WeChat's AI transformation is Tencent's real high-stakes gamble

"Many self - media outlets have mentioned that Tencent has been slow and missed some opportunities in AI. Do you think we're really slow? What exactly is the second half of the game?"

Recently, at the Tencent Cloud AI Industry Application Conference, Tang Daosheng, the senior executive vice - president of Tencent Group, posed this question to Yao Shunyu, who had joined Tencent not long ago and served as the chief AI scientist.

Yao Shunyu paused for a moment and replied, "It seems this should be a question I ask you." He also said that AI is a long - term game. "I don't think ChatGPT and Claude Code will be the only super apps. There will definitely be a continuous stream of new opportunities emerging."

This seemingly light - hearted banter unexpectedly hit on the most core controversy of Tencent's AI in the past three years.

Since the emergence of ChatGPT, the evaluation of Tencent's AI has almost always been accompanied by a keyword - slowness.

Compared with Baidu, which was the first to release ERNIE Bot, Alibaba, which continuously increased its investment in the open - source ecosystem, ByteDance, which grabbed a large user base with Doubao, and the emerging DeepSeek, Tencent has not been very prominent for a long time.

This impression has even become a consensus in the capital market.

However, on June 2nd, after the news that WeChat's AI agent was about to be released, Tencent's stock price soared by more than 10% during intraday trading, and its market value increased by about HK$414.8 billion in a single day. However, just a few trading days later, Tencent's stock price began to decline, and most of the gains were erased.

Behind this sharp fluctuation reflects the market's most real ambivalent sentiment towards Tencent's AI. On the one hand, Tencent seems to have missed the most exciting first half of the large - model era; on the other hand, it holds the most difficult - to - replicate card in the entire industry - WeChat.

This is also the biggest mystery of Tencent's AI at present.

At this juncture, we review Tencent's AI path in the past three years and try to answer three questions: Where exactly is Tencent slow? Can WeChat be its trump card to catch up? And when the AI competition enters the second half, where does Tencent stand?

01 Is Tencent Slow?

In November 2022, ChatGPT was released. It reached one million users in five days and one hundred million monthly active users in two months.

At that time, Tencent was in a rare trough since its listing. Throughout the year, Tencent's revenue was 554.6 billion yuan, a 1% year - on - year decrease; the net profit attributable to shareholders was 188.2 billion yuan, a 16% year - on - year decrease. This was the first time since its listing in 2004 that Tencent had a simultaneous decline in annual revenue and net profit.

With the tightening of game licenses and continuous regulatory pressure, the stock price had evaporated by more than half from its peak in 2021 to the end of 2022. In an internal speech at the end of 2022, Ma Huateng criticized several businesses by name. "Cut many businesses if necessary. Don't blindly follow competitors."

It's difficult for a company that is reducing costs and increasing efficiency to respond to a technological revolution at the fastest speed at the same time. This is directly reflected in the release rhythm of the Hunyuan large model.

Baidu released ERNIE Bot in March 2023. Although the product was not yet mature, it preemptively occupied the public's perception of China's large models. Alibaba, Huawei, and iFlytek followed closely and released their products intensively.

In contrast, Tencent didn't officially release the Hunyuan large model at the Global Digital Ecosystem Conference until September 7, 2023, and made it accessible through Tencent Cloud. Nearly ten months had passed since the release of ChatGPT.

Moreover, Tencent's logic is to start internally and then expand externally. At the time of release, more than 50 Tencent businesses and products, such as Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Financial Technology, Tencent Meeting, Tencent Docs, WeChat Search, and QQ Browser, had been connected to the Hunyuan large model for testing.

This is in contrast to the practices of Baidu and Alibaba. While other companies were competing for the market perception of the Chinese version of ChatGPT, Tencent chose to use AI to transform its own ecosystem first.

This path choice has its internal rationality, but it also comes at a cost. In the public perception battle for large models in 2023, Tencent had almost no voice.

In 2024, Tencent began to try to enter the consumer market.

In May 2024, Tencent Yuanbao was launched. However, the result was not ideal. According to QuestMobile data, as of March this year, Yuanbao's MAU (Monthly Active Users) was about 57.346 million, while the MAUs of ByteDance's Doubao and Alibaba's Qianwen reached 345 million and 166 million respectively.

