Tencent Conference Call: Chinese tech companies prioritize efficiency, with capital expenditures lower than their Western counterparts, as evidenced by DeepSeek

On Wednesday, the 19th, Tencent released its financial report for the fourth quarter of last year and for the entire year of 2024. The report shows that Tencent's net profit in Q4 increased by 90% year-on-year, and the annual capital expenditure surged by 221% year-on-year to a record high.

Regarding capital expenditure, Tencent executives pointed out at the earnings conference that R&D spending would not put pressure on profit margins. The capital expenditure is mainly used for purchasing GPUs to support advertising and gaming businesses, large language model training, and cloud services. These capital expenditures have actually brought good profit margins and high returns. Additionally, with breakthroughs in DeepSeek technology, the industry as a whole (including us) can utilize existing GPU resources more efficiently, improving large model training efficiency, without the need to purchase GPUs on a large scale as originally anticipated.

In terms of advertising business, Tencent noted that the company's advertising growth rate surpassed the industry average, partly because Tencent's AI-enhanced technology provided advertisers with a higher return on investment (ROI).

Regarding the gaming business, Tencent executives stated that by the end of last year, the company's deferred revenue (mainly from the gaming business) grew by double digits year-on-year, and it is expected to gradually convert into reported revenue in the first half and second half of 2025 and during certain periods in 2026. During this year's Spring Festival, the daily active users of the company's five highest-grossing games increased year-on-year, indicating the popularity of games, which is a positive leading indicator. The gaming business will directly and indirectly benefit from AI technology enhancements.

In terms of e-commerce strategy, Tencent executives stated that the mini-program store is a long-term strategic direction for Tencent. For example, the red envelope gift feature is widely used by many people to send gifts to friends, who will input their delivery addresses. This actually helps Tencent improve its delivery address mapping and makes it easier for many users who have already entered their addresses to complete transactions. Moreover, the gift recipient may continue to pass the gift on to others.

Regarding AI, Tencent executives mentioned that AI agents can be deployed in WeChat and QQ. One of Tencent's advantages in the future competition of large models lies in the high daily usage time and frequency of opening the WeChat app. Additionally, WeChat offers a rich variety of activities, covering social, entertainment, and learning, allowing AI agents to deeply integrate into the mini-program ecosystem to meet users' diverse needs.

In terms of capital allocation, Tencent revealed that AI is a key investment area. The company aims to invest in the future (such as AI) while also providing current returns to shareholders through dividends and stock buybacks. This year, it has announced a dividend payment of HKD 41 billion, and the scale of stock buybacks has been reduced from HKD 100 billion last year to HKD 80 billion, leaving HKD 20 billion as a flexible buffer to decide whether to further invest in AI based on actual conditions.

The following is the full Q&A:

Q1: Regarding the impact of AI capital expenditure on finances. As we increase our capital expenditure in the AI field, additional depreciation and R&D expenses will inevitably drag down our profit margins. In recent years, as we focused on high-quality growth, profit margins have significantly improved. So, in the future, how do we balance growth and profitability improvement?

People often say that the only two inevitable things in life are death and taxes. As the management of the company, if we say that a decline in profit margins is unavoidable, that would clearly be irresponsible And we indeed do not believe this is inevitable.

Regarding R&D spending related to AI, in fact, throughout Tencent's development history, we have increased R&D investment in different projects every year. Therefore, we do not believe that R&D spending itself will put pressure on our profit margins.

In contrast, capital expenditure (CapEx) is a more complex topic. Indeed, we raised capital expenditure to a higher stable level in the fourth quarter of last year, and over time, this additional capital expenditure will gradually translate into additional depreciation in the coming years.

However, it is worth exploring where this capital expenditure is specifically used to determine whether depreciation will indeed put pressure on profit margins. The most direct use of capital expenditure is to purchase GPUs to support our advertising technology and gaming business. From our financial report data, it can be seen that this capital expenditure has actually brought good profit margins and high returns. Liu Chiping also mentioned in previous discussions that these investments are effective.

