Tencent: Starting from Modest Buybacks, Is Super-AI a Calculated Public Plan?

$TENCENT(00700.HK) Q4 performance is significant in the context of AI, but the actual results are not that good. That evening, the ADR did not rise or fall, which seems to reflect the entangled nature of capital. However, the next day in the Hong Kong stock market, where major funds were distributed, Tencent fell nearly 4%, directly indicating the dissatisfaction of capital and leading to a wave of adjustments in the Hong Kong stock market.

Source: Longbridge, Dolphin Research

So what exactly are mainstream funds dissatisfied with?

Dolphin Research believes that it is not related to the business itself, but mainly the issue of capital allocation going forward: The initial plan of 80 billion for buybacks and an average of 90 billion for Capex has inadvertently "offended" two groups of people:

1) Value investors focused on shareholder returns; 2) Investors in the domestic computing power industry chain.

Why is it called the "initial" plan? Dolphin Research believes this is not Tencent's final capital allocation for 2025, whether for buybacks or capital expenditures. Not only did the management mention in the conference call that there is a 20 billion buffer for Capex, but Dolphin Research also believes that the actual reserved space is even larger.

However, industry chain investors need not be too agitated, as increasing capital expenditures by Tencent does not mean that the entire industry chain will benefit. Tencent is different from Alibaba; most of Alibaba's computing power needs are externally supplied, mainly based on customer demand for chip purchases. However, the computing power purchased by Tencent is mainly used for its own ToC business, and commercial monetization will take some time. Therefore, even if it plans to invest an equivalent scale of 100 billion in Capex, Tencent needs to be more meticulous in its calculations and ensure that the money is spent wisely.

On the contrary, Dolphin Research believes that investors focused on organic growth should eliminate some concerns and actively seek low positions that match risk and return, increasing their stakes in this top seed player most likely to give birth to ToC AI Agents and most hopeful to occupy the AI super entrance.

1. Crazy for AI, who did it offend?

Tencent's budget for capital expenditures in 2025 is in the low teens of revenue. Based on the current revenue growth expectation of 9%, this corresponds to a minimum of 72 billion and a maximum of 108 billion in capital expenditures (Capex). Narrowing the range further, it is around a median of 90 billion.

Is this number considered large or small? The market is in heated debate, with both sides holding their ground, but ultimately it depends on where one sits.

From Dolphin Research's perspective, at least from Tencent's viewpoint, this budget growth rate is not small. Although it may not be as passionately proclaimed as Alibaba's 380 billion over three years, it does not hinder the fact that this is Tencent's "most willing to spend money" moment in recent years 1. Actual Doubling of Capital Expenditure

At first glance, the median Capex of 90 billion this year has only increased by 17% compared to last year's 76.8 billion. However, it should be noted that Q4 Capex surged to 36.7 billion, driven by a sudden spike in demand for emergency stockpiling, and it cannot be ruled out that this was also due to advanced bookings under the upgraded restrictions on high-end chip purchases (the chips procured are mainly used for training).

Regardless of which factor caused it, the situation in Q4 is an exception. Dolphin Research estimates that if we follow the previous investment intensity, the annual scale should be around 50-55 billion, which is the high single-digit revenue proportion originally guided, meaning that in 2024, it actually "spent" more than 20 billion compared to the budget.

Therefore, according to Tencent's own investment pace, this year corresponds to an actual capital expenditure of 900 + 200 = 1,100 billion, compared to last year's 500-550 billion, which is a doubling pace. This level of investment is rare for Tencent, which has been tightening its purse strings over the past few years.

Unfortunately, the linear extrapolation of market expectations has already run ahead:

For Q4/2024's overall investment, combined with the news of Tencent buying cards in the fourth quarter, the market has raised expectations by 10 billion, and compared to the updated expectations, it beat by 10 billion (=200-100).

