Ma Huateng changes Tencent’s AI channel for the fourth time
Muhe Business Finance
2026.11.21 15:55
· From Beijing
Full text 5111 words
Text | Muhe Business Finance, author I Gong Zheng, editor I Yang Jing
On November 11, at Penguin Island in Shenzhen, the internal venue of Tencent's 28th Anniversary LinkTime, Ma Huateng, who was still distributing anniversary cakes during the day, showed another side that night: "The waves have come over. The waves in the distance have swallowed up the hills one after another. You can't just sit and enjoy it. The land battle has turned into a naval battle, and the army has turned into a navy."
The boss is good at creating a sense of crisis. I don’t know whether this kind of provocation method is still suitable for a large company like Tencent, which is said to be “suitable for retirement”. But this time, the market is willing to believe that Tencent is really anxious.
This is not a whim of Ma Huateng.
Looking at his past statements since this wave of AI, we know that this is the fourth turn he has had to make after overturning his judgment on AI three times in the past three years. Tencent's eagerness and anxiety today, to a certain extent, come from the rounds of decisions made by the big boss back then.
The rapid progress from competitors is just an external variable. The key lies in yourself, how you think, and then decide what to do.
Everyone says that AI has just begun, and Tencent still has a chance. At least it now has a WorkBuddy to support it, and there is still fire to ignite.
Before becoming anxious, Ma Huateng had already changed the course three times.
"AI is not an independent track. It is a bottom-level plug-in that serves games, advertising, recommendations, and medical imaging. It is used to optimize existing business efficiency. It does not make a separate AI entrance product for the C-side."
Before ChatGPT swept the world in 2023, the media summarized Ma Huateng’s judgment on the AI wave at that time based on his various statements.
In his view, AI is not an independent track, but a bottom-level plug-in that serves games, advertising, recommendations, and medical imaging. It is used to optimize existing business efficiency and does not solely produce AI entrance products for the C-side.
The command method relies on AI Lab to integrate visual and NLP capabilities (natural language processing) into internal scenes. Intelligent advertising placement and short video content review are all products of this set of ideas.
Looking at it now, is this kind of judgment "a big mistake"? But everything cannot be viewed apart from the time and space at that time.
Before 2023, AI has actually already quietly gone through two cycles in the country. The role of AI at that time was indeed a kind of technical empowerment.
One is the rise of voice + computer vision craze between 2012 and 2022.
AlphaGo beats Lee Sedol, breaking the national level. The four domestic AI tigers—SenseTime, Megvii, Yitu, Yuncong, etc.—are taking off rapidly.
Ordinary people’s memories focus on three aspects: facial recognition is being implemented everywhere, voice transcription is available at every forum venue, Baidu and others are launching self-driving cars, and Carrot Run is the first to take to the streets.
Picture | Face recognition in various buildings has only happened in the past ten years.
The second wave of enthusiasm is between 2018 and 2021. Baidu, Harbin Institute of Technology, Zhiyuan, etc. began to make pre-training models.
The boundary of the era at that time was that everyone did pre-training, and the goal was not to make a general-purpose C-end product that can chat, but to train the base and split it into capabilities to provide vertical businesses such as search, recommendation, customer service, review, and medical text. However, this wave is more limited to the technical circle and cannot be felt by ordinary people.
Perhaps because of this situation, it has led to "paralysis" of Tencent's judgment on the general trend of AI.
It has not set aside an independent main line of large models - that is, Tencent has not set aside a separate long-term track with independent budget, independent computing power, independent team, and independent assessment goals. It is specifically used to develop a general large base and is not dependent on existing businesses such as WeChat, games, and advertising, allowing it to make no money in the short term and not immediately improve business efficiency.
This is about the character of Tencent. Tencent has always been accustomed to waiting for the trend to be implemented in mature scenes, rather than rushing to the front line to define new trends.
However, in 2023, after ChatGPT detonated the industry, Ma Huateng's awareness rapidly upgraded.
