Hong Kong stock concept tracking | Large-scale, intelligent body procurement welcomes policy dividends Token consumption is expected to continue to grow at a high speed (with concept stocks)
The State Council recently issued the "Opinions on Promoting the Expansion and Improvement of the Service Industry." The opinions pointed out that it is necessary to deepen the implementation of the "AI+" action, accelerate the research and use of intelligent programming tools, and support the procurement of large models and intelligent body services.
The State Council recently issued the "Opinions on Promoting the Expansion and Quality Improvement of the Service Industry." The opinions clearly state that by 2030, the total scale of the service industry will surpass 100 trillion yuan. The opinions point out the need to deepen the implementation of the "artificial intelligence+" action, accelerate the research and development and use of intelligent programming tools, and support the procurement of large models and intelligent body services. Promote the development of 5G-A networks, strengthen the research and development of sixth-generation mobile communication (6G) technology. Moderately advance the construction of the mobile Internet of Things. Develop satellite internet application services. Promote the orderly layout and construction of computing resources and edge computing, and improve the intelligent computing cloud service system.
It is reported that since the outbreak of large model technology, domestic technology companies have experienced fierce "battle of large models", but how to move from high computing power consumption and research and development investment to profitability, complete the commercial closed loop, has always been a sword of Damocles hanging over the entire industry. The release of the above-mentioned "Opinions" provides the most powerful breakthrough solution - procurement support under the endorsement of policies. This marks the transition of Chinese AI applications from "conversational chat" to the productivity stage of "working and handling tasks", providing AI companies with the most desired hematopoietic ability of real money. It heralds the explosive development and procurement wave of vertical large models in more professional service fields such as education, healthcare, consulting, and law in the future.
The industry expects that 2026 will be the true "year of explosion of AI application and commercialization" in China. Data shows that as of March this year, the overall user scale of AI applications has reached 446 million, an increase of 134.95 million since November 2025, with a growth rate of 43.4%. Among them, the total monthly activity of the top five AI applications in China (non-unique) has soared from 355 million in the third quarter of 2025 to 722 million in just half a year. Dou package, Qianwen, and DeepSeek rank in the top three, with monthly active user scales of 345 million, 166 million, and 127 million respectively, adding over 130 million users in a single quarter. At the same time, the average activity rates of the three in the entire quarter reached 33.5%, 17.1%, and 21%, showing a rapid increase in user stickiness.
Recently, AI large model manufacturers at home and abroad have intensively released new models, and the industry's technological iteration continues to accelerate. Among them, Alibaba has continuously released multiple new models, including the all-modal large model Qwen3.5-Omni, CoPaw1.0 version, Qwen3.6-Plus model, etc.; KNOWLEDGE ATLAS released the GLM-5V-Turbo multi-modal Coding base model; Jieyue Xingchen launched the Step3.5Flash new version; Google released the Gemma4 open-source large model.
On April 4th, the internationally renowned large model API third-party call platform OpenRouter announced through its official X account that the daily call volume of Alibaba's Qianwen 3.6-Plus new model exceeded 1.4 trillion tokens, breaking the global record for daily single-model calls on the platform. On April 2nd, the Volcano Engine official account announced that as of March this year, the token usage of Dou package had exceeded 12 trillion, doubling in three months, an increase of 1000 times since its release in May 2024.
Open Source Securities stated that AI applications are transitioning from life scenarios to productivity tools, evolving from Chat to Agent, and are expected to drive continuous increases in reasoning depth and call chains, thereby continuously driving the rapid growth of underlying token consumption. With the exponential growth in token calls on the demand side, the supply side's core hardware procurement costs have significantly increased, and the cloud computing industry is expected to enter a new stage with the core of AI computing power premiums.
China Post Securities stated that as AI applications penetrate deeply, token consumption continues to grow explosively, and the bottleneck of computing power supply is becoming increasingly prominent. The imbalance between computing power supply and demand is seen, and the future looks promising for the continuous rise in computing power consumption from leasing to cloud to large models. Suggestions for focus: 1) computing power leasing/IDC (expectations of price increases in computing power demand growth); 2) cloud services (the most direct "selling shovels" to enjoy user growth dividends); 3) large models (benefiting from the surge in API calls); 4) IM entrance; 5) computing power hardware (underlying support for computing power).
Related concept stocks:
Alibaba-SW (09988): Alibaba releases the domestic strongest programming model Qwen3.6-Plus, achieving breakthroughs in programming and intelligent capabilities. In the programming capability sub-list of the international renowned large model competition Arena, Qwen3.6-Plus preview version ranks first in China, making Alibaba the second-ranked AI organization globally in programming capabilities.
MINIMAX-W (00100): On April 12th, MiniMax M2.7 was officially open-sourced globally, collaborating with Huawei Ascend, Moore Threads, Muxi, Kunlun Chip, NVIDIA, as well as domestic and foreign chip manufacturers and inference platforms such as TogetherAI, Fireworks, Ollama, and completing model access and inference adaption work on the first day of open source.
KNOWLEDGE ATLAS (02513): Based on the revenue in 2024, KNOWLEDGE ATLAS is the largest independent large language model manufacturer in China and the second largest in the world, with a market share of 6.6%. The daily average token consumption of the company in 2022, 2023, 2024, and 2025 H1 were 5 billion, 21 billion, 0.2 trillion, and 46 trillion, respectively.
Related Articles

CITIC SEC: humanoid Siasun Robot&Automation is about to increase in volume, rapid development of motor winding equipment.

GMTEight List of A-share restricted sales and lifting restrictions | April 22nd

CNT GROUP(00701) will resume trading on April 22nd.
CITIC SEC: humanoid Siasun Robot&Automation is about to increase in volume, rapid development of motor winding equipment.

GMTEight List of A-share restricted sales and lifting restrictions | April 22nd

CNT GROUP(00701) will resume trading on April 22nd.

RECOMMEND

DeepSeek’s $10 Billion Valuation Financing Rumor And Four Layers Of Logical Judgment
21/04/2026

48 Hours Witness AI’s New Battlefield: Alibaba Joins, Tencent Open Sources, Manycore Tech Lists — Has The World Model Reached Its ChatGPT Moment?
21/04/2026

Public Fund Giants Reverse Course With Heavy Hong Kong IPO Cornerstone Allocations Exceeding HK$2.2 Billion Year‑To‑Date
21/04/2026