The second use is for GPUs used in training large language models (LLM). There was a time last year when the industry generally believed that each new generation of language models required an order of magnitude more GPU resources. However, with the technological breakthrough of DeepSeek, this trend has changed. Today, the entire industry, including us, is able to significantly improve the efficiency of large language model training using existing GPU resources, without needing to increase GPU procurement at the pace previously expected.

The third direction of capital expenditure is related to our cloud business. We purchase GPU servers and then lease them to customers, generating returns. Although this may not be the highest return segment in our business portfolio, it is still a positive return business that can cover the costs of GPUs and related depreciation expenses.

Finally, the capital expenditure for consumer-side inference may face pressure in the short term. This is indeed an additional cost pressure, but we believe it is a controllable cost pressure, as this part of capital expenditure is only a portion of total capital expenditure. Moreover, we are optimistic about the future because we believe that over time, our consumer-facing inference activities will monetize through advertising revenue and value-added services, just like other activities on Tencent's platforms.

Overall, while we understand your concerns about the increase in capital expenditure and its impact on long-term profitability, we are actually very optimistic that we can continue to drive business growth while protecting profit margins.

In consumer-facing inference products, there are indeed many ways to reduce unit costs through technological means, software optimization, and better algorithms. This is also a factor to keep in mind.

Q2: Regarding the mini-program store. We launched the WeChat mini-program store and red envelope gift features in the third quarter, which have been warmly welcomed by users. Could the management share this year's strategy and key initiatives in the e-commerce field?

The mini program store is a long-term strategic direction for us. Any specific initiatives are just one of our many efforts to build this ecosystem in the long run.

I want to remind everyone of the positioning of the mini program store. It is essentially a unified platform that connects various parts of the WeChat ecosystem. Through a standardized product information data structure, product information can flow freely between different components of the WeChat ecosystem. The goal is to enable consumers to find quality products and merchants.

If you observe the different components of the WeChat ecosystem, you will find that it includes social infrastructure, content, search, mini programs, transaction platforms, and WeChat Work. These components together form a unified WeChat ecosystem.

The mini program store indeed hopes to attract users to engage deeply. If you look at the red envelope gift feature, it is actually a function within the social component of the WeChat ecosystem. This is just one of our many features aimed at fully leveraging the potential of the entire WeChat ecosystem.

From initial feedback, as you mentioned, this feature has been warmly welcomed by users. We see many people using this feature to send gifts to friends during the Spring Festival. This gifting behavior actually amplifies the word-of-mouth effect of quality products because you will only gift products that you truly believe are good.

For the recipients of gifts, this behavior further amplifies the impact of the initial purchase. They will enter their delivery address, which actually helps us build the infrastructure for a delivery address map, making it easier for many users who have already entered their addresses to complete transactions.

Generally, recipients of gifts sometimes further gift the items to their friends. Therefore, overall, I think this word-of-mouth effect is very positive, and the feedback from merchants is also very encouraging. But this is just one component of the entire WeChat ecosystem, right?

Therefore, we believe that over time, we will continue to patiently build this ecosystem and platform, viewing it as a marathon rather than a sprint. We believe that we can go far and long on this path. If you look at the GMV (Gross Merchandise Volume) of the mini program store, it continued to grow very rapidly in the fourth quarter of last year.

Q3: Regarding enterprise services. I would like to ask management if they could elaborate on the demand and adoption rate of the Ask service over the past two months, as well as your outlook for demand growth in the coming quarters. Additionally, besides the mentioned Tencent Meeting and Tencent Docs, could you share some of the SaaS product solutions we plan to launch that may help with cross-selling and upselling to cloud customers?

The demand for iService is actually very strong. We are currently supply constrained. One of the reasons for the significant increase in capital expenditures in the fourth quarter of last year was that we received a batch of urgent orders to purchase GPUs to support inference and cloud services. We can only fully meet the growth in iService demand once we install these GPUs in the data centers, which will take some time

Therefore, I would like to say that in the first quarter, we may not have fully captured the growth in demand, but over time, with the arrival and installation of GPUs, we will be able to meet more demand.

