In the first half of 2026, the core theme for A-shares and Hong Kong stocks, excluding short-term geopolitical noise and market sentiment, is the transition of AI from infrastructure narratives to application deployment validation. While the market debates whether North American cloud providers' capital expenditure has peaked, Chinese internet giants have quietly begun a new round of ecological positioning. This time, the core competition is no longer for traffic portals, but for AI agents that can understand, plan, and execute tasks.
The AI Agent Arms Race: A Qualitative Shift from "Q&A" to "Task Execution"
If 2025 was the year of AI coding, 2026 undoubtedly belongs to the AI Agent. Previously, user interaction with AI was confined to a "question and answer" dialogue phase. Now, products from various companies have evolved to a "proxy execution" stage, capable of directly invoking tools and performing complex tasks.
Tencent: WorkBuddy Stakes Out Office Ecosystem, WeChat Agent Builds a Super Portal
Tencent's AI application strategy is shifting from "internal efficiency improvement" to "ecological positioning", with the core being WorkBuddy and the WeChat AI Agent. WorkBuddy, officially launched in March 2026, was developed by the Tencent Cloud CodeBuddy team based on the same Agent intelligent architecture, targeting non-technical workplace users. Its core capability extends AI coding to execution-level tasks like calling tools and processing documents. Since its launch, data has grown rapidly, with PC monthly visits reaching 8.85 million in its launch month, surging to 20.97 million by June 2026—a 2.4 times increase from three months prior. By early July, its monthly active users had reached 20 million. WorkBuddy has now officially been listed on the HarmonyOS PC app market, becoming the first desktop office agent on the platform. Guohai Securities estimates that Tencent's Financial Technology and Enterprise Services revenue will grow 7% year-on-year in Q2 2026, driven by the iteration of the Hunyuan model, WorkBuddy's commercialization, and cloud product price increases. Furthermore, the WeChat AI Agent leverages its 1.4 billion monthly active users and vast mini-program ecosystem. Using an API-First technical path, it operates faster and offers greater stability than competitors' visual simulation models. Major brands like JD.com, Meituan, Didi, and Ctrip have already taken the lead in joining.
Alibaba: Qianwen Office Integrates Internal Resources, DingTalk Ecosystem Awaits Potential
Alibaba's path to AI Agent has been more circuitous. Since early 2026, the company has launched several Agent products, including Qoder Work, Wukong, and MuleRun, through an internal competition process. Most recently, "Qianwen Office" has begun small-scale testing, deeply integrating the aforementioned products, led by the newly appointed CEO of DingTalk, Yusen Chen. Currently, a local desktop client is available, with web-based and DingTalk-integrated versions to follow. In the long run, whether Qianwen Office can leverage DingTalk's B-end advantages to form a differentiated barrier will be a key indicator for observing the effectiveness of Alibaba's enterprise service smart transformation.
Meituan: Xiaotuan 2.0 Completes the "Execution" Loop, Local Service Scene Barriers Begin to Show
On July 27, Meituan announced a comprehensive upgrade of its AI-native assistant, "Xiaotuan". The core breakthrough of Xiaotuan 2.0 is moving from information inquiry to transaction execution. Users only need to state their request, and the AI can, using real-time information, assist in completing actions like placing orders, hailing taxis, and making reservations. For merchants without online booking, Xiaotuan also supports AI-powered outbound calls to automatically complete phone consultations and reservations. Currently, Xiaotuan's information services cover over 2,800 cities nationwide, having completed more than 700 million merchant information verifications. Haitong Securities' inaugural coverage report indicates that AI and robotics are the most important sources for Meituan's future re-pricing, improving consumer decision-making and delivery scheduling efficiency in the short term, and potentially having a disruptive impact on the existing service industry model in the long term.
