Wall Street heavyweight Morgan Stanley has released a new in-depth report titled "Consumer AI in China: Beyond the Chatbot Battle," shifting the investment narrative from user adoption to monetization. The analyst team highlights that China, with its mature mobile internet distribution networks and deeply integrated daily-life applications, is leading globally in consumer AI adoption. The next phase of competitive advantage, they argue, will belong to platforms that simultaneously control user entry points, proprietary data, merchant supply, and transaction execution capabilities.

The investment focus in China's consumer AI market is rapidly moving toward how user intent translates into revenue. In their latest research, Morgan Stanley projects that by 2030, the monetizable revenue space for China's consumer AI sector will approach 300 billion yuan (approximately 294 billion yuan), with an estimated 99% derived from advertising and transaction commissions. Based on this framework, the bank identifies TENCENT (HK: 00700) as the clearest beneficiary in the consumer AI landscape, while e-commerce and cloud computing giant Alibaba (HK: 09988) remains the top industry pick based on its full-stack AI capabilities. The analysts also view positively vertical platforms such as Meituan, major online travel platforms, and KANZHUN.

At the core of this near-300 billion yuan market outlook is what Morgan Stanley's quantitative models call "merchant-funded intent monetization," which encompasses advertising and AI-attributable transaction commissions. The 99% figure combines these two streams, with transaction commissions being the dominant force. The underlying business logic is straightforward: as consumers use AI for decision-making and purchases, platforms monetize through merchant advertising spending and commissions from completed online transactions.

For TENCENT, the bullish thesis centers on WeChat, the world's largest social and internet platform, which seamlessly connects social relationships, mini-programs, and payments into a task-execution environment. For Alibaba, the value lies in synergistic coordination among cloud computing, models, e-commerce operations, online transactions, and order monetization. Meituan represents what Morgan Stanley describes as the perfect fusion of AI large-model computational decision-making with local service fulfillment.

Consumer AI Enters the Transaction Battle: Morgan Stanley Mines 300 Billion Yuan Opportunities

The diffusion base for consumer AI in China is already extensive, and the next key question is how usage habits and task value monetization will deepen. Citing CNNIC data, Morgan Stanley notes that China's generative AI users grew from approximately 249 million in December 2024 to around 602 million by December 2025, building upon the world's largest smartphone, mobile payment, and integrated internet platforms. This massive user base underpins the bank's projection of nearly 300 billion yuan in monetizable revenue space.

In Morgan Stanley's AlphaWise survey sample, 80% of respondents use AI at least weekly in their personal lives, while 77% of employed respondents use it at least weekly at work—both figures significantly higher than the 54% recorded for the U.S. in corresponding surveys. However, the bank cautions that the China survey covered approximately 2,000 consumers aged 18-59 across first through fourth-tier cities and cannot be directly treated as national penetration rates; the U.S. and China surveys also had different timing.

The competitive landscape of Chinese consumer AI applications reveals both scale leadership and multi-platform usage patterns. According to Morgan Stanley's analysis, as of July 2026, the monthly active users (MAU) for Doubao, Qwen, DeepSeek, and Yuanbao stood at approximately 399 million, 161 million, 124 million, and 51 million respectively. Doubao's DAU/MAU ratio is about 42%, while Qwen and Yuanbao are both around 17%, and DeepSeek at approximately 25%. Notably, 64% of respondents report using at least five different AI tools in the past month.

The analysts observe that Doubao possesses advantages in traffic and product iteration, DeepSeek benefits from technical mindshare and user stickiness from its initial global popularity, Qwen excels in transaction connectivity, and Yuanbao serves as the capability incubation and user feedback function within the Tencent ecosystem. Usage is also expanding to the 40-59 age demographic, though privacy and security remain the primary concerns for low-frequency and non-users.

When evaluating the investment value of AI applications, Morgan Stanley emphasizes that standalone app user scale must be assessed alongside retention, task completion, and commercial conversion. This explains why the bank favors existing ecosystems like WeChat and Taobao—embedded AI features can directly improve existing behaviors such as shopping, food ordering, and travel booking. Approximately 75% of survey respondents find AI within shopping apps useful, and about 57% say such features would increase their purchase frequency. These are usage evaluations and purchase intentions; actual revenue effects still require verification through real conversion data.

The survey also reveals that the proportion of respondents who have paid for AI decreased from 41% to 35%, with only 21% preferring subscriptions and an average maximum monthly willingness to pay of about 41 yuan. This suggests a two-tier business model: maintaining free entry points for the mass market while charging subscription or per-use fees to high-consumption users, alongside monetization through merchant advertising, recommendations, and completed transactions.

Morgan Stanley breaks down the 2030 revenue space as follows: transaction commissions at 281 billion yuan, advertising at 10 billion yuan, and subscriptions and per-use fees at 3 billion yuan. Total scale is projected to expand from approximately 54 billion yuan in 2026 to 294 billion yuan by 2030, with a long-term base case of approximately 1.6 trillion yuan by 2040. Importantly, this measurement mainly represents the migration of existing internet revenue into AI-assisted or agentic AI processes, not entirely new revenue created for the industry.

Regarding the massive transaction commission outlook, the bank clarifies that measurement benchmarks must be based on attributable completed orders, with the same merchant payment not double-counted across advertising and commissions. The true incremental gains come from higher conversion rates, transaction frequency, and monetization efficiency compared to scenarios without AI.