At the shareholder meeting in May 2023, he left a widely circulated judgment: At first, I thought this was a once-in-a-decade opportunity for the Internet. The more I thought about it, the more I felt that this was an industrial revolution similar to the invention of electricity that had not happened in hundreds of years. But taking out the light bulb a month early will not matter in the long run. Many companies are rushing to release products just to boost their stock prices. Tencent does not make this choice.
Although he is not active physically, it is Ma Huateng’s “first major turn” in his thinking: he admits that the magnitude of AI far exceeds that of ordinary Internet trends, but chooses to play slowly, polish the bottom layer, and refuses to follow the trend and grab the starting spot.
Then in September, the Hunyuan large model was officially released. At that time, Wen Xin Yi Yan and Tong Yi Qian Wen had been released one after another, seizing the high ground of public opinion. Overseas, OpenAI continues to iterate, and GPT-4 is implemented.
However, Tencent is immersed in the psychological atmosphere of "I choose to do it". It has integrated the ability to use large models into its internal business and is not in a hurry to launch independent C-end products.
This is actually another manifestation of Tencent's "habitual path dependence" - using its familiar ecological playstyle and waiting for the scene to mature before harvesting.
There is a correct part in this set of judgments: the ultimate value of large models must be implemented in the scene, and simply grabbing the demo has no long-term significance.
But there are also misjudgments. Looking at it now, the ecology, developers, and user mentality of large models will quickly solidify during the launch window period.
If you are slow, you will fall behind in the starting line of C-side users and developer ecosystem. This step paved the way for Yuan Bao's subsequent pursuit pressure.
Ma Huateng's third channel change occurred from 2024 to early 2026, and the core symbol is "Betting on Yuanbao."
Seeing Doubao and Kimi quickly "getting up high", Ma Huateng revised his previous judgment of "not rushing to make a C-side entrance".
He decided to benchmark the mainstream large-model apps and launch the independent product Yuanbao. He also used the fission experience of WeChat red envelopes and used 1 billion yuan in red envelopes to attract new users, trying to migrate the social relationship chains of WeChat and QQ into AI-native scenarios.
This method of injecting chicken blood does indeed bring about a "gust of wind". Yuanbao was once connected to the DeepSeek model, which spurred a surge in users in the short term, with peak daily activity reaching 50 million.
But big bosses are sometimes "ruthlessly educated" by the market. After the red envelope activity ended, users continued to fall back.
QuestMobile data shows that in September this year, Doubao had 419 million monthly active users, Qianwen 158 million, DeepSeek 123 million, and Yuanbao only 37.33 million. The frequency of user openings was also significantly lower than Doubao. The outcome of the "entry" competition has begun to take shape.
At the shareholder meeting in May this year, Ma Huateng frankly reflected on himself: "It turned out that a year ago we thought we were on the boat, but then we found that the boat was leaking. Now I feel like I am standing on it, but I can't sit down. I hope the boat speed is faster."
This turn only proves one thing: the Internet is the Internet, and AI is AI. The gameplay of Internet traffic fission is not fully suitable for large model tracks.
When users choose AI products, the core is model ability, reasoning quality, and tool capabilities, rather than the social relationship of "if you give me a red envelope, I will use it." In order to verify this conclusion, Tencent paid extremely precious time and computing power costs, and even missed the window period.
After three rounds of review, we can see that AI has a clear curve of circulation in Tencent, a large company.
Before 2023, AI is a tool.
From 2023 to 2024, AI will be a revolution, but there is no rush.
From 2024 to 2026, we will directly build an independent APP and use the social fission traffic strategy to pursue it.
"Um, which one of these is not the boss's intention to implement it?"
The fourth course change
It's the boss who says the opposite. At the LinkTime meeting, it was the boss himself who shouted, “The Army is going to become the Navy.” It was the boss himself who shouted this fourth channel redrawing.
The core decision this time can be summarized in one sentence: it is to abandon the route of independent C-side large-model apps like Doubao and Kimi, and no longer bet on an independent AI entrance. This is mainly because the gap is too big and the market capacity is almost eaten up by friendly forces.
