According to IDC, by 2025, China’s AI application public cloud services market is expected to maintain rapid growth and the market size will exceed 13.7 billion yuan. In this space, leading cloud vendors hold a dominant position thanks to their full-stack AI capabilities and rich application scenarios.
According to Zhitong Finance APP, IDC says that by 2025, China’s AI application public cloud services market will maintain rapid growth and the market size will exceed 13.7 billion yuan. Large cloud vendors hold a dominant position in this space due to their full-stack AI capabilities and rich application scenarios.
Baidu Intelligent Cloud ranks first with a market share of 30.7% and relies on comprehensive enterprise-level AI application scenarios, including intelligent customer service, content creation, and knowledge management for wide implementation. Alibaba Cloud stands out in scenarios like smart offices and marketing creativity with its intelligent voice, customer service, and visual AI capabilities. Tencent Cloud continues to address consumer internet, media, finance, and other scenarios based on visual AI capabilities and intelligent customer service. Huawei Cloud, which leverages the deep cultivation of Pangu large models in government affairs, finance, manufacturing, and other industries, firmly takes the fourth place.
The nature of competition in the AI application market is shifting from an “arms race” on model parameters to a “race on implementation value” in specific scenarios.
Instead of isolated model API calls, users want complete applications that can truly solve business problems and improve efficiency. Whether it’s intelligent customer service, content generation, digital human marketing, or enterprise knowledge base Q&A and code-assisted development, cloud providers must encapsulate the capabilities of large-scale models into ready-to-use products to appeal to the broadest range of enterprise clients. Given this, major vendors must rapidly shift the focus of their AI application services investments from underlying modeling capabilities to industry solutions, data integration, workflow orchestration, and other “last mile” capabilities.
The “hidden stream of computing power” behind your applications: The large-scale model training and inference market continues to expand.
A thriving AI application market does not emerge from thin air. Every response from an intelligent customer service system and every marketing copy generated consumes the inference capabilities of large-scale models. Meanwhile, enterprise model fine-tuning and training to create differentiated applications forms another layer of essential demand: the large-scale model training and inference public cloud services market. Although this market is smaller than the application layer, it is a market with stable growth and high customer stickiness.
By 2025, the market size of public cloud services for large-scale model training and inference is expected to reach 7.94 billion yuan, presenting a different competitive environment from the aforementioned AI application market.
Alibaba Cloud leads by a wide margin with 42.2% market share and is the preferred platform for large-scale model training and inference thanks to its long-term accumulation of AI computing power and complete MLOps toolchain. Huawei Cloud (13.1%) is widely recognized in government and enterprise markets through its Ascend AI chip and fully autonomous and controllable stack. Amazon Web Services (7.1%) maintains its dominance among domestic and international companies due to its global GPU resources and advanced model training frameworks.
The rapid growth of the large-scale model training and inference market is driven by three key factors:
First, the explosive growth of generative AI applications has sharply increased the demand for training and inference.
The rise of generative AI applications, from text generation to image creation, code assistance to multimodal understanding, has created a huge demand for model training and inference. Rather than just calling pre-trained models for inference, enterprises need to fine-tune them based on their own data to build differentiated AI capabilities.
Second, the application of intelligent agents (agents) drives the demand for complex reasoning.
As intelligent agents move from concept to implementation, complex features such as multi-step task planning, tool invocation, and long-context reasoning have become standard features. This places higher requirements on model inference efficiency, concurrency, and response latency, leading enterprises to seek more specialized training and inference services.
Third, scheduling, managing, and optimizing computing power has become an essential need.
The demand for GPU computing power for large-scale model training and inference is rapidly increasing, but computing resources are scarce and expensive. How to efficiently schedule heterogeneous computing power, optimize model inference performance, and reduce cost per token has become a central challenge for enterprises. This has increased the demand for professional services such as AI computing power management platforms, model inference optimization, and elastic scaling.
Differentiation signals suggested by the market
It is noteworthy that the growth of the training and inference market is not evenly distributed. While the top three vendors (Alibaba Cloud, Huawei Cloud, and Amazon Web Services) together account for more than 62% of the market share, small and medium-sized AI computing power service providers are rapidly being eliminated. IDC concludes that computing power scheduling efficiency and model optimization capabilities are replacing “raw computing power pricing” as key factors in customer selection. This indicates that the training and inference market will become even more concentrated and computing power providers without engineering optimization capabilities will struggle to remain competitive.
IDC Outlook: Four irreversible market trends
Trend 1: AI becomes more industrialized and application value becomes a core metric.
The rise of the token economy has lowered the threshold for companies to experiment with AI, but the real commercial value lies in application implementation. In the future, vendors that can provide end-to-end AI application solutions or support companies building industry-specific applications quickly will gain a competitive advantage. IDC believes the market is accelerating its transition from being “technology feasibility driven” to “business ROI driven.”
Trend 2: Integrated training and inference platforms will become the mainstream procurement standard.
As model iterations accelerate and application scenarios become more complex, enterprises need a seamless platform that covers the entire model training, fine-tuning, deployment, and inference process. Integrated training and inference not only improves development efficiency but also reduces the total cost of ownership (TCO) of AI applications through continuous optimization. IDC observed that by 2025, more than 35% of large enterprise customers will consider “integrated training and inference capabilities” as a core criterion when selecting a vendor.
Trend 3: Multi-cloud and hybrid cloud strategies will become the norm.
With data security, cost optimization, and supplier risk in mind, more and more companies are adopting multi-cloud strategies to deploy AI applications. This requires AI cloud service providers to offer open API standards, flexible deployment options, and a consistent cross-cloud experience. Enterprises are reevaluating their single cloud binding strategies.
Trend 4: Industry verticalization and scenario refinement proceed in parallel.
Meanwhile, vertically focused AI applications are increasingly in demand in industries such as finance, healthcare, manufacturing, and education. Meanwhile, common scenarios such as marketing creativity, smart offices, customer service, and code development continue to deepen. Vendors need to balance industry depth and scenario breadth. IDC predicts that within the next two years, the growth rate of industry-customized AI solutions will outpace the growth rate of general-purpose AI applications.
IDC Recommendation: How should vendors and users act?
Recommendations for cloud vendors:
Move from “providing models” to “providing business templates and the ability to build low-code agents,” lowering the barrier to enterprise implementation.
Invest in integrated training and inference engineering capabilities rather than simply expanding your computing power pool. Managing computing efficiency will become a key differentiator in a competitive market.
Aggressively adopt multi-cloud ecosystems to reduce the risk of customer churn caused by lock-in strategies.
Recommendations for enterprise users:
Prefer cloud providers with industry-specific solutions and closed-loop training and inference capabilities to avoid dependence on a single model or computational resources.
When choosing a model API or training inference platform, be aware of the cost of migration between models and build in standardization and openness into your long-term evaluation framework.
For agent-type applications, we recommend starting with non-critical business scenarios (internal knowledge Q&A, writing assistance, etc.) and gradually moving to automated processes.
Lu Yanxia, research director at IDC China, said China’s AI public cloud services market is in an important transition stage from “technology-driven” to “value-driven”. The token economy has expanded market potential, but only AI applications that truly address business challenges can deliver lasting value to enterprises. In the future, vendors with robust model capabilities, application ecosystems, and engineering implementation strengths will lead the next wave of AI industrialization.
