Ma Huateng's anxiety is actually a kind of "epochal vertigo".
Pony Ma hasn't been this serious for a very long time.
The other day, Tencent ushered in its 28th birthday, a day that was supposed to be celebrated collectively by the entire company. However, in their internal speeches, Pony Ma and other core management threw out heavy phrases such as "life and death survival".
On Maimai, someone claimed "Pony Ma hardly wore a smile the whole time".
Tencent has not made a public response to what the actual situation is. But judging from the on-site transcript, the wording used by Pony Ma and others is full of a strong sense of urgency, and one can imagine how serious the atmosphere was.
For example, "the waves have already rushed over", "the waves in the distance are already swallowing hills one by one, you can't still sit there and enjoy the view", "everyone must hurry up, stand up quickly to prepare for new tactics and find new high ground", "land warfare turns into naval warfare, and the army turns into a navy".
In fact, if you only look at the latest performance, Pony Ma does not need to speak so heavily.
In the third quarter, Tencent's performance was fairly good: total revenue reached 207.3 billion yuan, a year-on-year increase of 8%.
Among the three major segments, marketing services revenue increased by 19%, followed by fintech and enterprise services at 9%. Value-added services, mainly driven by games, performed flat, but still posted a 3% increase.
On the profit side, gross profit was 115.9 billion yuan, up 7% year on year, with a gross profit margin of 56%; operating profit was 59.1 billion yuan, down 7% year on year, mainly due to a large increase in operating expenses for new AI products; net profit was 68.4 billion yuan, down 6% year on year.
In the previous quarter, Tencent's net free cash flow outflow was 69.1 billion yuan. This result was also expected: it invested tens of billions of yuan in AI-related businesses such as large-scale computing power procurement, which dragged down free cash flow.
Overall, this is not an exceptionally outstanding performance, but it is not far from external expectations, and all negative factors can be attributed to the long-term layout centered on AI. It is far from bad enough to make Pony Ma furious.
Pony Ma's anxiety over the past period of time is not aimed at the current performance, nor at a certain team or a certain business of Tencent.
A more reasonable explanation is:
It stems from the vertigo caused by the changes of the era — the giants of the mobile Internet era, who are not fully prepared at all, have been thrown into the surging waves of the AI era.
This sense of vertigo is not unique to Tencent, ByteDance, Alibaba and other large companies also have it.
The common problem for the giants is that they need to guard their old businesses to make stable profits to support their operations, while also making good progress in the new AI business to grab a ticket for the new era.
However, after a period of vigorous promotion, large companies have found one after another that the effect cannot be said to be completely non-existent, but it is far from satisfactory.
Pony Ma's anxiety is not personal, but an era-wide dilemma spreading across the technology industry.
1
Every time it releases its performance, Tencent will name several "top students" as highlights.
In this quarter's financial report, the recognized highlights include the Hy4 series foundational model, the office Agent platform WorkBuddy, Channels, as well as new and old games such as Delta Action, Valorant: Source Energy Operation and Honor of Kings.
This is not an overstatement.
Multiple tests show that although the Hy4 launched not long ago is not yet SOTA, it already has the strength to rank among the first camp of global large models. WorkBuddy firmly occupies the first place in its niche track, with MAU exceeding 32 million.
However, on two tracks where competitors are running very fast, Tencent is still moving somewhat slowly — this may have caused Pony Ma to experience FOMO sentiment to a certain extent.
One track is AI cloud services.
Tencent stated in its financial report that the revenue growth of enterprise services in the previous quarter was mainly driven by a 35% year-on-year increase in cloud service revenue, benefited from rising demand for AI-related services and continuous expansion of the international market.
But as we all know, over the years, the cloud has never been Tencent's strength, nor its development priority.
The arrival of the AI era has not really changed this situation.
According to data from Sullivan, in the domestic full-stack AI cloud service market in the first half of this year, Alibaba Cloud accounted for 42% of the market share, Volcano Engine accounted for 20%, and Baidu Intelligent Cloud accounted for 16%. The three together occupy more than 3/4 of the total share. Tencent Cloud only has 5%, ranking fifth after Huawei Cloud (9%).
Therefore, Tencent's cloud service revenue increased by more than 30% in the previous quarter, more or less benefited from the low base effect.
In fact, in horizontal comparison, Tencent Cloud's growth rate is not outstanding. In the same period, Alibaba Cloud's revenue increased by 36% year on year, and Baidu's AI cloud infrastructure revenue increased by 94% year on year.
