By Gu Yekai, Cheng Huan, People’s Daily
Special effects and rendering in movies, as well as everyday applications of generative artificial intelligence (AI), facial recognition, and real-time translation, all rely on intelligent computing power. Today, as large AI models and generative AI continue to advance, the demand for computing power is rising rapidly.
According to a recent report on China’s AI computing power developmentin 2025, which was jointly released byInternational Data Corporation (IDC) and Chinese IT firm Inspur Information, China’s intelligent computing power is growing faster than expected.
Based on half-precision (FP16) intelligent accelerator card performance, China’s intelligent computing power reached 725.3 EFLOPS in 2024, surging 74.1 percent year-on-year. It is expected to rise further to 1,037.3 EFLOPS in 2025, 1,460.3 EFLOPS in 2026, and 2,781.9 EFLOPS in 2028. Meanwhile, the market size hit $19 billion, surging 86.9 percent from a year ago.
At the construction site of a large-scale intelligent computing center in Jinan, east China’s Shandong province, massive modules were lifted and positioned with precision, coming together like building blocks to form a state-of-the-art facility. It took just 120 days for this Meta Brain computing power factory of Inspur Information, consisting of 119 prefabricated containers, to enter full operation from groundbreaking.
As a fundamental pillar of digital economy, computing power falls into three main categories: supercomputing power, general-purpose computing power, and intelligent computing power. Designed specifically for AI applications, intelligent computing power supports the training and execution of AI algorithms and models. Across industries, demand for intelligent computing power is surging.
The Meta Brain computing power factorybuilt in just 120 days is a testament of the rapid growth in China’sintelligentcomputing power. “In 2024, China’s intelligent computing power grew three times faster than general-purpose computing power,” said Zhou Zhengang, vice president of IDC China.
In recent years, China has been ramping up its computing power infrastructure at national, regional, and corporate levels. “Over the next two years, China’s intelligent computing power is expected to continue its rapid expansion,” Zhou added. “In 2025, it is projected to grow another 43 percent compared to 2024, and by 2026, it will have doubled from 2024 levels.”
The adoption of intelligent computing is accelerating, with many enterprises actively embracing large AI models. According to an IDC survey, 42 percent of companies are conducting preliminary tests and proof-of-concept validations for large models, while 17 percent have already integrated them into production and real-world applications.
However, in manufacturing—an industry characterized by complex processes, diverse product categories, and fragmented, high-volume data—implementing AI models comes with challenges such as high costs and difficulties in data collection.
“For example, in quality inspection, obtaining a comprehensive set of defect samples is a major hurdle,” said Jia Jiaya, founder ofsmart manufacturing companySmartMore.
“In the past, production lines must be maintained operational two to three months in order to collect defect samples. But now, the deep learning of our industrial multimodal AI model allows us to accomplish the task in just a week—or even less,” Jia explained.
This AI model has already generated hundreds of defect types and is widely applied in sectors such as consumer electronics and new energy, according to Jia.
Beyond building their own intelligent computing infrastructure, many enterprises are turning to cloud computing services for digital transformation. For instance, Huawei’s Pangu large modelhave been deployed across more than 30 industries and 400 application scenarios, spanning manufacturing, pharmaceutical research, coal mining, and steel production.
At theGui’an Supercomputing Center in southwest China’s Guizhou province, rows of server cabinets hum with high-speed operations. With 1,000 high-performance graphics processors and over 700 servers, it boasts a combined computing power of 15 quadrillion calculations per second now
Diverse industry-specific needs are driving new requirements for computing power. “The financial sector requires high security and low-latency computing environments. The medical field needs to process vast amounts of medical imaging data. Manufacturing demands real-time production optimization, and the internet industry relies on large-scale user data processing and content recommendations,”said Liu Jun, senior vice president of Inspur Information. These demands call for computing power infrastructures that deliver high performance, low latency, robust security, scalability, and cost effectiveness.
Industry experts emphasized that after years of scaling up computing capacity, the next critical step for China’s computing power industry is transitioning from mere expansion to unlocking high-value applications and maximizing efficiency.
Discussion about this post