For 2025's capital expenditure, Tencent's guidance falls in the range of 80-100 billion. If we add back the extra 10 billion spent in Q4, this year's budget is only 90-110 billion. This is still much lower than the 120 billion that investors in the computing power supply chain expect, based on comparisons with Alibaba and ByteDance, not to mention the more optimistic expectations directly blind-bidding at 150 billion. Compared to Alibaba's "going with the flow," Tencent's "caution" is quite disappointing.

2. But Buybacks Discounted, Total Amount Decreased Instead of Increased

Having spent less money, are Tencent's shareholders happy?

However, they are not. The money saved on computing power has not been correspondingly added to shareholder returns (buybacks + dividends), instead, the overall shareholder return has decreased by 20 billion Hong Kong dollars compared to last year (121 billion vs 144 billion).

In the commentary, Dolphin Research also emphasized—"Shrinking shareholder returns, after the market value rises, means that the yield will be greatly discounted. Tencent's 100 billion buyback last year attracted many value investors and insurance funds seeking stable returns. If subsequent buybacks are not increased, a mere 2.5% shareholder return significantly diminishes investment attractiveness."

From the perspective of value investors, there is naturally dissatisfaction—you can increase Capex, but you cannot cut my earnings because of it. For example, when Meta and Google expanded their Capex budgets in the past two years, they also provided value-investing shareholders with a "big gift" of reassurance—large buybacks + first-time dividends. Although the shareholder returns of both companies are not high compared to their trillion-dollar market values, at least the absolute value of the return scale has increased

Interestingly, when combining <(1)-(2)>, if we look at the total shareholder returns and capital expenditures, Tencent's overall capital planning appears to be shrinking:

Last year's "total operating cash outflow" of HKD 144 billion + RMB 76.8 billion, totaling RMB 209.3 billion, seems to indicate that Tencent's capital allocation has become somewhat "cautious" compared to this year's RMB 201.3 billion (HKD 121 billion + RMB 90 billion).

This contradicts conventional judgment.

After all, if there is a 9% revenue growth and a 12%-15% net profit growth next year, then operating cash flow should also expand proportionally. Why be so stringent with expenditures outside of operations?

During the conference call, management mentioned that there would be a buffer of RMB 20 billion in capital expenditures, which comes from the reduction of RMB 20 billion in shareholder returns. But the question is, from the perspective of total cash outflows for shareholder returns + capital expenditures, which is RMB 221.3 billion, the growth compared to last year is less than 6%, still lower than this year's potential profit/operating cash flow growth rate.

So, where is Tencent spending the excess money?

2. "Savings Group" Tencent vs. "Living Paycheck to Paycheck" Meta

At the end of 2023, when Tencent announced a HKD 100 billion repurchase plan, Dolphin Research discussed Tencent's capital allocation issues in "Tencent Raises Its Gun Again: One, Two, Three, Who is the Wooden Man?." We mainly did a simple calculation and concluded:

For Tencent to comfortably meet normal operations and the HKD 100 billion repurchase, it needs to efficiently manage its investment portfolio, including ROI-driven new external investments and timely disposal of assets in the existing asset pool to meet the cash flow needed for external investments.

This is different from the reduction of holdings in JD, Sea, and Meituan for physical dividends in 2022-2023; the HKD 100 billion repurchase requires real cash, preferably in USD, to avoid the inconveniences of foreign exchange controls. Later, we also saw Tencent's reduction actions over the past year; while non-listed companies are harder to track, for listed companies, known examples like Nubank, Nio, BOSS Zhipin, and Reddit are all US-listed targets, and after selling, they can immediately obtain USD for repurchase.

So in 2025, with Tencent needing to increase capital expenditures, how will management allocate the five key funding needs: basic operations, external investments, capital expenditures, shareholder returns, and debt repayment and interest? What strategic thinking is implied behind the current allocation? What are the differences in funding allocation strategies between Tencent and the social giant Meta, which has also initiated massive investments due to AI?