From Token Growth to Profit Growth: Internet Giants' AI Monetization Pathways

The competitive advantages of TENCENT, Alibaba, and vertical platforms converge on their ability to integrate models into executable business processes. WeChat's "Xiaowei" assistant remains in an expanding testing phase, covering chat and file processing, native WeChat functions, and mini-program services, with trial collaborations involving JD.com, Meituan, Ctrip, and Trip.com. Critical operations still require user confirmation. The dedicated WeLM-80B model has approximately 80 billion total parameters with about 3 billion activated per token, demonstrating the direction of controlling inference costs for specific applications, though this cannot be directly converted into proportional cost savings.

On the Alibaba front, the bank recognizes Qwen's shopping assistant embedded within Taobao, covering the path from shopping inspiration to after-sales service, while also requiring return discipline on the large-scale customer acquisition spending for the standalone Qwen app. Meituan's "Xiaotuan" is transitioning from providing answers to enabling orders, ride-hailing, and reservations. OTA platforms can connect travel recommendations to inventory and bookings, and KANZHUN can embed AI into recruitment matching and employer processes.

While these advantages support commercial expansion, the analysts note that profit realization still depends on whether incremental revenue can cover model, inference, and marketing investments. Additionally, the "AI answer mode" based on chatbots or agents may divert traditional search advertising, while AI music and video generation models expand competitive content supply. Even if actual production costs decrease for traditional music and video enterprises, existing content barriers may face erosion.

Overseas model and application development is expanding "AI-handleable tasks" into more complex workflows. OpenAI's GPT-6 Astra emphasizes computer operation, browser usage, and multi-step task execution, capabilities expected to broaden the scenarios agents can serve. Efficiency improvements in task execution will also affect token consumption per task. Furthermore, OpenAI confirmed it suspended new subscriptions and upgrades for the ChatGPT Pro 20x tier at $200 per month effective September 10, with existing subscriptions continuing normally. The official announcement directly attributed this suspension to Astra's computational demands, highlighting the surge in computational requirements driven by expanding AI application token inputs.

Additional evidence of direct AI application demand expansion comes from global AI application leader Anthropic, which signed agreements with Google and Broadcom in April, expecting to add multiple gigawatts of next-generation TPU capacity starting in 2027. The agreement with Amazon includes commitments for over $100 billion in AWS technology purchases over the next decade and up to 5GW of additional capacity. In July, Anthropic announced a cooperation with AMD for up to 2GW of MI450 series GPU deployment, with the first 1GW expected to begin in the first half of 2027. These capacities will be deployed in phases, gradually converting into training and inference service supply.

The AI application token expansion driving AI infrastructure demand is clearly reflected in the strong earnings of industry leaders, South Korea's record semiconductor exports, and long-term capacity agreements. NVIDIA (NASDAQ: NVDA) reported second-quarter fiscal 2027 revenue of $96.2 billion for the period ending July 26, 2026, up 106% year-over-year, with data center revenue reaching $89 billion, up 117%. South Korean official data shows semiconductor exports grew 209% year-over-year in August 2026, accelerating to 270.1% growth in the September 1-10 period. These figures collectively reflect the combined effects of demand, pricing, and product mix.

For Chinese internet companies, the investment insight lies in following two transmission paths to revenue: large cloud vendors capturing enterprise model deployment, inference services, and application development needs, while consumer platforms use AI to improve advertising matching, order conversion, and merchant operational efficiency. The overseas data supports the judgment of active global AI investment and strong user computational demand in AI applications. However, the growth strength of Chinese platforms themselves should be verified through cloud business orders, AI feature retention, transaction conversion, and cash flow.

From the perspective of underlying technology and unit economics, the benefit mechanism for Chinese internet giants lies in serving larger real demand at lower successful-task costs while retaining the commercial value generated by transactions. A single shopping or travel agent task may require multiple rounds of retrieval, reading product and inventory data, comparing options, calling tools, and verifying results. Input tokens thus include not only user queries but also the materials and tool returns the model needs to process. Long-context prefilling increases computational work, while the generation phase is constrained by model weights, KV cache capacity, and memory bandwidth limitations.

Quantization, model routing, cache reuse, and batching can reduce some overhead, but identical token counts do not correspond to fixed computational costs. AI usage growth brings inference spending, but platforms must also control costs through technical optimization to convert commercial revenue growth into profit growth. The key question is how existing users, business data, and transaction systems help platforms improve transaction efficiency, combined with cost control, to determine whether incremental commercial revenue can generate incremental profits.

Based on Morgan Stanley's selected API pricing, simple Q&A single-token costs range from 0.0006 to 0.012 yuan, while large software agents or extended multi-agent sessions range from 1.2 to 24 yuan. These fees do not yet represent complete service costs. Platforms with existing traffic and authorized business interfaces can reduce customer acquisition costs and task execution friction, then use domain data to improve success rates. Higher success rates mean the same model call budget can generate more effective orders and commercial revenue through more efficient pathways.

As China's consumer AI enters a growth window worthy of serious attention, Morgan Stanley suggests that the most valuable investment metrics to track are incremental contribution profit per thousand AI tasks, repurchase and retention rates, and cash flow after capital expenditures.