According to the latest report by International Data Corporation (IDC) on the "China Intelligent Computing Cloud Infrastructure Market (AI IaaS) (First Half of 2025)", China's AI IaaS overall market experienced a year-on-year growth of 122.4% in the first half of 2025, reaching a market size of 19.87 billion yuan. The GenAI IaaS market alone grew by 219.3%, totaling 16.68 billion yuan, while the Other AI IaaS market contracted by 14.1%, with a market size of 3.19 billion yuan.
In detail, the AI IaaS market is experiencing explosive growth as demand across various sectors embraces AI. On the supply side, since last year, cloud service providers have been steadily investing in AI infrastructure, rapidly reaching and serving end-users via diverse deployment models like public clouds, edge clouds, and private clouds. As a result, the resource supply and pricing within the computing power market remained generally stable in 2025, with only minor fluctuations.
On the demand side, strong performance was evident in multiple sectors such as the internet, automotive, mobile manufacturing, finance, and government, which continued to increase expenditures related to intelligent computing power and AI applications. The automotive industry and smart driving suppliers are intensifying their competition by ramping up their investments in computing power to gain a competitive edge in smart driving solutions. Additionally, the government sector saw significant increases in computing expenditures catalyzed by the DeepSeek initiative at the beginning of the year, while the embodied intelligence industry is accelerating its implementation, with related computing demand expected to be gradually released in the second half of 2025.
A tipping point is rapidly approaching, with expectations around DeepSeek gradually materializing. In 2025, the focus in the industry has shifted from large-scale model training to inference. In the first half of the year, inference scenarios accounted for 42% of the GenAI IaaS market, while training scenarios decreased to 58%. Following the positive impact of the DeepSeek event at the beginning of the year, the peak of deployment in central and state-owned enterprises and government sectors has nearly concluded. Simultaneously, leading companies in the sector are beginning to trial proof-of-concept (PoC) validations and small-scale deployments of generative AI applications within their business systems. It is anticipated that as internal enterprise multimodal generation and real-time inference scenarios continue to flourish, inference infrastructure will become a core branch of AI IaaS alongside the development of the AI Agent market.
The pattern of computing power supply is evolving towards a diversified ecosystem as AI applications in enterprises grow rapidly, and the generative AI and large model markets enter the inference era. With existing computing resources no longer in short supply, cloud vendors and major computing customers are increasingly focused on optimizing the cost structure of inference services and beginning to conduct horizontal comparisons of various models, service platforms, and accelerated computing chips. Currently, mainstream domestic and international cloud computing vendors are actively developing self-research chips, and it is expected that domestic computing power will enter a new growth cycle driven by internal factors.
In terms of market structure, the GenAI IaaS service market is driven by the computing power demand for training and inference scenarios, with its market share rising to 84% in the first half of the year, while the Other AI IaaS market's share is down to 16%. The market structure for GenAI IaaS has essentially consolidated, with market concentration further increasing. Alibaba Cloud continues to lead the market by substantially increasing capital expenditures on AI infrastructure and enhancing computing power reserves, catering to diverse AI scenario demands through various AI IaaS services. Volcano Engine, backed by the cost advantages of ByteDance's computing scale, adopts a relatively aggressive sales strategy, providing cost-effective computing resources in the AI market. Baidu (09888), leveraging Kunlun chips and the Baige platform, focuses on the computing needs of major clients across sectors such as automotive, e-commerce, and audiovisual, further expanding its market share. Tencent (00700) has actively embraced AI infrastructure investments this year, leading to rapid growth in GPU-related revenues.
Within the telecommunications operators, existing intelligent computing resources were rapidly absorbed in the first half of the year. The intelligent computing business is swiftly implementing in central and state-owned enterprises and government sectors while maintaining an appropriately proactive pace in intelligent computing construction. China Telecom is building a distributed intelligent computing network relying on the Xirang platform to provide computing resource coverage across different regions and industries, significantly boosting its intelligent computing revenue this period. China Mobile continues to increase its investment in computing resources and recently announced an upgrade to its "AI+" action plan, leading to rapid development of related intelligent computing businesses. China Unicom launched an upgraded version of Unicom Cloud AI in July, alongside upgrades to its "Unicom Stargazing" computing scheduling platform 2.0, enhancing AI full-stack capabilities and continually promoting the commercialization of intelligent computing.
At the same time, a clear differentiation is emerging among mid-tier and emerging service providers. Some are expanding their GenAI IaaS business scale through ecosystem partnerships with leading players, while others face challenges including contract defaults on large computing orders, effectively withdrawing from the market competitive landscape.
IDC's analyst for Chinese enterprise research, Chen Qijin, noted that in the first half of 2025, China's AI IaaS market continues to maintain an unexpectedly high growth trajectory. IDC predicts that by 2029, the market size will approach 150 billion yuan, with inference computing's share nearing 80%. From a technological perspective, in 2025, multimodal models and video generation models will achieve significant progress, while the performance and complex task processing capabilities of various models will further enhance, providing numerous upgrade opportunities for AI applications. On the demand side, pre-training scenarios are gradually converging, with industry users expected to focus on post-training and inference scenarios related to AI-driven businesses in the future. On the supply side, cloud vendors and telecommunications operators are continuously promoting technology iteration of AI infrastructure to achieve synergetic evolution between software and hardware and architectural innovation, as the computing power and chip market remains in a rapid iteration phase. The training and inference demand for generative AI is driving the core value of AI IaaS services from "efficient resource supply" to "innovation empowerment". New AI applications will propel exponential growth in AI computing power, while the technological iteration of AI computing power will further spur the emergence of new AI applications.