Tencent organized manpower re-engineering, with half of the force continuing to guard the cash cow "army" of games, advertising, and social networking, while the other half served as the "naval force" to engage in AI-native business, embedding agents into the WeChat ecosystem, mini programs, and enterprise service links, and taking the parallel route of B-side WorkBuddy and C-side small and micro businesses.
Why do we have to go this way? We might as well put ourselves in Ma Huateng's perspective and look at the industry. Each company has taken a differentiated route and established its own position.
Byte Doubao, relying on the short video ecology, has firmly captured the minds of popular C-end AI assistants and has formed stable user habits.
Alibaba Tongyi Qianwen relies on Alibaba Cloud and e-commerce transaction links to embed AI into business operations and local life. The open source model Qwen has long led the world in downloads of Hugging Face, and the developer ecological moat is very deep.
Baidu Wenxin, along the line of search + enterprise services, continues to deepen its efforts in government and enterprise, and relies on the full-stack AI cloud to win the bid for central enterprise projects.
Startups such as DeepSeek, Kimi, and Zhipu are taking the open source base route to quickly capture the developer community.
In comparison, Tencent’s core moat is WeChat’s social relationships. However, if search, information acquisition, consumption decision-making, and office productivity are all taken over by external AI agents, WeChat may slowly become a mere communication channel—the traffic is still there, but the value outlet is intercepted by others.
This is Ma Huateng's biggest worry and even fear - "the wave swallows up the hills." It’s not that Yuanbao cannot beat Doubao today, but when AI becomes a new generation of interaction paradigm, if Tencent cannot find a native AI carrier that adapts to its own ecosystem, the traffic advantage it has accumulated over the past two decades will be slowly eroded away.
Of course, Tencent has not lost so much that it has no shorts left. God has given it a WorkBuddy, which has already shown signs of commercialization, with annual revenue exceeding US$500 million. Hunyuan Hy4 has achieved good results in the domestic rankings of Agent Arena. Xiaowei is embedded in WeChat and relies on mini programs, payments, and video accounts. It is an ecological container that is difficult for others to copy.
But these chips are still "chess pieces used to play chess" and are not enough to form a crushing "king". Opponents are not just waiting around, everyone is simultaneously increasing their computing power, talents, and scenarios.
Tencent must wait for no time, change the organizational atmosphere of big factory pensions, and use it to block positions, so as to leave a way for itself to survive. After all, I have changed course three times in the past. I believe Ma Huateng already knows that the window period of AI changes faster than that of mobile Internet.
How should Tencent fight next?
Putting aside the boss's "loneliness", this fourth turn, in the eyes of capital markets and industry observers, is a multiple-choice question full of tension: short-term profits are under pressure, and long-term value is highly uncertain.
Judging from Tencent's just-released third quarter report, capital expenditures were 81.4 billion yuan, and AI computing power procurement brought free cash flow to a periodic negative turn; after excluding the AI business, the operating profit of the old business increased by 8% year-on-year, but AI-related businesses brought a profit drag of about 15.2 billion yuan, higher than the 10.5 billion in the previous quarter.
Many investment banks such as UBS and Morgan Stanley have lowered their target prices. Their core concerns are the same: AI investment is huge and the return cycle is long, which continues to eat up short-term profits.
But what is optimistic is that most institutions still maintain overweight and buy ratings. The logic is: the basic market of games, advertising, and financial technology is strong enough to withstand continued money burning. The current stock price has priced in the pain of AI investment in advance.
As for the opinions of industry insiders, they are divided into two groups.
One group is cautious. Tencent has missed the best window for independent C-side AI entrance. The route of small micro-enterprises embedding WeChat is expected to be restricted by WeChat product restraint, privacy, and experience, making it difficult to let go like an independent App.
Whether the Agent ecosystem can come out depends on whether small program developers and merchants are willing to reconstruct services in the WeChat system. This is a long ecological cultivation project, "slow work makes careful work", whether Tencent is willing to turn into a "patient company" this time, and so on.