Regarding different software solutions, you mentioned Tencent Meeting and Tencent Docs, while WeChat Work is another very powerful product. In fact, it is our highest revenue SaaS product, with revenue doubling year-on-year in the last quarter.

In addition, we have also seen security access software and audio-video communication software, including real-time communication and live streaming software, which are also being sold to our cloud customers. Many of these products can also be empowered by AI technology, providing additional value to customers.

In terms of consumer-facing applications, we do have a large number of different consumer applications, and you can expect more new products to be launched in the future.

Q4: Regarding consumer-facing applications. In addition to the significant breakthroughs made by the Hunyuan large model overseas in the past few months, there are also EMA Copilot (an AI assistant) and enhanced search functions. I wonder how these products will gradually evolve over time, and whether Hunyuan will position itself as an AI gateway that integrates all search and discovery entry points, thus complementing the WeChat super app?

I believe AI is still in a very early stage, so it is difficult to talk about what the final state will look like. But what I want to say is that each product will continue to evolve and become very useful or even more powerful tools for users.

For example, Yuanbao can become a very powerful AI-native assistant. Mail Copilot can serve as your personal knowledge base and also as a shared knowledge base for team collaboration. WeChat may launch many different features in the future.

In addition, I believe our other products will also integrate AI experiences, including QQ, browsers, etc. Therefore, we will see more and more consumer-facing AI products. At the same time, each product will continue to evolve.

If you look at Yuanbao, it indeed integrates many different functions, but it will not be the only entry point. Each of our products will try to find unique use cases, leveraging AI to provide an excellent experience for users. Meanwhile, our different products can also work together to provide suitable pathways for user growth of AI products.

Therefore, I believe this will be a continuously evolving process that will help us build a series of consumer-facing AI solutions and applications.

Q5: My question is actually a continuation of the previous question. We have seen Tencent Yuanbao show very strong growth momentum since this year. Could management elaborate on the strategy for further growing the user base in a competitive market? Yuanbao is a combination of AI chat and AI search; which part does Tencent have more interest in? Additionally, related to this, what will be the future share of AI search in the overall search market (measured by search query volume)? And how does Tencent plan to monetize AI search in the future? Currently, Yuanbao is a product that combines chat and search functions, but over time, it will develop into a multifunctional AI assistant serving different types of users.

Its user base will include students (for learning), knowledge workers (for completing work tasks), and users who need to conduct in-depth research (for exploring different topics).

Therefore, Yuanbao will have many different application scenarios. Yuanbao's unique advantages are, first, its innovation capability, continuously adding new features and characteristics to meet user needs. Second, the content ecosystem, Yuanbao can access Tencent's content ecosystem, especially public accounts and video accounts, which are high-quality sources of information. Third, the multi-model strategy, Tencent's multi-model strategy allows users to access the best models and utilize model combinations to meet complex needs.

In the future, as I mentioned earlier, many of our high daily active user (DAU) products will begin to add different AI features and characteristics, some of which will be integrated with Yuanbao to create synergies. Our product array will support each other and develop together.

Therefore, we believe these ongoing efforts will keep us competitive, and we also have some unique advantages.

Regarding the broader question you mentioned about the relationship between AI prompts and traditional search, I think different people will have different views, and this will take time to verify. But from a macro perspective, if we look back at the history of web search, including early web directories, and compare our behavior using AI prompts with traditional search, I believe AI search has the potential to replace traditional search.

Because ultimately, web directories, traditional search, and AI prompts are all mechanisms for accessing the internet knowledge graph. But among these mechanisms, AI prompts bring new technologies, new efficiencies, and new transactional capabilities achieved through agent AI, which traditional search cannot achieve.

As for the monetization methods of AI prompts, time will provide the answer, but we have already seen in the Western market that the initial monetization model was through subscription models, followed by performance advertising. In China, I believe it will start with performance advertising, followed by value-added services.