The Rationale Behind Capital Expenditure: Why Chinese Internet AI Investment is Far from its Ceiling
The market's primary concern about the AI industry chain is whether overseas cloud providers' capital expenditure growth has peaked. This concern was amplified in June when Meta announced it would commercialize its older generation of surplus computing power, an act initially misinterpreted by the market as "excess computing capacity". However, the reality is that capital expenditure expectations for the four major North American cloud providers—Google, Amazon, Microsoft, and Meta—have been revised upwards to over $725 billion for 2026, a significant increase from $410 billion in 2025. Their cloud business revenue maintains high growth, supporting the increased spending. In contrast, Chinese internet giants' capital expenditure "firepower" is far from being fully unleashed. For example, Tencent completed an offshore convertible bond issuance of approximately 31.5 billion yuan in June, with funds primarily directed towards AI computing power, large models, and the commercialization of WorkBuddy. According to Goldman Sachs, the ratio of capital expenditure to operating cash flow for major Chinese cloud service providers remains at a healthy level, indicating that their AI infrastructure investments are not overly squeezing cash flow. This is thanks to the stable growth in operating cash flow for Chinese cloud providers in 2026 and a relatively controllable pace of capital expenditure expansion, without the aggressive over-investment seen among their US counterparts. According to the latest investment bank forecasts, Tencent's full-year capital expenditure for 2026 could increase to over 180 billion yuan. Yet, even so, Tencent's ratio of capital expenditure to cash flow and net profit remains significantly lower than that of overseas giants like Google and Meta. Tencent's substantial cash reserves provide the capacity for long-term, sustained AI investment. This difference in "capital expenditure potential" underscores the unique business structure advantage of Chinese internet companies. Their core businesses—such as gaming, advertising, e-commerce, and local services—still generate stable operating cash flow, making AI investment more of an "incremental" addition rather than a "substitution" of existing resources, providing a thicker financial safety net.
From "Valuation Recovery" to "Profit Delivery + AI Commercialization"
The first half of 2026 saw a clear shift in the logic driving the Hong Kong tech sector. Reflecting on the AI evolution process, from the "capital expenditure arms race" in Q4 2023, to the explosion of inference computing demand in Q3 2024, and the first true AI application boom with AI Coding in Q2 2025, the number of users and tokens for leading companies' large models has grown exponentially, achieving a closed-loop business model for overseas models. Currently, AI application catalysts are landing intensively. The concentrated launches of Tencent's WorkBuddy, Alibaba's Qianwen Office, and Meituan's Xiaotuan 2.0 mark the transition of Chinese internet companies' AI commercialization from "proof of concept" to "large-scale deployment". Having completed a bottom-up repair, the investment logic for Hong Kong tech stocks in the second half of the year has shifted from risk appetite recovery to three core themes: profit delivery, enhanced shareholder returns, and validation of AI commercialization. In terms of valuation, as of July 27, the Hong Kong Stock Connect Internet Index's latest price-to-earnings ratio was only 19.02 times, placing it in the 5.57th percentile over the past three years—meaning its valuation is lower than 94.4% of the time in that period.
Key Expectations and Discrepancies to Watch
While the narrative of AI Agents is grand, multiple uncertainties remain in their implementation. Regarding the pace of commercialization, WorkBuddy currently uses a subscription-plus-credit model, focusing on expanding its user base in the short term. While its long-term monetization potential is viewed favorably, user willingness to pay and conversion rates require continuous monitoring. In terms of the competitive landscape, the entry barriers in sectors like AI office and AI local services are not insurmountable. Competitors such as ByteDance's Trae Work and Zhipu's AutoClaw have already entered the arena, increasing the crowdedness of the track. As for the return on capital expenditure, the characteristic of upfront AI infrastructure investment lagging behind commercialization deployment means the ROI verification cycle is long. The market's patience for capital expenditure will depend on the application side's ability to consistently deliver impressive data.
Related Index and ETF
The CSI Hong Kong Stock Connect Internet Index has its top ten constituent stocks featuring major tech giants and AI application companies across various fields, with significant leading advantages. The Hong Kong Stock Connect Internet ETF (Huabao, 513770) tracks this index. Its strategic advantage lies in the purity of its index constituents, which strictly filters out hardware companies, directly focusing on the delivery of the AI Agent commercialization closed-loop for players like WorkBuddy (Tencent), Qianwen Office (Alibaba), and Xiaotuan 2.0 (Meituan). It is currently the most efficient Beta tool for capturing the market's expectation of "AI applications transitioning from concept to large-scale deployment".
Conclusion
The Chinese internet industry is at a critical narrative switch, moving from "valuation recovery of traffic platforms" to "re-evaluation of the AI agent ecosystem". Tencent, through WorkBuddy and the WeChat AI Agent, is building a dual-driven ToB+ToC Agent ecosystem. Alibaba, via Qianwen Office, is attempting to integrate internal resources to regain ground in the B-end market. Meituan, leveraging its local service scene barriers, has taken the lead in creating a closed-loop for "proxy execution". The development of the AI industry is far from its endgame, but when the first truly viable application appears, capital markets often provide significant positive feedback. The AI Coding boom in Q2 2025 already validated this rule, and the wave of AI Agents in the second half of 2026 may be writing the next chapter.
Comments