The State Council has recently issued guidelines aimed at expanding capacity and improving quality within the service industry. The document sets a target for the sector's total scale to exceed 100 trillion yuan by 2030. It emphasizes the deep implementation of the "AI+" initiative, calling for accelerated development and adoption of intelligent programming tools and supporting the procurement of large language models and AI agent services. The guidelines also promote the advancement of 5G-Advanced networks, strengthen R&D for sixth-generation mobile communication (6G) technology, advocate for moderately超前 construction of mobile Internet of Things, and support the development of satellite internet application services. Furthermore, they call for the orderly advancement of computing power layout and edge computing infrastructure, alongside improvements to intelligent computing cloud service systems.
Since the breakthrough of large model technology, domestic tech companies have engaged in an intense "hundred-model battle." However, the path to profitability and achieving a commercial closed-loop amidst high computational costs and R&D investments has remained a significant challenge for the entire industry. The release of these guidelines provides the most powerful solution yet: procurement support with policy backing. This signals that China's AI application is transitioning comprehensively from "casual conversation" to a productivity tool stage focused on "getting work done," offering AI firms the much-needed financial viability. It indicates that vertical large models will experience explosive growth and a procurement boom in specialized service fields such as education, healthcare, consulting, and law. The industry anticipates that 2026 will mark the true beginning of the large-scale application and commercialization of AI in China.
Data shows that as of March this year, the overall user base for AI applications reached 446 million, an increase of 134.95 million since November 2025, representing a growth rate of 43.4%. The combined monthly active users (non-deduplicated) of the top five domestic AI applications surged from 355 million in Q3 2025 to 722 million within just six months. Doubao, Qianwen, and DeepSeek ranked as the top three, with monthly active user bases of 345 million, 166 million, and 127 million respectively, adding over 130 million new users in a single quarter. Their average activity rates for the entire first quarter were 33.5%, 17.1%, and 21% respectively, indicating rapidly increasing user engagement.
Recently, AI large model developers both domestically and internationally have intensively released new model versions, indicating continuous acceleration in industry technology iteration. Alibaba consecutively launched several new models, including the full-modal large model Qwen3.5-Omni, CoPaw version 1.0, and the Qwen3.6-Plus model. KNOWLEDGE ATLAS released the GLM-5V-Turbo multimodal coding base model. Step星辰 launched the new Step3.5Flash version. Google released the Gemma4 open-source large model.
On April 4th, the internationally renowned third-party API platform for large models, OpenRouter, announced via its official X account that the daily token consumption for Alibaba's new Qianwen model, Qwen3.6-Plus, exceeded 1.4 trillion tokens, setting a new global record for single-model daily calls on the platform. On April 2nd, Volcano Engine stated via its official account that by March this year, Doubao's token usage had surpassed 120 trillion, doubling in just three months and growing 1,000-fold since its release in May 2024.
Kaiyuan Securities noted that AI applications are transitioning from lifestyle scenarios to productivity tools and evolving from Chat towards Agent functionality. This shift is expected to drive increased inference depth and longer calling chains, thereby promoting sustained high-speed growth in underlying token consumption. With demand-side token call volume growing exponentially and supply-side core hardware procurement costs rising significantly, the cloud computing industry is poised to enter a new stage centered on AI computing power premiums.
Zhongyou Securities indicated that as AI applications penetrate more deeply, token consumption continues its explosive growth. The bottleneck in computing power supply is becoming increasingly prominent, leading to a potential imbalance between supply and demand. They express optimism about future price increases spanning from computing power leasing to cloud services and large models. Recommendations focus on: 1) Computing power leasing/IDC (benefiting from demand growth and price hike expectations); 2) Cloud services (direct beneficiaries of user growth); 3) Large language models (benefiting from surging API calls); 4) IM entry points; 5) Computing hardware (providing foundational compute support).
Related concept stocks: BABA-W (09988): Alibaba released Qwen3.6-Plus, described as the most powerful domestic programming model, achieving breakthroughs in programming and AI agent capabilities. In the programming sub-ranking of the international model arena Arena, the Qwen3.6-Plus preview version ranked first in China, making Alibaba the world's second-ranked AI institution for programming capability. MINIMAX-WP (00100): On April 12th, MiniMax M2.7 was officially open-sourced globally. Collaborating with chip manufacturers and inference platforms including Huawei Ascend, Moore Thread, MetaX, Kunlun Chip, NVIDIA, TogetherAI, Fireworks, and Ollama, the model completed access and inference adaptation on its first open-source day. KNOWLEDGE ATLAS (02513): Based on 2024 revenue, KNOWLEDGE ATLAS is China's largest independent large language model developer and the second-largest LLM provider overall, holding a market share of 6.6%. The company's average daily token consumption for 2022, 2023, 2024, and the first half of 2025 was 500 million, 2.1 billion, 0.2 trillion, and 4.6 trillion tokens, respectively.
Comments