A very interesting detail occurred during the Spring Festival in 2026. Yuanbao spent a huge amount of money to launch a social fission activity of "sharing 1 billion yuan in cash red envelopes". The core mechanism was to attract new users through WeChat groups. However, within a few days after the red - envelope links flooded everywhere, they were restricted due to triggering WeChat's external link compliance rules.

The official WeChat account "WeChat Pie" issued an announcement: After receiving user complaints, Yuanbao's Spring Festival activity induced sharing, harassed users, and disrupted the ecological order. According to regulations, its links were restricted from being directly opened in WeChat.

This fiasco reflects, to some extent, Tencent's identity misplacement in the AI era: With a super - social platform of 1.4 billion monthly active users, why can't it incubate a super - app in the AI era?

The rise of Doubao is essentially a classical Internet - style entrance defense war. Relying on its abundant traffic dividends and the user - acquisition gene of its app factory, ByteDance managed to create an independent super - entrance, Doubao, to anchor the throne of consumer - end traffic in the AI era.

In contrast, Tencent launched the independent app Yuanbao and encountered growth bottlenecks, which was criticized by the industry as "begging while sitting on a gold mountain".

Tencent's biggest moat is WeChat. Asking users to leave WeChat and look for AI in another independent dialog box is not only using its own weakness to confront ByteDance's strength in user acquisition but also a huge waste of WeChat's natural scenarios.

02 The Card of WeChat

Tencent doesn't need to find an entrance for AI. The entrance is already there, and no competitor can replicate it.

At the earnings conference call for the third quarter of 2025, Liu Chiping described a clear vision: "WeChat will eventually launch an AI agent to help users complete many tasks within WeChat using AI."

He believes that WeChat's ecosystem has a very powerful communication and social ecosystem, with a large amount of data that enables the agent to understand users' needs, intentions, and interests; it has a very powerful content ecosystem, including official accounts and video channels; it has a mini - program ecosystem, which basically covers most use cases on the Internet; it has a business ecosystem that allows people to buy goods, and a payment ecosystem that allows people to complete payments almost immediately.

Liu Chiping said, This is almost an ideal assistant for users, understanding their needs and being able to perform all tasks within this ecosystem.

At an internal executive meeting in December 2025, the WeChat team summarized several trends for entering the AI field: WeChat must have built - in AI tools that do not rely on third - party systems; WeChat essentially consists of three parts: interpersonal social interaction, information acquisition, and efficiency tools. Among them, social relationships themselves cannot be replaced by AI, and the role of AI can only occur at the information and efficiency levels.

This statement reflects the WeChat team's understanding of its own boundaries: AI will not subvert the social core of WeChat, and its value lies in transforming the other two aspects.

In 2026, the built - in AI Agent of WeChat was accelerating its implementation.

In March, foreign media The Information reported that Tencent was secretly developing an AI agent for WeChat. The project was classified as top - secret, led by Zhou Hao, the technical leader of WeChat, and reported directly to Zhang Xiaolong. It was planned to conduct a gray - scale test in mid - 2026 and be fully launched in Q3.

In June, there were reports that Tencent was launching an embedded AI agent for WeChat. The prototype test had been completed, and the compliance approval process required for public launch would be initiated as soon as this month.

An informed person who had watched the early demonstration introduced that users can enter the dialog box of the AI agent by swiping right on the WeChat main interface. On this interface, users can input instructions, and the agent will automatically call millions of WeChat mini - programs to complete tasks such as finding a coffee shop according to taste preferences and price requirements and placing an order.

So, as the trump card of Tencent's AI layout, can the WeChat AI agent enable Tencent to catch up and have an impact on the AI battle among domestic giants?

Judging from the market reaction, many investors are obviously willing to believe this story. However, there is still a long engineering process between the vision and reality. There are a large number of WeChat mini - programs, and issues such as service quality, interface stability, merchant cooperation, payment processes, and profit distribution are all not trivial matters; at the same time, in the future, users, merchants, and platforms may all launch their own Agents, and the lack of constraints may easily lead to process chaos.