(1) Meta: A Value Growth-Oriented Allocation Strategy — Spend More When You Earn More, Invest in Yourself, and Give Back to Shareholders

Meta's overall funding allocation strategy is synchronized with its income and expenditure, meaning that when business performance is good, capital expenditures will also expand proportionally, ultimately keeping net cash fluctuations within a small range. The only exception was in 2023, where net cash increased unusually, mainly due to one-off cost-saving effects (the U.S. economy rebounding after a downturn increased advertising from Chinese cross-border e-commerce/game companies while reducing VR investments at the request of shareholders).

In terms of specific net cash outflows, Meta primarily spent on soft and hard asset investments and shareholder returns. Especially after significantly increasing shareholder returns in 2021, these two major expenditures were almost evenly split. This is the most reasonable allocation strategy — the money earned from operations should either be invested in development or distributed to shareholders.

(2) Tencent: A Calculated ROI Allocation Strategy — Earn More but Spend Less, Balance Internal and External Investments

Tencent's funding allocation is quite different from Meta's. In terms of net cash changes, Tencent prefers to maintain net cash inflows greater than net expenditures. Especially after the internet industry collectively "tightened its belt" in 2022, Tencent's operational efficiency has shown a noticeable improvement, and the proportion of net cash increases has been continuously rising.

In terms of specific capital expenditures, 2021 was a watershed year.

Before 2021, Tencent's largest capital outflow was external investments, followed by investments in itself. Shareholder returns only began to show improvement in 2022, and it wasn't until the market valuation was severely suppressed in 2024 that shareholder returns barely approached half of cash outflows.

Regarding soft and hard expenditures, excluding the anomalous periods of 2H24, 1H23, and 2H20, historical figures do not seem to completely follow operational fluctuations. Compared to the significant ups and downs of other expenditures, it appears that Tencent has firmly anchored itself within an absolute scale range, for instance, the fluctuation range has been between 20-30 billion every six months since 2020.

This allocation strategy, whether before 2021 or during 2022-2024, actually reflects Tencent management's "prudent" attitude towards capital management: a. Before 2021, Tencent had an investor mindset. Although during this period Tencent's internal business was developing rapidly, and thus did not care whether cash was net outflow, the allocation attitude was relatively proactive. However, the management preferred to weigh the preferences for internal and external investments from the perspective of "return on investment" (ROI), comparing the results of internal team investments with external companies, which is another interpretation of a horse racing mechanism. Therefore, sometimes it would appear that compared to directly investing in business expansion, Tencent preferred to leverage the data advantages of ecological traffic for targeted external investments.

b. In the two years after 2022, Tencent adopted an accountant mindset, with the overarching idea being "cost-cutting." This led to a more cautious and meticulous accountant mentality, trimming different branches of business to make them more focused and reduce resource waste.

The question is, in 2025 when AI Agents are destined to flourish domestically, will Tencent still maintain its original "cautious" attitude towards capital usage?

Dolphin Research believes that Tencent will still adopt a gradual approach, but has left itself with a backup of up to 35 billion. Of this 35 billion, 20 billion is recognized by the company as a buffer for capital expenditures, but the remaining 15 billion is unlikely to be fully used for buybacks; it will still prioritize investments, whether internally or externally.

The reasons are twofold:

(1) Limited dollar deposit reserves, high friction costs for direct currency exchange for buybacks

(2) AI represents a "WeChat-level" major opportunity, with potentially higher investment ROI

Next, Dolphin Research will elaborate on these points.

Three, Tencent's "Dollar Account"

In the past two years, when the market was under extreme pressure, there was a wave of "dividend buyback" among Chinese concept internet companies, using real money to support themselves. This is the best measure for a mature enterprise to express confidence in its development.

However, when it began to be implemented, an awkward issue arose. Among Chinese concept companies, most of their main businesses are in China, and under normal operations, they spend and earn in RMB, which is not a problem.

