Yuanbao withdrawn? WorkBuddy into
Qifan Business
2026.11.20 09:38
· From Shandong
Full text 5344 words
Text | Qifan Business
On the evening of November 18, Tencent President Liu Chiping admitted in an earnings call that Yuanbao did not meet the company’s initial success expectations and related investments had been significantly reduced. Another AI product, WorkBuddy, has delivered different results: its revenue in the last seven days has exceeded US$500 million in annualized terms, and its revenue in the third quarter has been more than five times that of the second quarter.
The different situations of the two products occurred during the period when Tencent’s AI investment was most intensive. In the third quarter, Tencent’s capital expenditure reached 81.433 billion yuan, a year-on-year increase of 527%, and free cash flow had a net outflow of 69.1 billion yuan. Computing power procurement, model training and product development are becoming an increasingly expensive bill.
Source: Tencent third quarter financial report
Tencent’s requirements for AI investment return have already emerged as early as the second quarter earnings call. Management emphasized at the time that subsequent investment in computing power needed to match actual business returns. Three months later, Yuanbao reduced investment, WorkBuddy accelerated commercialization, and small businesses began to enter the WeChat ecosystem at a lower cost.
Tencent is still willing to spend money on AI, but it is no longer willing to spend money on all AI products in the same way.
01 Things that Yuanbao cannot buy
As of the end of September this year, the combined monthly active accounts of WeChat and WeChat reached 1.449 billion. The social relationships, content consumption, payment and mini-program ecosystem formed around WeChat allow Tencent to have one of the most mature traffic distribution systems in the domestic Internet industry.
Another expectation was placed on Yuanbao: to establish an independent AI assistant entrance outside of WeChat.
The rapid growth of products such as Doubao and DeepSeek has gradually made direct questions to AI a new usage habit. Search, content generation and complex task processing begin to compete for user time that originally belonged to search engines, content platforms and office software. Tencent needs an AI product that can directly face users, which not only demonstrates the capabilities of the hybrid model, but also reserves a place for the next generation of Internet portals.
Tencent has invested considerable product resources in Yuanbao. Search capabilities continue to expand, external services are gradually connected, and products try to move from general question and answer to more complex task processing. Relevant materials in the third quarter show that Yuanbao Search has established a 100-billion-level index, and its monthly active users are still growing.
On November 18, Liu Chiping gave a different evaluation: Yuanbao did not meet initial expectations, Tencent has significantly reduced related investment, and its focus has shifted to improving product capabilities.
Yuanbao’s users are still growing, but Tencent has decided to reduce investment.
This set of information is more worthy of study than simply comparing the user sizes of Yuanbao and Doubao. Tencent has not announced Yuanbao’s independent customer acquisition cost, retention rate and payment conversion rate. The outside world cannot determine whether the problem lies in user quality, frequency of use or commercialization efficiency. What management has confirmed is that the product investment results have not met expectations.
Tencent's Internet business in the past was accustomed to looking for business opportunities from its user scale.
WeChat relies on relationship chains to build retention, games maintain payment through content and social relationships, and advertising business benefits from user scale and usage time. Between traffic, products and commercialization, a relatively mature operating system has been formed.
It is difficult for an independent AI assistant to directly copy this set of paths.
Users can install Yuanbao, Doubao and DeepSeek at the same time, or switch between different products. Models are updated very quickly, and a certain capability that leads today may be caught up in a few months. A marketing campaign can bring downloads and visits, but it may not necessarily change the user's long-term habit of using which assistant.
WeChat’s traffic can help Yuanbao gain exposure, but it cannot directly establish a migration threshold for it like a relationship chain. Users stay in WeChat because their contacts, chat history and daily services are already concentrated there. When using AI assistants, users are more concerned about whether the problem at hand can be solved.
The cost structure has also changed.
When a user sends a WeChat message, there is no need to call a large model for each message. When Yuanbao handles complex problems, searches for information, or performs tasks, model inference and tool invocation will continue to consume computing resources. As user usage frequency increases, service costs may also increase simultaneously.
Traditional Internet products often hope to expand the number of users as quickly as possible before looking for advertising, transaction or subscription revenue. AI assistants also need scale, but scale itself is not enough to prove business value. Without sufficiently stable usage demand and revenue sources, growth may also mean higher expenses.
