Tencent has already obtained the ticket to the AI era.

The five stamps on the ticket are all affixed. Tencent has completed boarding, and the ship has officially set sail. What is worth looking forward to next is where this ship of the AI era will eventually head.

On the evening of August 12, Tencent released its financial report for the second quarter of 2026. At the opening of trading the next day, its share price fell by 3.29%, and once dropped by more than 4% during the trading session.

This financial report that triggered the share price decline is actually not unsatisfactory at all: its single-quarter revenue reached 204.785 billion yuan, a year-on-year increase of 11%, exceeding 200 billion yuan for the first time in history; its Non-IFRS net profit hit 68.4 billion yuan, up 9% year on year. What really made the market uneasy are two figures: free cash flow stood at -13.8 billion yuan, the first time it has turned negative since the company's listing; capital expenditure reached 527.84 billion yuan, a sharp year-on-year surge of 176%, 60% higher than the market expectation of 321 billion yuan.

Investment banks split into two camps that very night. Citi raised the target price to HK$765, noting that "the resilience of core businesses is prominent, and the effectiveness of AI applications is emerging"; Morgan Stanley cut its target price by 15%, worrying that "earnings from the second half of 2026 to 2027 may stay flat". For the same company and the same financial report, there is a divergence of HK$215 between the two valuations of HK$765 and HK$550.

The divergence centers on only one point: Is the huge sum of money Tencent is spending now buying it a ticket for the AI era, or is it paying to build a dock for a ship that may never arrive? To answer this question, we need to go through this financial report line by line.

1 One Sum of Money, Two Calculation Methods

The first thing Tencent President Martin Lau did at the earnings call was to present investors with a new calculation framework.

He divided Tencent's businesses into two segments: one is the mature core businesses — games, advertising, social networking, and fintech, which maintain steady growth and generate abundant free cash flow on their own; the other is AI-native businesses, including self-developed large models, brand new AI applications, and supporting computing power infrastructure, which belong to one-time concentrated investment.

Under the first calculation method, Tencent's financial report looks "very unsatisfactory": Non-IFRS operating profit reached 75.6 billion yuan, with a year-on-year growth of only 9%, and the operating profit margin dropped from 38.5% in the previous quarter to 37%; free cash flow turned negative directly.

Under the second calculation method, the picture is completely different: after excluding the loss impact of new AI products, Non-IFRS operating profit is about 86.1 billion yuan, representing a 19% year-on-year increase. The AI business lost about 10.5 billion yuan in this quarter (compared with 8.8 billion yuan in the previous quarter), dragging down the group's profit growth rate by a full 10 percentage points — but this is a result of active choice, not a deterioration of the core businesses.

To use an analogy: a person with an annual salary of one million yuan spends 300,000 yuan a year on a part-time doctoral program. Looking at his bank statement, you may think he is "getting poorer"; but looking at his payroll, his income is actually rising. Tencent is exactly at this stage now.

The persuasiveness of this "two-segment division" depends on a prerequisite: the investment in AI can really bring returns in the future, which is also the part the management spent a lot of time demonstrating that night. Martin Lau made a confident remark throughout the earnings call: "A batch of computing power equipment we paid deposits for a few months ago can now be resold at a profit of more than 30% compared with the original purchase price."

This remark seemingly refers to the shortage in the computing power market, but actually points to one key fact: the huge sum Tencent spent on purchasing computing power has a guaranteed bottom line. Even if all AI businesses fail, the equipment can still be resold at a profit; to go a step further, "even if we stop developing self-developed applications and simply lease out our computing power, the related business itself can be profitable."

For this huge investment, the worst case is no loss — that's the confidence for Tencent to spend 52.8 billion yuan on capital expenditure in a single quarter.

2 Why Not Lease Out Computing Power

But Tencent is not planning to make money by leasing out computing power, at least not for now.

James Mitchell, Chief Strategy Officer of Tencent, explained this choice at the earnings call: "If we lease our computing power to third parties, we can recover the equipment depreciation cost almost immediately. But in fact, we have a different layout, and we are executing a longer-term strategy."

The management has set a clear priority for the computing power allocation under this strategy: The first priority is to train the Hunyuan series of large models; the second priority is to run inference for WorkBuddy; the last priority is to lease cloud computing power.