If AI cloud business fails to develop well, what Tencent loses is far more than just revenue.
Having a large-scale cloud with a huge number of users means that giants can get better prices and more stable supply in upstream computing power chip procurement, and can also provide landing scenarios for self-developed chips.
At the same time, AI cloud is the core scenario for enterprise customers to access MaaS services. Every real task they run on the AI cloud will not only consume tokens, but also precipitate scenario data and industry know-how to feed back model iteration, thus forming a flywheel of "more customers → more scenario data → stronger model → more customers".
On this highly valuable track, Tencent still needs to speed up its progress.
The other track is chips.
At present, Tencent has formed three self-developed chip product lines: Zixiao (AI inference), Canghai (video transcoding), and Xuanling (smart network card). But up to now, there has been no news that these products have been sold to the public on a large scale.
The latest financial report did not mention the progress of the three chips, nor did it propose to develop general-purpose chips.
In contrast, the chip businesses of other large manufacturers not only supply external customers, but are even sprinting for IPO.
Baidu's Kunlunxin submitted its listing application in January this year, with a valuation of up to 500 billion US dollars. Alibaba's T-head launched its Hong Kong listing process in the second half of the year, which is expected to be listed in 2027, and Morgan Stanley estimates its market value could reach 860 billion US dollars. ByteDance is also very active, it is said to have formed multiple product lines with a team size of more than 1,000 people.
Without the support of self-developed chips, Tencent must lock in computing power as much as possible under the condition of unstable external supply.
It needs to frequently use its "money power".
Reflected in the financial report, quarterly capital expenditure has been rising: 31.9 billion yuan in the first quarter, 52.7 billion yuan in the second quarter, and reached an astonishing 81.4 billion yuan in the third quarter. About 70% to 80% of this money flows to AI computing power hardware, and the rest is spent on the construction of supporting data centers.
On last night's conference call, Tencent President Martin Lau said that the current capital expenditure is actually infrastructure investment to launch AI-native businesses. Once completed, this set of facilities can be continuously used to train new generation models and continuously produce tokens.
"That's why after the investment in 2026 and 2027, the trend will get better in 2028 — because we can continuously use the infrastructure we built in 2026 and 2027," he said.
At the same time, Tencent hopes to continue to increase the proportion of domestic computing power.
Martin Lau mentioned a worrying situation: the United States has tightened overseas computing power restrictions. "Looking at the impact of the latest tightened restrictions on our AI business now, I would say this is very regrettable."
He believes that it does set everyone back, and everyone can use less advanced computing power. For the training of new models, "in the short term, it may take us longer to launch the next generation version. But as time goes by, the domestic supply chain will gradually catch up."
"In the long run, we believe that all Chinese players will have to rely on domestic supply chains and domestic chips, and we see that this industry is actually progressing very fast. With the introduction of the latest restrictions, we expect this process to be further accelerated, because all R&D forces will have to focus on accelerating the domestic supply chain," Martin Lau said.
As for the cloud business, James Mitchell, Tencent's Chief Strategy Officer, said that he sees no impact in practice.
He said: "Because our cloud business now mainly runs inference, especially for efficient models such as Hy4 flash. We have now reached a stage where these inferences can be run very efficiently with a surprisingly small number of domestic chips."
2
Another very practical anxiety for Pony Ma is: Tencent has too many middle managers.
Tencent is known for treating its employees well, rarely laying off people through harsh means, and was once called a "retirement paradise" for office workers. The ensuing problem is that the number of middle management personnel continues to expand, taking up a large amount of salary expenses, and making the whole company become bulky.
On traditional battlefields, many of Tencent's business segments have been retreating steadily under the attack of ByteDance. Especially in content business, Tencent failed to win the short video war, and also failed to seize the opportunity of micro dramas, leading to a disadvantage in user duration.
On the conference call, Mitchell admitted that the total user duration of Channels in the previous quarter increased by more than 20% year on year, and there is still room for further growth. "It should be pointed out that our DAU usage duration is still less than half of that of the industry leader."
It's just that Tencent's game business is still growing steadily, and the two major social positions WeChat and QQ are still stable, which largely covers up the poor performance of other fronts.
On the AI battlefield, Tencent's shortcomings of huge organization and slow response have been continuously amplified.
Tencent has invested a lot in its foundational model but lagged behind for a long time. It was not until Yao Shunyü took office to integrate the company's entire AI business that it gradually found its direction and took the initiative, which reflects the drawbacks of bloated middle management.