Tencent has not released Yuanbao's independent operating data, so it cannot be concluded that it has suffered serious losses. Management's decision to reduce investment at least shows that the company is unwilling to continue to maintain its original investment intensity based solely on user growth.
Yuanbao is still improving search, service connection and task execution capabilities. It needs to give users clearer reasons to use it, rather than just being one of many general Q&A tools.
Tencent can bring users to Yuanbao, but it has not proven why these users must stay for a long time.
For a company that owns WeChat, re-establishing an independent entrance may not be easier than acquiring the first batch of users.
02 WorkBuddy starts turning work into income
WorkBuddy has already achieved commercialization results that Yuanbao has not yet disclosed.
Tencent disclosed that WorkBuddy has more than 10 million daily active users and daily consumption of large language model tokens exceeds 100 trillion. The annualized revenue in the last seven days has exceeded US$500 million. The third quarter revenue has been more than five times that of the second quarter. About 60% of China's top 100 companies have become customers.
The $500 million is the annualized run rate of the last seven days of revenue, not the annual revenue that has been realized. It still provides an important signal: some enterprises and users are already willing to continue to pay for AI capabilities.
China Unicom Huasheng's financial reconciliation work shows where this need arises.
After introducing the agent, the company hands over the forms and matching rules to the system for processing, and a financial staff member can complete the entire month's data collection in about two hours.
The case did not disclose specific procurement costs, and there were no independent audit results, so it was impossible to directly convert how much money was saved. What can be seen is that financial work that was originally scattered in different regions and required repeated verification has begun to be processed centrally through intelligent agents.
When companies purchase AI tools, what ultimately needs to be answered is not how many functions it has, but whether it can shorten working hours, reduce errors, or allow the same number of employees to complete more tasks.
WorkBuddy begins to shift the value of AI from answering questions to completing work.
Tencent is also trying to link charging methods with actual usage.
Traditional enterprise software is usually charged based on account number, term or functional module. If employees use it more, software vendors may not gain additional revenue. WorkBuddy attracts users through free quota and then sells Tokens to high-frequency users.
Complex tasks require more model reasoning and tool invocation, and companies may increase purchases as a result. Liu Chiping mentioned in the earnings call that some heavy users spend as much as 40,000 yuan per month purchasing tokens.
This number does not represent the average price per customer, but it shows that some users are willing to pay higher fees for high-frequency AI work.
Token charging is not without contradictions.
Enterprises want to use fewer computing resources to complete their work, and Tencent needs to generate sufficient revenue from these tasks.
Suning.com has already begun to manage this cost. In a case disclosed at the Tencent Global Digital Ecosystem Conference, Suning.com used the observability capabilities of WorkBuddy Enterprise Edition to troubleshoot extremely time-consuming tasks, adjust prompt words, models, and tool configurations to reduce token consumption by 10%.
This figure does not disclose both the quality of task completion and the actual cost changes, and cannot be directly equated to a 10% reduction in total costs. But companies have begun to focus on how much computing resources are needed to complete the same task, rather than just letting AI take on more work.
The reduction in tokens consumed by a single task may reduce customer fees and may also make companies willing to expand AI to more processes. Whether platform revenue increases or decreases depends on the number of tasks, charging levels and actual usage.
Tencent needs to handle two accounts at the same time: making customers feel that AI is worth buying, and allowing service costs to be controlled as model efficiency improves.
WorkBuddy's revenue increased fivefold in the third quarter, but Tencent did not disclose the number of paying customers, average customer price, customer retention rate and independent inference costs. Whether the revenue mainly comes from new customers or the expanded use of old customers cannot be determined from the existing data.
There is another frontier in the enterprise software market.
Businesses such as finance, supply chain, customer management, and human resources often run on professional systems such as ERP and CRM. If WorkBuddy wants to become a unified operating portal for these systems, it must cooperate with the original software manufacturers.
The Beijing News Shell Finance previously reported that after UF ERP is connected to WorkBuddy, users still need to log in to their UF account, and the relevant fees are also paid within UF’s original system. The revenue distribution mechanism among some partners is still being explored.
WorkBuddy can help users call enterprise software, but it does not mean that the revenue from the original software will be transferred to Tencent. Models, business systems, data interfaces and implementation services are provided by different companies, and the fees paid by customers will also remain in different links.
Entering the enterprise workflow does not mean having the right to charge for the entire workflow.
