AI-Driven Computing Power Demand Sparks Accelerated Construction of Computing Network 'Highways'

[baoshi.rao]

Large-scale model algorithms are propelling artificial intelligence toward more general-purpose applications, driving a surge in demand for intelligent computing power. Around computing hub nodes, 130 backbone optical cables have been constructed, significantly enhancing data transmission performance. The industrial chain continues to expand, with China now leading the world in the production of computing products such as servers and computers... The 2023 China Computing Power Conference, recently held in Yinchuan, Ningxia, revealed new developments and trends in the computing power industry, highlighting its role as a cornerstone of digital and intelligent transformation across society.

From scientific research to industrial production and daily life, computing power is permeating every aspect of production and living, much like water and electricity. In recent years, a series of important policies and initiatives have been introduced at both national and local levels, along with the implementation of major engineering projects, driving rapid development in the computing power industry.

Minister of Industry and Information Technology Jin Zhuanglong presented the latest data at the conference. As of now, the total scale of operational data center racks nationwide exceeds 7.6 million standard racks, with a total computing power capacity of 197 exaflops (197EFLOPS), ranking second globally. The computing power industry has begun to take shape, with China leading the world in the production of computing products such as servers, computers, and smartphones.

Particularly, the development of AI technologies like large-scale models has further increased the demand for computing power. According to the China Comprehensive Computing Power Index (2023) released by the China Academy of Information and Communications Technology (CAICT) at the conference, general-purpose computing power currently accounts for 74% of the total computing power scale, while intelligent computing power accounts for 25%, a 60% year-on-year increase.

'This reflects the significant role of large-scale models in driving the demand for AI computing power,' said Yu Xiaohui, President of CAICT. He noted that China's computing power structure is optimizing, with intelligent computing power demand showing explosive growth.

Huawei Chairman Liang Hua also emphasized that rapid advancements in industries such as autonomous driving, life sciences, and smart manufacturing have led to escalating demands for computing power to support ultra-large AI models and massive datasets. Intelligent computing power suitable for large-scale model training has become the primary driver of computing power growth.

With the surge in intelligent computing power demand, multiple stakeholders are ramping up high-performance intelligent computing supply, accelerating key technological breakthroughs, and unlocking the empowering effects of AI computing power.

Minister Jin Zhuanglong stated that the Ministry of Industry and Information Technology will continue to optimize infrastructure layout. This includes refining the top-level design for high-quality development of computing facilities, coordinating computing resources, increasing the supply of high-performance intelligent computing, enhancing the reliability of computing networks, and improving efficiency in resource utilization.

Enterprises are also intensifying their AI innovation efforts to upgrade computing power infrastructure. In Zhongwei, Ningxia, the 'China Unicom Intelligent Computing Training-Inference Integrated Hub Construction Action Plan' was recently launched. During the conference, China Unicom staff introduced plans to build the first integrated training-inference intelligent computing hub node in western China, which will provide AI training-inference services and boost the intelligent computing industry in Ningxia and the broader western region.

Meanwhile, China Telecom is strengthening independent R&D in AI frameworks, algorithm models, and scheduling platforms, building the Galaxy AI algorithm platform, and collaborating with partners to develop industry-specific large models and intelligent applications. China Mobile is systematically advancing the integration of general computing, intelligent computing, supercomputing, and quantum computing, while initiating the construction of Asia's largest standalone intelligent computing center.

'Computing power infrastructure construction is in full swing. While accelerating the development of national hub nodes and large-scale data centers, the importance of edge computing infrastructure is becoming increasingly prominent,' said Zhang Zhiyong, Chairman of China Tower. He emphasized the need to continuously improve public computing power layouts, provide edge intelligent computing and shared computing services based on AI hardware and software, expand computing power supply, and achieve ubiquitous computing power and pervasive intelligence.

The realization of 'ubiquitous computing power' relies on the support of computing networks. Data from the Ministry of Industry and Information Technology and the National Bureau of Statistics show that China has deployed 130 backbone optical cables around computing hub nodes, significantly improving data transmission performance. In the first half of this year, investment in new information infrastructure such as 5G and data centers grew by 13.1%, while investment in integrated new infrastructure like industrial internet and smart transportation increased by 34.1%.

Industry insiders believe that in practice, heterogeneous computing power and software incompatibility are common. Connecting existing computing centers with different architectures through networks can reasonably allocate, share, schedule, and release more computing power while lowering the barrier to application.

"To maximize the utility of computing power and directly impact GDP, two technical challenges must be addressed: first, ultra-broadband, ultra-low-latency network connections for large-scale computing centers nationwide; and second, computing power scheduling," said Gao Wen, an academician of the Chinese Academy of Engineering.

During the conference, China Mobile, in collaboration with multiple supercomputing centers, AI computing centers, and cloud service providers, launched the largest nationwide "Baichuan" computing power integration initiative. The platform features a national computing network map that simultaneously integrates general computing, AI computing, supercomputing, and quantum computing. It collaborates with national-level supercomputing centers, authoritative AI computing providers, and leading cloud vendors for integration, and has partnered with CETC and Boson Quantum to achieve optical quantum computing integration. Currently, the platform offers a total computing power supply capacity exceeding 10 EFLOPS.

"We aim to provide a new model for computing power supply, building a unified computing network where computing power is as accessible as electricity," said Wang Xiaoyun, Chief Scientist at China Mobile.

Industry experts noted that the accelerating digitalization, networking, and intelligentization across various sectors have driven exponential growth in computing power demand, expanding the diversity of computing network development. With the continuous optimization of China's computing network layout and the further activation of national computing resources, the empowering effect of computing power will be further unleashed, driving the digital economy to new heights.