However, due to being listed overseas, the real money needed for buybacks is in dollars or Hong Kong dollars. If there is not enough foreign currency:

a. Currency exchange must comply with the foreign exchange bureau's regulations on capital inflow and outflow;

b. Additionally, due to the VIE structure, the funds earned by the domestic operating entity (subsidiary) must be remitted to the overseas holding listed entity (parent company), which also requires paying a withholding dividend tax of 5%-10% (considered as a dividend action from the subsidiary to the parent company).

c. Moreover, it goes without saying that when the market value fluctuates significantly and temporary buybacks are needed, one must also endure some costs of exchange gains and losses.

Unless there are overseas operations (operating entities) that can directly earn dollars, the above troubles do not exist. Therefore, we see that last year, companies like Alibaba announced significant increases in buybacks, and they all temporarily issued dollar bonds 1. Where does Tencent's US dollars come from?

Compared to other Chinese concept assets, Tencent has a more diverse way of obtaining US dollars:

(1) No loss: Direct overseas revenue

Tencent also has an international business, with the main source of income being overseas game revenue, projected to generate 58 billion RMB in 2024, equivalent to over 8 billion USD. However, before being "forced" to implement an accelerated overseas strategy in 2021, Tencent's overseas game revenue was at most 4 billion USD a year. Excluding 30-50% of the developer's share, the actual net income is less than 3 billion USD.

(2) Low interest cost: Borrowing US dollars

Without US dollar revenue, it is necessary to "borrow" from external sources. Tencent mainly uses two channels to borrow US dollars:

1) Issuing US dollar bonds

This is also the main operational method for Alibaba and JD.com to temporarily exchange US dollars for repurchase, but there are differences in execution.

It was precisely due to the previous "investor" strategy, which required foreign exchange for global investments, that Tencent foresaw the need for foreign exchange funding and made advance plans. In 2014, Tencent launched the "Global Medium-Term Note Program." This program authorized Tencent to issue multi-currency bonds to a consortium of international investment banks such as Goldman Sachs and HSBC at relatively favorable interest rates (compared to the market rates at that time) on an irregular basis to obtain the required foreign exchange.

The program initially had a quota of 5 billion USD, but it was quickly raised to the current quota of 30 billion USD year by year. Over the past decade, Tencent has issued a total of 25.7 billion USD in notes, along with an additional 3 billion HKD.

Since these notes are mostly bonds with a maturity of over 5 years, up to 40 years, once issued, they lock in interest costs for many years, which tests Tencent's long-term business planning vision. Tencent is quite adept at "timing," choosing to issue bonds during periods of relatively low global interest rates, such as 2014-2015, 2018-2019, and 2020-2021, and the locked-in periods have become increasingly longer. After the global interest rate hike cycle began in 2022, Tencent paused its issuance activities.

As of the end of 2024, the principal balance of Tencent's outstanding US dollar notes that have not matured is 19.5 billion USD.

2) Direct Borrowing from International Investment Banks

In addition to long-term notes, Tencent also borrows foreign exchange directly from international banks. Among these, long-term US dollars account for more than 90% of the total foreign currency borrowing, while short-term borrowing involves multiple currencies, primarily for temporary emergency needs. Generally, Tencent signs floating interest rates and uses interest rate swap contracts to hedge against some of the risks of interest rate fluctuations.

(3) The Last Resort: Net Asset Sales

Generally, borrowing US dollars can meet Tencent's foreign exchange needs, as its main market is still in mainland China, where normal operations and routine investments do not require much US dollars. However, in exceptional circumstances where borrowing costs are too high, in addition to bearing more tax losses from currency exchange, Tencent can also resort to selling assets (especially shares in US-listed companies) to obtain US dollars.

As is well known, Tencent's investments are consistently successful, with enviable investment returns. Buying and selling generally follow a strict valuation algorithm, and they will not forcibly cut losses just to cash out US dollars unless necessary.