Tencent already has Tencent Cloud's enterprise customers, technical service team and infrastructure. Enterprise WeChat, Tencent Documents and Tencent Meetings also provide ready-made work scenarios. WorkBuddy can use these channels to enter enterprises without having to build a sales network completely from scratch.
In the third quarter, Tencent Cloud service revenue increased by 35% year-on-year. The official financial report mentioned that WorkBuddy demand, as well as more computing power being allocated to model-as-a-service and GPU leasing, have jointly promoted cloud business growth.
Tencent did not disclose WorkBuddy's independent contribution to cloud revenue, nor the allocation of costs among different businesses. Application revenue, model services, and cloud infrastructure may drive each other, but how much profit each creates cannot be separated from existing data.
China Unicom Huasheng's financial reconciliation explains why companies need AI, Suning.com's Token optimization reflects that customers have begun to calculate usage costs, and UFIDA ERP reveals the revenue boundary in the enterprise software ecosystem.
WorkBuddy has crossed the threshold of direct payment. What needs to be verified next is whether enterprises will continue to expand usage and whether revenue growth can cover increasing computing and delivery costs.
03 Tencent begins to recalculate the priority of AI
Tencent’s requirements for return on AI investment preceded the contraction of Yuanbao’s investment.
On the second quarter earnings call, management has proposed that model training requires an initial investment, and subsequent inference computing power purchases must consider actual business returns. Tencent still needs to build AI infrastructure, but it will not infinitely increase capital expenditures that do not match business needs.
The Beijing News Shell Finance previously reported that Tencent Chief Strategy Officer James Mitchell also said that the company is prioritizing investment in WorkBuddy while lowering the priority of other products in the new AI product portfolio.
During the third quarter earnings call on November 18, management further explained the investment in different products.
After the launch of the new Hunyuan model, Tencent began to observe actual usage demands brought about by the improved model capabilities. Yuanbao did not meet initial expectations and related investments have been reduced. WorkBuddy, with its rapidly growing usage and revenue, has reason to continue expanding.
What Tencent has adjusted is not the direction of AI investment, but the priority of obtaining resources for different businesses.
Financial data puts the change into clearer context.
In the third quarter, Tencent's Non-IFRS operating profit was 68.073 billion yuan, a year-on-year decrease of 6%. After excluding the impact of new AI products such as Hy, Yuanbao, CodeBuddy, WorkBuddy and Xiaowei, the operating profit of the same caliber was 83.3 billion yuan, a year-on-year increase of 8%.
The difference between the two sets of figures is approximately 15.2 billion yuan, reflecting the impact of new AI products on operating profits in the current quarter under Tencent’s disclosure standards. It is not the entire AI cash payout, nor can it be treated as a loss for Yuanbao or WorkBuddy respectively.
The similar impact previously disclosed by Tencent was approximately 8.8 billion yuan in the first quarter, approximately 10.5 billion yuan in the second quarter, and expanded to approximately 15.2 billion yuan in the third quarter. The difference in operating profits in different quarters continues to increase, indicating that new AI products have had an increasingly obvious impact on Tencent's overall profit performance.
Changes in cash flow are more intuitive.
In the third quarter, Tencent’s capital expenditure reached 81.433 billion yuan, and capital expenditure payment was 65.1 billion yuan. Advance payments for large computing power purchases resulted in a net free cash flow outflow of 69.1 billion yuan; after excluding related advance payments, free cash flow was positive 16.9 billion yuan.
Tencent's total cash at the end of the third quarter was approximately 525.7 billion yuan, and its mature businesses still have strong profitability. Companies do not lack funds to continue investing in AI. What really needs to be solved is where the funds should be invested and what they can get after investing.
Model training, application inference, and cloud service expansion all require computing resources. The computing power purchased by Tencent must not only support hybrid model iterations, but also meet the needs of internal products such as WorkBuddy, and also serve enterprise customers who purchase external models and GPU resources.
Computing power can be allocated among different businesses, but product demand does not grow at the same rate. Tencent must choose between long-term technical capabilities and short-term operating performance.
This choice is also reflected in Xiaowei.
As of the third quarter, the number of daily active users of small micro-enterprises exceeded 30 million, and they have entered more than 30 touch points within WeChat. Tencent is expanding its connections with services such as mini programs and WeChat stores, and using relatively lightweight models to control inference costs.