This priority ranking itself is a strategic declaration. Leasing out computing power brings meager profits, while training models stakes on the discourse power of the next decade.

The bet is paying off. The official version of Hunyuan Hy3 released in July has ranked among the top three in the world in terms of token consumption calculated by OpenRouter. The management announced a more aggressive roadmap at the earnings call: Hy4 to be released later this year will achieve a full jump in parameters and performance, with "performance exceeding that of larger-scale competing products"; the subsequent Hy5 will continue to approach and finally reach the top level of the industry. After reaching SOTA (State-of-the-Art), Tencent will build a gradient model matrix with multiple specifications — models of different parameter scales will adapt to different costs, scenarios and product lines, and each tier can generate stable profits.

The ambition behind this roadmap is no longer just to "keep up with the first echelon", but to "stand in the first echelon and then make profits by leveraging scale and cost structure".

3 Yuanbao Downgraded, WorkBuddy Upgraded

Underlying the models are applications. At this layer, Tencent has just completed a key internal horse race ruling. The management confirmed at the earnings call that "after we found that WorkBuddy is gaining strong momentum, we decisively tilted resources to it and lowered the priority of other new AI products in our portfolio."

Among the "other products" whose priority has been lowered is Yuanbao.

It is worth noting that Yuanbao is not a loser — its DAU exceeded 50 million in February this year, and its MAU reached 114 million, making it one of the largest AI assistants in China in terms of user scale. According to conventional logic, a product of this scale should deserve more resources. But Tencent's choice is to focus all its firepower on WorkBuddy.

The reason is hidden in the data. The financial report shows that the monthly visits of WorkBuddy on PC end reached 20.97 million, ranking first among domestic AI office agents. The user retention rate disclosed in the Q1 financial report was as high as 60%. More importantly, its business model has been proven viable — the gross profit margin of WorkBuddy's paid user business and MaaS business is already on a par with the overall gross profit margin of Tencent Cloud at the current stage.

On one side is a C-end assistant with huge traffic but unclear monetization path, and on the other side is an office agent with high retention rate and high gross profit margin. Tencent chose the latter, and Martin Lau positioned it far more than a simple tool: "WorkBuddy is essentially a brand new platform, a highly flexible workstation adapted to general artificial intelligence, with its core value lying in the execution of tasks. Its underlying layer is a scheduling framework that can call multiple large models to solve complex tasks — Hunyuan is one of the core models, but not the only one."

This choice reminds people of the most famous self-revolution in Tencent's history: When WeChat was born, QQ was still at its peak, but Tencent still bet all the resources of the whole company on WeChat. The key to the horse race mechanism has never been equal distribution, but to dare to place heavy bets on the winning projects and cut the underperformers when the outcome is initially clear.

4 The Second Amplification of WeChat

If WorkBuddy is Tencent's offensive weapon in the AI era, then WeChat is its deepest moat. And this moat is being re-excavated by AI.

On June 20, WeChat launched its native AI assistant "Xiaowei", which is based on the self-developed vision-language model WeLM, supports text and voice conversations, and can directly operate WeChat's native functions — opening mini-programs, sending messages, making WeChat calls, AI searching, and summarizing group chats. It is currently in gray test, and the plan is to expand its access in the third quarter.

This is the biggest revision of WeChat in the past six years. But what is really worth recording is the demonstration Martin Lau made for it at the earnings call: "In the PC era, QQ was only a social communication tool. After entering the mobile internet era, WeChat was born, and the WeChat ecosystem amplified the overall value of QQ by more than ten times. Now ushering in the AI era, the WeChat ecosystem is embracing its second huge growth opportunity, and it will evolve into an application and ecosystem centered on AI in the future."

There is a bold inference hidden in this remark: every migration of computing platforms (PC → mobile → AI) means an order-of-magnitude amplification of ecosystem value. If this rule holds, WeChat's AI transformation is not just a function upgrade, but a repricing with tenfold leverage.

Martin Lau also outlined a broader picture: "In the future, users only need to issue a complex instruction to their exclusive agent, and the agent can automatically complete the entire transaction process. The vast majority of mini-program merchants will also deploy exclusive merchant agents. In the long run, merchant agents can directly connect with user agents."