Q6: Regarding the advertising business. We have seen a healthy 17% growth in the advertising business, can the growth rate in the fourth quarter be seen as a reference for the entire year of 2025? Additionally, regarding AI-driven advertising enhancement technologies, we have seen from Meta's machine learning and Advantage+ shopping ads that these AI technologies can drive accelerated growth for global peers. So, with more AI applications landing this year, is it possible for the advertising business to accelerate growth further?

Regarding the advertising business, we are very satisfied with the growth rate in the fourth quarter, which significantly exceeds the industry average. The growth mainly comes from organic growth rather than external factors, and the growth covers almost all industries we monitor

We believe this is because in almost all industries we monitor, the AI-enhanced technology we deployed provides advertisers with a higher return on investment (ROI) than ever before, while also outperforming other platforms. You compared us with some global peers, and I think that is the right approach, but it is important to note that global peers often tend to fully load ads in the early stages of their product evolution.

For example, in the short video sector, we tend to gradually increase the ad load of new products, and this strategy remains unchanged. Therefore, I believe we have differences in ad loading speed compared to some global peers. But overall, as long as there are no drastic changes in the macroeconomic environment, we are quite satisfied with the performance of our advertising business.

Q7: Regarding the gaming business. How should we view the growth prospects of domestic and international gaming businesses this year, especially considering the high base effect of the domestic gaming business in the second half of this year? Additionally, what is the impact of AI on the gaming business? You mentioned the extension of game lifespans, but also referred to changes in competition and cost structures.

Regarding the gaming business, you mentioned the high base effect in the second half of this year. I think this is the "curse" of the industry—if you do well, people will worry about the base effect a year later.

But I want to point out some observable facts about our gaming business, one being the growth of deferred revenue. At the end of last year, our deferred revenue (most of which comes from the gaming business) grew by a high double-digit percentage year-on-year. This deferred revenue will gradually convert into reported revenue in the first half of 2025, the second half of 2025, and part of 2026.

Another is user behavior during the Spring Festival. User behavior during the Spring Festival is an important indicator of game popularity. This year during the Spring Festival, the daily active users (DAU) of our top five revenue-generating games all increased year-on-year, which is a positive leading indicator.

In addition, there are some subjective views. We subjectively believe that several games are gradually developing into evergreen games, one of which we mentioned is "Delta Force." This is a very important opportunity and growth driver.

Secondly, as we mentioned in our prepared remarks, we have many new games in development, and we are very excited about these games.

Thirdly, we believe that the gaming business will benefit directly or indirectly from AI technology enhancements. One is in a direct way, where game developers use AI technology to create more content faster and serve more users more effectively.

The second is in an indirect way, which may be a trend spanning decades, not just a story for the second half of this year. As humans use AI more broadly, we believe people will have more time and the demand for high-autonomy activities (such as gaming) will also increase, as AI empowers people with more capabilities and freedom

Therefore, for users, interactive entertainment (such as games) is one of the best ways for them to express themselves in a highly autonomous manner (rather than passively).

I would like to add that when we think about competitive dynamics, we believe that AI will make evergreen games more enduring. We have already seen how AI helps us execute and amplify our evergreen strategy. Part of this is reflected in content production: high-quality content can now be produced in a shorter time, allowing for more frequent game updates to maintain user interest.

In PvE (Player vs Environment) experiences, games can become more exciting with smarter bots, resembling PvP (Player vs Player) more closely. In PvP, AI can significantly improve aspects such as matchmaking, balance, and onboarding, thereby enhancing the user experience. All of this will help already popular large-scale games become even more popular and attractive to users.

Q8: Regarding commercial payment business. It is encouraging to see commercial payment revenue shift from negative growth in the third quarter to flat in the fourth quarter. Can you share your observations on the trends in commercial payment activities for the first quarter of 2025?

Regarding commercial payments, our observation is that transaction volume is actually continuing to grow, but the average selling price (ASP) is still under significant pricing pressure. Therefore, from a value perspective, revenue remains basically flat.

