According to IDC, the Chinese AI application public cloud service market maintained high-speed growth in 2025, with its market size exceeding 13.7 billion yuan. In this sector, leading cloud providers hold dominant positions by leveraging comprehensive AI capabilities and diverse application scenarios. Baidu AI Cloud ranked first with a 30.7% market share, achieving broad adoption through extensive enterprise-level AI application scenarios including intelligent customer service, content creation, and knowledge management. Alibaba Cloud demonstrated strong performance in areas such as smart office solutions and marketing creativity, utilizing its capabilities in intelligent speech, customer service, and visual AI. Tencent Cloud continued to advance in consumer internet, media, and financial scenarios, relying on its visual AI capabilities and intelligent customer service. Huawei Cloud secured the fourth position through deep cultivation in government, finance, and manufacturing industries with its Pangu large model. The core competition in the AI application market has shifted from a "arms race" over model parameters to a "battle for implementation" based on scenario value. What users need is not isolated model API calls, but complete applications that genuinely solve business problems and enhance efficiency. Whether for intelligent customer service, content generation, digital human marketing, enterprise knowledge base Q&A, or code-assisted development, cloud providers must encapsulate large model capabilities into out-of-the-box products to appeal to the broadest range of enterprise clients. Considering this, leading vendors should rapidly shift their investment focus in AI application services from underlying model capabilities to "last-mile" competencies such as industry solutions, data integration, and workflow orchestration.
The "Computational Undercurrent" Behind Applications: The Large Model Training and Inference Market Continues to Expand The prosperity of the AI application market does not emerge from nowhere. Every response from an intelligent customer service agent and every generation of marketing copy consumes large model inference capabilities. Meanwhile, the model fine-tuning and training that enterprises undertake to create differentiated applications constitute another layer of rigid demand—the public cloud service market for large model training and inference. Although smaller in scale than the application layer, this market exhibits higher growth stability and customer stickiness. In 2025, the public cloud service market for large model training and inference reached 7.94 billion yuan, presenting a competitive landscape distinct from the aforementioned AI application market. Alibaba Cloud led significantly with a 42.2% market share, becoming the preferred platform for large model training and inference due to its long-term accumulation in AI computing power and a comprehensive MLOps toolchain. Huawei Cloud (13.1%) gained widespread recognition in the government and enterprise market by leveraging its Ascend AI chips and fully independent, controllable full-stack capabilities. Amazon Web Services (7.1%) maintained an advantage among international enterprises and foreign-invested companies through its global GPU resources and advanced model training frameworks. The rapid growth of the large model training and inference market is driven by three main factors: First, the explosion of generative AI applications has spurred a surge in training and inference demand. From text generation to image creation, and from code assistance to multimodal understanding, the proliferation of generative AI applications has generated massive demand for model training and inference. Enterprises not only need to call pre-trained models for inference but also require fine-tuning models based on proprietary data to build differentiated AI capabilities. Second, Agent applications are driving complex inference needs. As Agents move from concept to implementation, complex capabilities such as multi-step task planning, tool invocation, and long-context reasoning have become standard. This places higher demands on model inference efficiency, concurrency, and response latency, prompting enterprises to seek more professional training and inference services. Third, computing power scheduling, management, and optimization have become essential. The demand for GPU computing power for large model training and inference is growing exponentially, yet computing resources are scarce and expensive. How to efficiently schedule heterogeneous computing power, optimize model inference performance, and reduce per-Token costs has become a core challenge for enterprises. This has spurred demand for a series of professional services, including AI computing power management platforms, model inference optimization, and elastic scaling.
Noteworthy Divergence Signals in the Market It is important to note that growth in the training and inference market is not evenly distributed. The top three vendors (Alibaba Cloud, Huawei Cloud, Amazon Web Services) collectively hold over 62% of the market share, while small and medium-sized AI computing service providers are being rapidly squeezed out. IDC assesses that computing power scheduling efficiency and model optimization capabilities are replacing "bare computing power price" as key factors in customer selection. This indicates that the concentration of the training and inference market will further increase in the future, and computing power providers lacking engineering optimization capabilities will struggle to maintain competitiveness.
IDC Outlook: Four Irreversible Market Trends Trend One: AI Industrialization Enters Deep Waters, with Application Value Becoming the Core Metric The rise of the Token economy has lowered the barrier for enterprises to trial AI, but the real commercial value lies in application implementation. In the future, vendors capable of providing end-to-end AI application solutions or supporting enterprises in rapidly building industry-specific applications will gain a competitive advantage. IDC believes the market is accelerating its shift from being "technology feasibility-driven" to "business ROI-driven." Trend Two: Integrated Training-Inference Platforms Become the Mainstream Procurement Standard As model iteration speeds up and application scenarios become more complex, enterprises require seamless, end-to-end platforms covering model training, fine-tuning, deployment, and inference. Integrated training and inference not only improves development efficiency but also reduces the total cost of ownership (TCO) for AI applications through continuous optimization. IDC observed that in 2025, over 35% of leading enterprise customers considered "whether it possesses integrated training-inference capability" a core evaluation criterion during vendor selection. Trend Three: Multi-Cloud and Hybrid Cloud Strategies Become the Norm Considering data security, cost optimization, and vendor risk, an increasing number of enterprises are adopting multi-cloud strategies for deploying AI applications. This requires AI cloud service providers to offer open API standards, flexible deployment options, and a consistent experience across clouds. Single-cloud lock-in strategies are being re-evaluated by enterprise customers. Trend Four: Industry Verticalization and Scenario Refinement Proceed in Parallel On one hand, industries such as finance, healthcare, manufacturing, and education exhibit growing demand for vertical AI applications. On the other hand, general scenarios like marketing creativity, smart office, customer service, and code development continue to deepen. Vendors need to find a balance between "industry depth" and "scenario breadth." IDC anticipates that within the next two years, the growth rate of industry-customized AI solutions will surpass that of general-purpose AI applications.
IDC Recommendations: How Vendors and Users Should Act Recommendations for Cloud Vendors: Shift from "providing models" to "providing business templates + low-code Agent building capabilities" to lower the barrier for enterprise implementation. Invest in engineering capabilities for integrated training and inference, rather than merely expanding computing power pools. Computing efficiency management will become a key differentiator. Proactively embrace the multi-cloud ecosystem to avoid customer churn risks associated with lock-in strategies. Recommendations for Enterprise Users: Prioritize selecting cloud vendors with industry solutions and closed-loop training-inference capabilities to avoid being locked into a single model or computing power source. Pay attention to cross-model migration costs. When choosing model APIs or training-inference platforms, incorporate standardization and openness into the long-term evaluation framework. For Agent-type applications, it is advisable to start with non-critical business scenarios (such as internal knowledge Q&A, assisted writing) and gradually evolve towards automated processes. According to Lu Yanxia, Research Director at IDC China, the Chinese AI public cloud service market is at a critical juncture transitioning from being "technology-driven" to "value-driven." The Token economy has raised the market ceiling, but only AI applications that genuinely solve business problems can deliver sustained value to enterprises. In the future, vendors possessing model capabilities, an application ecosystem, and engineering implementation prowess will lead the next wave of AI industrialization.
Comments