Just imagine the following scenario: you say "Help me book a table for four for dinner on Saturday night", your agent will automatically open the corresponding merchant agent on the Dianping mini-program, compare prices, reserve the table, place the order and make payment, with no manual intervention throughout the process. Between the tens of millions of mini-program merchants and 1.4 billion monthly active users on WeChat, human beings will no longer be needed as the "operation interface".

At that time, the commercial value of the WeChat mini-program ecosystem will be reactivated in a way that is impossible to estimate today.

5 Five Stamps on the Ticket

Back to the original question: Why do we say Tencent has got the ticket for the AI era?

Putting this financial report and the management's statements together, we can see a complete layout. Competition in the AI era essentially requires five capabilities — models, applications, computing power, scenarios and capital. The vast majority of companies only have one or two of them: OpenAI has models but no scenarios, ByteDance has scenarios and is supplementing its model capabilities, and startups have applications but no computing power. Tencent is one of the very few companies that have all five capabilities.

In terms of models, Hunyuan Hy3 ranks top three in the world, with a clear roadmap for Hy4 and Hy5; in terms of applications, WorkBuddy has achieved proven retention rate and gross profit margin, and CodeBuddy leads among AI programming tools in China; in terms of computing power, the high-end GPUs locked in advance have generated a 30% floating profit in the tight market, and NPO super nodes will be deployed by the end of the year; in terms of scenarios, WeChat's 1.4 billion monthly active users make it the largest AI application distribution entry in the world; in terms of capital, the game business generated 65.9 billion yuan of revenue in a single quarter with a year-on-year growth of 11%, and the advertising business has maintained double-digit growth for 11 consecutive quarters — the operating cash flow generated by these mature core businesses every year is the granary supporting the AI arms race.

Without any of these five stamps, the ticket will not be complete. Interestingly, four of them — applications, computing power, scenarios and capital — are not new gains in this quarter, they are the stock assets accumulated by Tencent over the past two decades. The arrival of the AI wave has suddenly given these assets new uses. The so-called "getting the ticket" is half from the aggressive investment of hundreds of billions of yuan in this year, and the other half from the precipitation and accumulation of 20 years of preparation.

6 A Ticket Does Not Equal the Other Shore

Of course, this ticket is not almighty.

At the earnings call, when James Mitchell was asked about the advertising growth rate, he took the initiative to cool down expectations rarely: "The advertising revenue growth rate has always fluctuated, and it is not recommended to make linear extrapolation based on any single-quarter growth rate." This restraint shows that the management is clearly aware that the 22% advertising growth rate comes from both the dividend of AI delivery tools and the dividend of traffic release from Video Account, the former is sustainable while the latter has a ceiling.

The real test lies in a longer time dimension. Martin Lau promised that the capital expenditure for AI-native businesses is "mainly one-time concentrated investment in this year and next year", and subsequent inference computing power will only be added under the premise that "considerable returns can be achieved" — this means that around 2027, Tencent must show the market a clear evidence chain that AI revenue can take over from the core businesses, otherwise the statement of "one-time investment" will face a trust discount. The gap between Hunyuan and the world's top models, the expansion of WorkBuddy from office scenarios to general scenarios, and the experience polishing of WeChat Xiaowei from gray test to full release, none of these links can fail.

But these are all problems to be solved to "reach the other shore", not doubts about "whether you can get on the ship".

The railway bubble in the 19th century burst many times, but the railway itself changed the United States; the internet bubble in 2000 destroyed countless companies, but made the survivors dominate the next era. History has repeatedly proved that huge capital expenditure in the era of technological revolution is the enemy of profits in the short term, but the price of admission in the long term. The real question has never been how much money you spend, but whether the money you spend has brought you something that others cannot take away.

What Tencent has got this time are the top three models in the world, commercialized agents, computing power locked until next year, and an upcoming AI-enabled WeChat ecosystem with 1.4 billion users.

The five stamps on the ticket are all affixed. Tencent has completed boarding, and the ship has officially set sail. What is worthy of our long-term attention next is where this ship will eventually head.

This article is from the WeChat official account "Lishang Business Review" (ID: libusiness), Author: Zhang Fan, Editor: Pingfan, published with authorization from 36Kr.