Liu Chiping said at the earnings conference call that even if small businesses are opened to all WeChat users in the future, their single-quarter costs are expected to be lower than Yuanbao’s investment level at the beginning of the year.
Yuanbao needs to allow users to actively open an independent application, while Xiaowei can enter the existing usage process of WeChat. Users already read content, communicate, pay and use mini programs in WeChat. AI can provide help in these scenarios without necessarily needing to create a new entrance.
Small businesses also don’t have to rely entirely on independent subscription fees.
If AI can improve content discovery, service matching and transaction processes, Tencent has the opportunity to gain indirect benefits through advertising, payments and other existing businesses. Liu Chiping once mentioned that if AI capabilities can bring an increase of 10% or 20% to WeChat’s existing business in the future, the commercial value will be considerable.
The related growth has not yet been verified by actual financial data. Microenterprises do not disclose independent revenue, nor their specific contributions to advertising or transaction conversions.
The several AI paths chosen by Tencent have shown different operating characteristics.
Yuanbao needs to continue to establish usage habits of independent products, WorkBuddy can obtain direct income from corporate work, and small businesses rely on WeChat to find opportunities to integrate AI with existing businesses. The hybrid model and computing infrastructure are behind these products, providing common technical capabilities for different applications.
Tencent did not announce whether Yuanbao’s reduced budget will be transferred directly to WorkBuddy, nor did it announce a reordering of all AI products based on profit margins. Model R&D and consumer products still require long-term investment, and value cannot be judged by just one quarter's revenue.
The management’s statement is enough to show that Tencent has begun to more clearly distinguish the investment pace of different AI businesses. The product's access to resources no longer only depends on whether it has the opportunity to become a new entrance, but also depends on the demand that has been demonstrated and whether the next stage of investment can create more value.
What Tencent did best in the past was to use WeChat, games and social relationships to build a huge user network and then find commercial income from it. AI has changed this familiar way of doing business: having traffic does not necessarily mean that you can establish the advantages of an independent assistant; having corporate customers does not mean that you can obtain all the income from the workflow.
Yuanbao’s investment has shrunk and WorkBuddy has received higher priority, but this does not mean that Tencent has found the final answer to the AI competition. Both products still need to prove their long-term value, and Xiaowei and Hunyuan also have their own unfinished tasks.
For a company that already owns WeChat, the key to the next AI competition may not be to recreate WeChat, but to find businesses that are truly worthy of continued investment.
Respect for the original work, if you are qualified, you should be appreciated.
According to the real-name requirement of the Cybersecurity Law, please bind your mobile phone number before posting a comment.
Express
09:37
Xi Jinping meets with Indonesian President Prabowo
PCB concept is partially active, Zhongjing Electronics is approaching the daily limit
09:36
Hong Kong large model stocks fell collectively, MINIMAX fell more than 10%
09:34
Huawei's concept of computing power is partially recovering, True Vision 4-connected board
The concept of computing power chips continues to adjust, and Temple shares fall by the limit for the second consecutive time
09:30
Foreign companies file 337 investigation application for specific sports shoes
09:29
A-shares opened: the three major indexes collectively opened lower, with significant adjustments in the optical communications, gold, and GPU sectors
09:25
The central bank launched a 379.5 billion yuan overnight reverse repurchase operation today
09:22
Hong Kong stocks opened: Hang Seng Index fell 1.57%, Hang Seng Technology Index fell 2.95%
09:10
State Post Bureau: Postal delivery business volume and revenue increased in the first 10 months
09:06
Spot silver's intraday decline expanded to 2%, at $41.99 per ounce
The MSCI Emerging Markets Equity Index is down 20% from its June peak
09:05
November LPR quotations are released: interest rates for 1-year and 5-year terms and above remain unchanged
09:01
Most domestic commodity futures closed higher, with coke and coking coal rising by more than 3%.
FTSE China A50 Index futures opened down 1.31%
09:00
今年以来西安口岸出入境外籍人员突破50万人次
苹果正研发自有AI代理,拟与OpenAI和Anthropic竞争
08:59
美光:2027年存储芯片供需形势会比今年更加紧张
习近平将同出席APEC领导人非正式会议的外国领导人举行双边会晤
08:57
三星任命存储与移动业务独立负责人
扫描下载App