However, this is still a channel. Especially since last year, under the pressure of a hundred billion buyback, the requirement for the investment portfolio's "self-sufficiency" has increased, meaning they will not use the main operating cash to increase additional investment targets. From the execution situation in 2024, Tencent's investment department even "exceeded its goals"—the scale of asset disposals was greater than the increase in investments.

From the proportion of shares held by Tencent disclosed by some listed companies, it can be seen that Tencent mainly reduced its holdings in companies like Reddit, NuBank, and Nio last year. However, from the types of assets disposed of, as shown in the figure below, most of the sales in 2024 were still non-listed companies—financial assets measured at fair value with changes included in profit and loss.

2. Is the Reserve of Foreign Exchange Enough for Buybacks?

Whether borrowed or obtained through asset sales, the US dollars will eventually need to be spent.

From the final US dollar deposits held by Tencent, the deposit balance has been increasing year by year, but by the end of 2024, the US dollar deposit balance was nearly 17 billion US dollars (equivalent to 122.2 billion RMB, all short-term), which is significantly less than the US dollar debt accumulated in recent years (borrowings + notes) of 35.7 billion US dollars.

Although out of the $35.7 billion, only $4 billion is due within a year. This makes the $17 billion in short-term deposits more than sufficient compared to the $4 billion in debt.

However, after deducting the $4 billion in short-term debt, the remaining $13 billion in deposits, combined with the expected overseas game revenue retention of about $3.5 billion this year (=$9 billion gross revenue * (assuming 40% developer share + 20% overseas channel fee)), totals approximately $16.5 billion in available funds.

Additionally, if the investment portfolio remains self-sufficient without assuming net selling, then this $16.5 billion in available funds ( readily available, at an exchange rate of 7.2, about 120 billion RMB), is theoretically the maximum amount that can be directly used for repurchase (without needing to exchange currency and pay additional dividend taxes).**

However, Tencent will not operate this way in practice; retaining a certain amount of dollar reserves is necessary, whether for repaying debts from as early as 2026 or for other temporary operational needs. Therefore, if we follow the situation from 2022-2023 and retain a portion of dollars, say $8 billion (60 billion RMB), then there would be at least a $2.5 billion gap for repurchase (promising $8 billion for repurchase, which is $11 billion, using $8.5 billion in deposits, leaving $2.5 billion).

Due to the relatively high global financing rates, based on Tencent's past operations that pursue ROI and cost-effectiveness, Dolphin Research believes that the motivation for large-scale external borrowing is currently low. Because as long as it is close to 5%, and there is no short-term appreciation potential for the RMB, the friction costs including fees to investment banks make direct currency exchange more cost-effective (requiring a 5% withholding dividend tax).

Therefore, this gap can only be filled from two sources: one is direct currency exchange, and the other is reducing holdings in investment portfolios that can receive dollars, such as high-shareholdings in Pinduoduo, Futu, etc., and companies with relatively high valuations or overlapping holdings, such as Spotify and Sea.

Of course, in this process, Tencent will prioritize filtering out targets with limited future development potential. Given that the current tariffs imposed by Trump have disrupted the market, most assets in Tencent's investment portfolio are in a relatively undervalued state, and Dolphin Research will not list a death list 3.0 as in previous articles.

We believe that Tencent, which is skilled in timing and stock selection, will likely wait for a wave of recovery before carefully selecting targets if it chooses to divest assets. This is because it involves the issue of market value fluctuations of the invested companies, requiring in-depth communication with the management of the other party to ensure timely responses If during this process, the Federal Reserve further opens a rate-cutting cycle, then borrowing dollars directly later would be even more delightful for everyone.

In summary, the current net cash of USD 13 billion (after deducting short-term debts) and the expected stable overseas game revenue of USD 3.5 billion (after deducting developer and channel shares) give Tencent the confidence not to rush into borrowing dollars during a high-interest rate cycle. Referring to the dollar notes issued by Alibaba at the end of last year, the interest cost (4.875%~5.625%) is higher than Tencent's historical financing costs.