As Pony Ma said: Tencent's organizational reform needs to be accelerated, and there does not necessarily need so many management positions in the future, do not let those cadres who only pass messages, do not want to move and occupy positions become obstacles, and all mountain-topism and territory consciousness must be completely broken.
It is not difficult to see that Pony Ma may also feel that the existing team is not handy enough to fight, and the organization needs to undergo a thorough transformation from inside to outside.
His vision is that half of the people are responsible for defending the existing business, and the other half go to fight new battles, using AI to do more things together, requiring "immediate feedback, immediate decision-making, immediate action". Tencent needs to break the past management processes, and the entire management mode and working methods need to be redesigned through AI.
On the conference call, when asked about organizational reform, Martin Lau said that it is at a very early stage.
"The insight we have seen is that AI can indeed be applied to various businesses to improve operational efficiency. But to achieve this, a fairly large-scale reorganization of the organization is actually required, and a large number of work processes need to be redesigned. If we can do that, existing businesses can be operated with fewer people," he said.
Martin Lau believes that AI enables small teams to explore new ideas, build new businesses and create new growth engines with the help of AI. If we can run existing businesses with fewer people and redeploy these people to create new growth drivers for existing businesses and build new businesses, that will be the ideal state.
In fact, Tencent's launch of large-scale organizational reform now is relatively slow among major large manufacturers.
Since the beginning of this year, ByteDance, Alibaba and other companies have adjusted many major businesses around the AI backbone, including former star products such as Feishu and Wukong. Relevant management personnel have also been replaced, and a group of younger outstanding talents have been placed in core positions.
In contrast, although WorkBuddy is connected with multiple Tencent products and has integrated QClaw, it has not yet promoted larger-scale organizational changes.
Pony Ma and Martin Lau's recent speeches have released a stronger signal of reform: only by making greater efforts to adjust can we build an organization that is more adapted to the AI era and truly competitive.
At present, Tencent has made many achievements in the AI field, but has not yet won a decisive victory. WorkBuddy and Hy4 are only leading at the phased stage, and both are facing many challenges.
Taking WorkBuddy as an example, Tencent revealed on the conference call that its annualized revenue in the past seven days has exceeded 500 million US dollars, more than twice that of Tencent Meeting which was launched seven years ago. In contrast, although Doubao is not good at office Agent, it has recently started to bind with Douyin E-commerce and local life. If it sells traffic on a large scale, it will be able to generate large-scale revenue in not too long a time.
How to further convert unstable advantages into winning momentum? In the long run, one of the most worthwhile actions for Tencent to take is to carry out major reforms on talents and organization. The so-called "50% layoff" is of course a rumor, but Tencent does need to show such courage and determination.
3
Pony Ma's anxiety is an era-wide dilemma — when the timing is not yet right and the conditions are not mature, there is destined to be no solution.
Including Tencent, most domestic Internet giants still have most of their bodies staying in the last era. Their elements such as technology, products, operations and organizations are still operating according to the past model, and continue to obtain considerable returns.
At this time, the AI era suddenly came, deeply shocking every company, and making CEOs and founders shrouded in FOMO sentiment. They are forced to think about the strategies and tactics of the next era, and it is easy to find no way out and fall into distress.
It can be said that the problems plaguing Tencent are also plaguing all other giants.
Taking Doubao as an example, it is a native AI application that should have built a business model completely different from traditional apps. But in fact, it has not figured out how to do it specifically.
Judging from recent actions, Doubao is accelerating its integration with Douyin E-commerce, not only actively pushing coupons and mall cards, but also opening up a fixed entry for it. It is still making money according to the logic of mobile Internet — selling traffic to others, especially e-commerce platforms with abundant budgets.
Doubao has gone back to the old path of mobile Internet: AI cannot develop independently of mobile Internet, and even needs to rely on the latter to realize monetization — this is obviously contrary to the vision that AI will devour everything and rewrite everything.
Another problem for giants is that AI has never found a truly viable landing scenario.
At present, the business model built by AI is mainly in the B-end, or mainly AI Coding, selling tokens to programmers. In the broader C-end, AI tries to penetrate every landing scenario, but none of them can immediately provide an epoch-making solution like super apps.
According to Martin Lau's disclosure, WeChat's built-in Xiaowei has achieved 30 million DAU, and the number of daily interactions continues to increase. At a certain point, it will become as powerful as Meta