For the AI Agent to truly work, it needs not only an entrance but also the coordination of the entire chain. The interfaces of mini - programs need to be stable, merchants need to cooperate in accessing, payments need to be seamlessly connected, inference costs need to be controllable, and data calls need to be compliant.

Meanwhile, competitors are not standing still. Alibaba and ByteDance are accelerating their expansion into service scenarios. Qianwen has been deeply integrated into Alibaba's ecosystem, including Taobao, Alipay, Fliggy, and Gaode; Doubao has also been deeply integrated into the Douyin e - commerce ecosystem.

Ma Huateng said at the shareholders' meeting in the first quarter of 2026: "We thought we got on the boat a year ago, but then we found the boat was leaking. Now we feel we're on it, but we can't sit down comfortably. We still hope the boat can go faster." This statement precisely describes Tencent's AI situation.

Tencent has spent three years transforming from a bystander to a bettor. Now, what it needs to prove is that it can make up for the lost time in the second half.

03 The Second Half of the AI Game

Tencent doesn't fail to understand AI. Instead, it attaches great importance to the input - output ratio.

In the past two decades, one of Tencent's most successful business practices has been to quickly amplify its advantages after seeing the trend clearly. Whether it's in games, payments, or the industrial Internet, Tencent is better at making heavy - duty investments after the business model matures, rather than being the first to take risks.

This gene is also reflected in AI.

In the past few years, while promoting the research and development of Hunyuan, the launch of Yuanbao, and the transformation of internal businesses, Tencent has also maintained strategic flexibility. Around 2025, Tencent successively invested in technology companies such as Jiyiwei Semiconductor and Lightelligence, and also appeared on the shareholder lists of many large - model companies such as Moonlight, MiniMax, and Zhipu.

However, in terms of capital expenditure, Tencent has been much more restrained.

In 2025, Tencent's capital expenditure was 79.2 billion yuan, and its R & D investment was 85.75 billion yuan, both hitting record highs. In contrast, according to foreign media reports, ByteDance's capital expenditure in the AI field in 2025 was about 150 billion yuan, of which about 90 billion yuan was used for AI computing power procurement.

But this restraint doesn't mean that Tencent doesn't attach importance to AI. In the past two years, Tencent has been waiting for an answer to a question: How can AI make money?

In the past two years, the most core narrative in the large - model industry has been model capabilities. Parameter scale, benchmark testing, inference ability, download volume, and user scale have constituted the main theme of industry competition. Although the models have been continuously improving, the business model is still vague.

What really made the industry start to see changes was the emergence of the Agent wave.

Since the end of 2025, Agent frameworks represented by OpenClaw have rapidly emerged. Different from traditional chatbots, Agents can not only answer questions but also call tools, connect services, and execute tasks. From querying information to completing transactions, from content generation to automatically executing workflows, large models have for the first time started to transform from dialogue tools into execution systems.

This means that AI has for the first time had a relatively clear commercialization path.

Token calls, tool usage, task execution, enterprise subscriptions, and transaction sharing, each link may form a source of income. To some extent, Agents have allowed Tencent to see the end - goal of AI commercialization.

This is why since 2026, Tencent's response to Agents has significantly accelerated. Tencent Cloud quickly launched the OpenClaw one - click deployment service; the WeChat Agent entered the testing stage; products such as Yuanbao, Enterprise WeChat, and Tencent Meeting also began to be reconstructed around Agent capabilities.

Looking back at Tencent's product history in the past two decades, we can find a recurring path. Tencent may not always be the first to invent technology, but it often becomes the one that turns complex technology into mass - market products. This is true for QQ, WeChat, WeChat Pay, and mini - programs.

Today, this logic is being replicated in the Agent era.

If the competition in the past two years focused on model capabilities, then after 2026, the focus of competition is gradually shifting to ecological capabilities.

Whoever can make AI truly enter the scenarios where users have formed habits, keep developers and merchants on the platform continuously, and enable Agents to complete the entire closed - loop from understanding needs to executing tasks and then to completing transactions will have a better chance of becoming the winner in the next stage.

And this is precisely the direction Tencent is betting on.

This article is from the WeChat official account "Market Value List" (ID: shizhibang2021), written by Xiang Qing and edited by Jia Xin. It is published by 36Kr with authorization.