Our interpretation of this trend is that consumer spending willingness is recovering, but there is still significant pricing pressure on the supply side. Therefore, we hope this is a good sign indicating that we are nearing the end of a tough market. As consumer demand continues to improve, competition among suppliers may gradually ease, ultimately translating into value growth. However, we need to observe this further in the future.

Q9: Follow-up question regarding AI and capital expenditures (CapEx). You mentioned that the capital expenditure to revenue ratio for 2025 will remain in the low double digits, similar to the ratio for 2024. This guidance implies that capital expenditure growth will significantly slow down. Can you elaborate on the logic behind this capital expenditure to revenue ratio? Is it because you foresee a slowdown in the demand for generative AI, or is it because the significant increase in 2024 is already sufficient to meet the generative AI demand for 2025?

Yes, the increase in capital expenditure at the end of 2024 should be sufficient to meet the demand for generative AI and other businesses in 2025, while maintaining sustainable development at the new normal operational level. There is a time lag between ordering GPU servers and fully deploying them to data centers. Therefore, for part of the time in the fourth quarter and the first quarter, we are in this state. However, as Liu Chiping mentioned, by the end of the first quarter, we are deploying these GPUs and benefiting from them, meeting both our internal demand for Yuanbao reasoning and the needs of external Tencent Cloud customers, generating direct revenue through leasing GPU servers.

From a more macro perspective, there was a time last year when people asked whether our capital expenditures were sufficient compared to our Chinese peers or global counterparts. Now, among listed companies, I believe our capital expenditures in the fourth quarter are the largest among Chinese tech companies. Therefore, we are leading among our Chinese peers.

Overall, the capital expenditure as a percentage of revenue for Chinese tech companies is lower than that of some Western counterparts. However, we believe this phenomenon is due to Chinese companies generally prioritizing efficiency and the effective utilization of GPU servers, which does not necessarily impact the final technological outcomes. The success of DeepSeek actually symbolizes and reinforces this point.

It is worth noting that this is a very dynamic situation. What we are providing is essentially our current expectations, but frankly, these expectations may change. If there is a sudden surge in demand, we will definitely increase our GPU orders. Therefore, I believe we will maintain a high degree of flexibility and dynamic responsiveness to market changes.

Q10: Regarding follow-up questions on the advertising business, could management share more details? You mentioned that advertising spending has increased across most industries, which verticals have benefited the most? Additionally, in terms of advertiser behavior, we have noticed significant growth in video accounts, media advertising, search advertising, etc. What changes have occurred in advertiser behavior, and how have these changes driven the growth of advertising revenue?

Regarding your question about our advertising business, we see year-on-year growth in industry categories including e-commerce, finance, fast-moving consumer goods, gaming, local services, education, and healthcare. This is a broad list of industries.

As we deploy these AI-enhanced technologies, we typically apply them first in video accounts, and then gradually penetrate into other parts of the WeChat ecosystem, Tencent's other products, and our advertising network. Therefore, over time, the benefits brought by these technologies will be increasingly felt. This is part of the advertising business.

Q11: Regarding capital allocation. Management mentioned that this year's stock buybacks will be less than last year, but on the other hand, dividends will increase. So, what are our priorities in capital allocation? At the same time, in the context of strong demand in the current AI industry, how should we view merger and acquisition investments?

Regarding capital allocation, I think the key principle is that we want to create returns for the company and shareholders through investments. This means that at different times, we will invest in different areas. Sometimes, we will invest heavily in the ecosystem and our core business.

We used to invest heavily in ecosystem partners and built a very large investment portfolio. Subsequently, we began to return cash to shareholders through dividends and stock buybacks. And now, AI has become our focus investment area, and we believe AI has great potential to create returns for the company in the future. Therefore, our current investment strategy is: to invest in the future (such as AI), while also providing current returns to shareholders through dividends and stock buybacks

I believe this is our core principle. From our financial capability perspective, we have a very strong ability to achieve this goal, as we have very robust cash flow at the operational level, along with a large and valuable investment portfolio, a significant portion of which consists of highly liquid assets.

Therefore, we have sufficient financial resources to invest in the future, whether in AI or other necessary investment activities. Our portfolio is essentially self-sufficient while also providing current returns to shareholders.