However, at the same time, on one hand, the dollar reserves are not completely sufficient, and on the other hand, historically, during high growth periods, Tencent's return to shareholders has not been high compared to Meta in terms of subjective willingness. Therefore, last year's 100 billion buyback was more like a "bottoming action" under the circumstances of no business development and continuous selling by major shareholders, and should not be seen as a subjective attitude shift towards regularly increasing buybacks in the future.

In addition, Tencent's buyback willingness this year (whether subjective or forced) can be seen to some extent from the small perspective of the global stock market crash triggered by tariffs:

—— After the tariff announcement on April 7, Tencent's market value fell by 12% in a single day, but Tencent did not significantly increase its buyback efforts on that day, unlike during the two event windows of the new online gaming regulations at the end of 2023 and the U.S. Department of Defense sanctions list at the beginning of 2025.

IV. AI is an industrial opportunity on the level of "recreating WeChat"

Returning to the initial question, Dolphin Research can provide an answer. The most likely allocation of the RMB 15 billion "backup" funds is to continue increasing business investments (including capital expenditures), rather than buybacks.

In fact, since the end of last year, Tencent has already changed its "wait-and-follow" attitude regarding this round of AI transformation. Perhaps it has seen the soaring traffic of Yuanbao, or perhaps it has noticed the "computing power deflation" trend brought by the Deepseek V1 version.

After DeepSeek made a splash during the Spring Festival, Tencent finally stopped hiding and launched a comprehensive high-profile promotion, giving all the top-tier traffic resources of the entire group to "Yuanbao." If the initial "Yuanmeng Star" received resources at the billion level, then "Yuanbao" must be at the hundred billion level. The last product that warranted the entire group’s involvement and received a public endorsement from Pony was still WeChat.

Dolphin Research believes that for Tencent, creating the next super agent is imperative. This may not only be about optimism for AI development but more critically about maintaining the position of the primary traffic entry point.

AI will certainly give birth to a super entry point, similar to WeChat at the billion-user level. But whether this super entry point will emerge from Tencent, no one can confidently judge. After all, in the past decade, no one expected that under the absolute advantage of WeChat's strict defense, Douyin could break through the "two WeChat and one Toutiao" social defense chain with a new content form (video information flow) and an addictive algorithm (privacy-invasive precise recommendations), ranking as another 1 billion-level "social + content" platform (of course, "one end," Toutiao, is Douyin's brother) The platform, along with TikTok, has also torn apart Meta's social network, even reaching parity with WeChat.

In other words, new killers can emerge at any time. Who has the highest probability of winning? Dolphin Research will share a few of its subjective judgments, not afraid of being proven wrong, just for discussion, after all, there is no 100% certainty in the AI era.

Judgment 1: No new forces will emerge in the AI super entrance

Dolphin Research defines the AI super entrance not as an all-powerful AI entrance. The more one desires it to be "all-powerful," the more likely it becomes a "chicken rib" in the eyes of users.

For vertical Agents, we believe that the opportunities for small and medium-sized enterprises and independent studios have not diminished. However, the more feasible product form for a general-level super Agent is—based on solving general search functions, it can also serve as a unified calling front for millions of vertical Agents behind it. Simply put, it is an "app store" for the AI Agent era.

Therefore, returning to AI Agents, the final circle of the super entrance can only be these giants. There are two reasons:

(1) Scene demand outweighs model scoring. The emergence of the epic milestone DeepSeek has, to some extent, leveled the AI starting line for domestic giants, and at least the significance of large model scoring has rapidly declined.

This is quite uncomfortable for platforms that previously invested heavily in developing their large models. Resource waste is a minor issue; the key is how to balance the issue between DeepSeek and its large models.

Adding a DeepSeek to the front end is simple, but the team deployment behind it will be quite complicated. After all, the giants cannot completely abandon the development of their large models, but in the case of limited or even exclusive resources for computing power and promotional positions, how to manage internal and external allocation is a significant problem.

