital economy and promote the transformation and upgrading of the company's business, the company's holding subsidiary digital expressway intelligent computing center (lingshui) co., ltd. plans to invest approximately 104 million yuan to build the hainan lingshui intelligent computing center project. the intelligent computing center project is scheduled to be completed and put into production before december this year, with a computing power scale of approximately 676pflops.

in fact, there are many listed companies like hainan expressway that "cross-border" to carry out intelligent computing center business. in early january this year, huafu fashion, one of the world's largest color spun yarn manufacturers and suppliers, announced that its wholly-owned subsidiary zhejiang huafu plans to invest in the aigc intelligent computing center project of shangyu huashang digital intelligence center, planning to build an ai computing power of 3000pflops, and the planned investment amount for the first phase of the project is 250 million yuan. in addition, many listed companies such as "msg king" lotus holdings, real estate developer damingcheng, and dye manufacturer jinji shares have begun to lay out new computing power tracks and promote intelligent computing center projects in the past two years.

if it is reasonable for operators, data providers, cloud vendors and other entities to enter the intelligent computing track, then the exploration of these cross-border players to open up a second growth curve is often prone to market doubts. for example, since huafu fashion announced its entry into the computing power track, the reporter saw on the investor interaction platform that investors have been asking questions to the secretary of the board of directors about huafu fashion's intelligent computing center project, such as "whether there has been a computing power order" and "whether the computing power project can be delivered and operated as scheduled". a senior industry insider analyzed to the securities times reporter that the construction of an intelligent computing center not only requires heavy asset investment, but also has high requirements for sales channels, technical strength and other aspects. if cross-border players do not have sufficient industry accumulation and competitive barriers, they may not be able to gain a foothold in the market.

the problem of supply and demand mismatch remains to be solved need to explore diversified profit models

building an intelligent computing center is not a one-time deal. on the contrary, the real challenge has just begun after it is completed and put into operation. from construction to use, how to make the intelligent computing center truly serve the development of local industries and become an effective investment is a problem that every intelligent computing center must face.

angel investor and senior artificial intelligence expert guo tao told reporters that the main profit models of the intelligent computing center currently include computing power leasing, technical services, data operations, etc. to achieve sustainable investment returns, it mainly depends on the following factors: first, market demand, that is, whether the computing power and services provided by the intelligent computing center can meet customer needs; second, cost control, including construction costs, operating costs, etc.; third, technological innovation, whether it can maintain a leading position in hardware equipment, software platforms, etc.; fourth, policy support, whether the government has relevant policies to support the development of the intelligent computing center.

during the interview, the reporter learned that some intelligent computing centers currently have problems of low utilization and mismatch between supply and demand. on the one hand, there is a mismatch between supply and demand of computing power. at present, some regions that have built intelligent computing centers do not have sufficiently large industrial support and application needs, and the phenomenon of server idling is common. under the background of the implementation of the "east data west computing" project, a number of intelligent computing centers have been built in the central and western regions, but the conditions for the cross-domain flow of computing power are not mature, and it will take some time to build a "one network" of national computing power, which has led to a large surplus of computing power supply in these regions. "compared with the continuous increase in the number of hardware, it is more necessary to consider how to make full use of the existing computing power." an hui, deputy chief engineer of the china academy of information and communications technology, said.

on the other hand, there is still a mismatch between the supply and demand of computing power, and there is a large gap in computing power for high-end applications. "in the context of the rapid progress of large model technology, we have very high requirements for the efficiency of model training. if a new idea is proposed and it is not implemented in time, it is likely that someone else will have done it in a few days." a young teacher of artificial intelligence at a 985 university told the securities times reporter that many domestic intelligent computing centers currently use domestically developed chips, and the efficiency of model training needs to be improved. "just like a rough house, it takes a lot of time to do adaptation work before 'moving in'," the teacher said, and the computing power of some intelligent computing centers cannot meet market demand.

at the same time as the mismatch between supply and demand, according to bai runxuan's observation, many intelligent computing centers have a single profit model, mainly relying on computing power leasing, which is not conducive to long-term sustainable development. "since the intelligent computing center is an infrastructure project with a large investment, it may face certain challenges to recover the investment cost. therefore, the intelligent computing center needs to explore diversified profit models, such as providing value-added services, conducting scientific research cooperation, etc., to improve profitability." bai runxuan said. guo tao also believes that the intelligent computing center can expand services such as data cleaning, analysis, visualization, and customized algorithm development and optimization, and explore new profit models and business models.

this year's government work report proposed "moderately advanced construction of digital infrastructure and accelerating the formation of a national integrated computing system". respondents generally believe that "moderately advanced construction" is by no means blind construction or repeated investment, but rather, on the premise of good planning, to form an industrial ecology of coordinated development of investment, construction, operation, management and use, and continuously improve the utilization rate of intelligent computing resources. "local governments should fully consider local industrial development needs and technological development trends when planning intelligent computing centers, and avoid blindly following trends. at the same time, they should strengthen communication and cooperation with other regions to avoid duplicate construction." bai runxuan said.

editor: li dan

END