Regarding this year's capital allocation, we have announced a cash dividend of HKD 41 billion and slightly reduced the stock buyback scale from last year's HKD 100 billion to this year's HKD 80 billion. This leaves us with a flexibility buffer of HKD 20 billion, to decide whether further investment in AI is needed within this year based on actual circumstances.

Comparing to the past two years, last year we paid a dividend of HKD 32 billion, with stock buybacks totaling HKD 100 billion, amounting to a total target return of HKD 132 billion.

This year, our target return is HKD 80 billion (stock buybacks) plus HKD 41 billion (dividends), totaling HKD 121 billion. If we include the excess portion of last year's stock buybacks (HKD 15 billion), this year's figure is actually similar to last year's target return.

Therefore, we are striving to provide a very good balance between current returns and future investments. At the same time, we have very strong financial resources to achieve this goal.

Q12: Regarding the discussion on expenditures, high-margin businesses, and monetization. Given that we have been pursuing a high-quality growth strategy over the past few years, and we have also seen the story of margin expansion—operating profit growth outpacing revenue growth. But we also note that over time, the gap between the two will gradually narrow. Therefore, I would like to connect these points, and ask how should we view the story of margin expansion in 2025 and the coming years?

To some extent, one key reason we have been able to achieve an increase in operating profit margin over the past two years is that we have enjoyed an improvement in gross margin. The reason for the improvement in gross margin is that, although the overall gross margin of our core business is about 50%, many new revenue sources (which contributed to most of the revenue growth) have gross margins as high as 70% to 80%.

With this change in revenue structure, the gross margin is pulled up, leading to gross profit growth outpacing revenue growth. Of course, over time, a base effect will occur, or more accurately, an asymptotic effect. If the overall gross margin gradually approaches the incremental gross margin, then the pace of improvement in overall gross margin will inevitably slow down. I believe this is unavoidable in the long term.

Nevertheless, when we observe our high-quality revenue sources, such as video account advertising, search advertising, value-added financial services, e-commerce transaction commissions, and self-developed games, the growth rate of these businesses typically exceeds that of overall revenue and contributes gross margins far higher than the overall gross margin. Therefore, we believe we will continue to enjoy gross profit leverage in the future, thereby driving the improvement of operating profit leverage From a gradual perspective, income growth will tend to stabilize. Of course, AI investment will actually have a certain offsetting effect on this trend.

Q13: Regarding AI and AI agents. We have seen many companies release their own large models and AI agents. How does management view the key competitive factors for future large models, and what is our competitive advantage? How should we enhance future user engagement by adding more features and functionalities?

I believe there are many types of AI agents, right? An AI agent is essentially a model that utilizes model capabilities and connects to different software tools to complete complex tasks. This is a very broad concept.

You can have standalone AI agents or AI agents embedded in different applications. We believe there will be a variety of AI agents in the future, but for us, we will be able to build standalone AI agents in the following ways. For example, by utilizing high-quality models. By leveraging our large user base across different software platforms, such as our browser, Yuanbao, etc.

At the same time, we can also deploy AI agents within WeChat and QQ. These AI agents can leverage the in-app ecosystem to provide excellent services to users by completing complex tasks. Taking WeChat as an example, first, one of our advantages is that WeChat users have a very high daily usage time, and the frequency of users opening the app is also very high.

The second advantage is that activities within WeChat are very diverse, not just entertainment and transactions, but also social communication and content consumption. Many people work and study on WeChat, and a large number of transactions are completed through WeChat. The WeChat mini-program ecosystem supports a variety of different activities.

Therefore, if we observe the mini-program ecosystem, we can easily build an AI agent based on the model that connects to many different mini-programs to help users complete various activities and complex tasks. I believe these are all very unique advantages we currently have.

Of course, these experiences need to be built with great care and patience to ensure that we provide the right experience for users while placing a high emphasis on their data security, sense of security, and comfort. These are all considerations we need to pay attention to when building these products. But over time, I believe these are all huge opportunities we face