Picture | Will Zhang Xiaolong also change his product restraint more?
There is also a group of people who give "optimistic expectations". Tencent's advantage has never been to build a new app from scratch, but to build an existing ecosystem. Transactions, private domains, and service links in WeChat are natural scenarios for Agent implementation. AI shopping guides, smart services, and personalized consumption assistants can all be integrated into mini programs without requiring users to download new software.
Take the consumption agent as an example. It does not necessarily have to grow into an independent app like Muse, but is hidden in the WeChat that everyone opens every day.
The specific measures Ma Huateng has given this round currently fall on the three lines of organization, computing power, and products.
At the organizational level, in March this year, the AI Lab, which had been in operation for ten years, was canceled and merged into the Hunyuan system. Yao Shunyu was introduced as the chief AI scientist. The computing power is uniformly dispatched by the headquarters and is no longer scattered among various BGs.
In terms of employment, there is a mandatory two-line division of labor, with half of the people focusing on the basics and the other half transitioning to AI-native business. The team's skills must be adapted to naval warfare and the old ideas of traffic operation in the past will no longer be followed.
On the product line, it’s a two-way bet. The B-side continues to amplify the commercialization of WorkBuddy office intelligence and seize the enterprise productivity market; the C-side focuses on WeChat Xiaowei, embedding the Agent into the WeChat native environment, and connecting with mini programs, Souyisou, and video accounts, and no longer concentrates resources on the independent App Yuanbao.
On the model side, we continue to iterate on Hunyuan, strengthen Agent, multi-modal, and tool calling capabilities, take into account the open source ecosystem, and attract developers.
It is foreseeable that when a company that is good at social networking and traffic abandons the traffic strategy and turns to the native AI ecosystem, it will undoubtedly be a huge challenge whether it can develop product intuition that is suitable for the era of intelligent agents. Tencent will definitely face pain.
Many people discuss Tencent AI and like to struggle with one question: Is it considered a laggard?
Perhaps a more appropriate perspective is: Ma Huateng is using four self-corrections to complete a cognitive iteration. He accepted the cost of trial and error, bet on Tencent's core ecosystem, and bet on a set of Agent answers of his own.
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
16:02
South-to-North Water Diversion Project "Tianhe" large model completed upgrade
15:46
Iranian Foreign Minister: Enriched uranium will not be shipped out of Iran
15:40
China Copper's 10 billion yuan small public debt offering application was accepted by the Shanghai Stock Exchange
15:17
“GTA6” drives PS5 price to continue to rise Pro version rises to 16,000 yuan
15:03
CITIC Food and Beverage Investment Strategy for 2027: Restructuring the paradigm at the bottom of the cycle, maintaining positive dividend growth and surprising growth
Human resources service direction national artificial intelligence application pilot base unveiled in Chongqing
13:56
US startup GG sues Apple in India, accusing NameDrop of infringement
13:40
The U.S. Securities and Exchange Commission is reported to be investigating potential accounting issues at Nvidia
12:57
People in many European countries went on strike, affecting transportation and other public services.
12:51
Russian Ministry of Foreign Affairs: Heads of State diplomacy plays a special leading role in Russia-China relations
12:40
Hongmeng Smart Xingjie V8 is launched, priced from 319,800 to 429,800 yuan
12:38
Argentine government approves country’s largest investment project
12:36
CITIC Securities: It is expected that a national mechanism electricity price policy will be launched in 2027
12:15
Bitcoin fell below $56,000, down more than 6% on the day
12:05
The National Development and Reform Commission held a video conference to supervise the payment and use of government investment funds and project reserve work
11:44
Bitcoin falls below $57,000, hitting lowest level since September 2024
11:36
The Mongolian section of Hanoi has entered the cold flood season from 2026 to 2027
11:11
软银孙正义曾探讨以1万亿美元收购SK海力士
11:05
美货运“龙”飞船脱离空间站启程返回地球
10:31
美总统特使与卡塔尔高官讨论美伊斡旋工作
扫描下载App