What can be confirmed is that when DS, promoted by the state, helps with the first step of user education (as of February 2025, the monthly active users of AI chatbots totaled 258 million, with a penetration rate exceeding 20%, and this popularization speed far exceeds that of short videos), then the core of the upcoming competition, shifts from technology to seizing user usage scenarios, returning to what the giants are best at. This, in turn, raises the existing traffic entrance advantage of the giants—having basic traffic makes it easier to win at the starting line, such as Doubao.

With the support of Douyin's billion-level traffic resources, the model scoring is not optimal for Doubao, which has already created a significant gap with its peers in just one year.

(2) The capital threshold may not have decreased. Large companies that already have access also have such defensive needs and will be more proactive in accelerating user penetration, or more accurately, the process of grabbing users. As user stickiness increases, the inference call volume of a super Agent will also increase exponentially.

The computing cost and user promotion cost required behind this demand can be seen from the scale of capital expenditures by Alibaba, Tencent, and ByteDance in this area; it is still a significant amount. In other words, the "computing power deflation" promoted by DeepSeek may instead accelerate the pace at which large companies personally enter the field, and has not resolved the capital threshold in the competition circle, making it very difficult for small and medium-sized companies and even early-stage unicorns that need capital infusion.

Judgment Two: The Final Showdown is Tencent vs. ByteDance

As mentioned earlier, a super entry point requires both traffic and the ability to call vertical Agents, while the platform itself must have sufficient capital to cultivate user habits in the early stages.

Whether it is traffic or calling ability, at present, if the platform can directly adopt a "take-it-or-leave-it" approach, it undoubtedly increases the odds of winning in this competition.

Among the giants in China's AI final circle, there are only a handful of players with the potential to become super entry points in the ToC space, possessing the above capabilities: Tencent, ByteDance, Alibaba, and Baidu:

(1) Those with a billion-level and high-sticky traffic, namely Tencent and ByteDance.

(2) Those with calling ability (platforms that can persuade vertical Agents to hand over their API interfaces), mainly WeChat with mini-programs (Tencent), Alipay (Alibaba), Douyin (ByteDance), and similar Baidu intermediary pages.

However, it should be noted that since Alipay only has functional calls and does not have user chat + search information usage scenarios, it is difficult for user habits to be refined. Therefore, Alibaba's AI ToC has chosen to bet on the "Quark Super Box."

However, in terms of specific implementation, there are also differences among the giants:

(1) Internal and external large model resource allocation: Tencent, which places more emphasis on racehorse ROI and has already mastered balancing resource allocation under the competition of similar projects internally and externally, undoubtedly places more importance on DeepSeek at present.

Baidu follows, as the first company to launch a large model and go all-in on AI, it is completely impossible to abandon self-research; this is the only pillar after the decline of the traditional search story. However, its original entry point is too weak, necessitating the introduction of DeepSeek as a necessary gimmick.

But Alibaba and ByteDance choose to continue to believe in themselves.

Although Alibaba chose the weaker traffic Quark on the C-end, it did not introduce DeepSeek. ByteDance's Doubao also did not embed DeepSeek Dolphin Research and believes that the simultaneous development of internal and external large models will always test the management's resource allocation and team balance capabilities in the short term. During the adjustment process, there may be personnel turbulence, which could disrupt the product advancement rhythm. Alibaba and ByteDance do not have this issue, but this is based on their models being quite good and the parent company continuously directing traffic. Tencent will encounter problems, but considering its extensive experience in competition, it may ultimately not face significant issues.

(2) Front-end product form strategy: These four companies can be divided into two camps.

Tencent and ByteDance have similar strategies, relying on their own social traffic foundation, advancing AI+ empowerment and independent apps simultaneously.

Alibaba and Baidu have similar strategies, relying more on search engine traffic for empowerment. Although Baidu also has independent apps, its traditional search traffic is gradually being eroded, and the Wenxin Yiyan app has already emerged in the first tier.

AI+ empowerment is considered the ultimate product form under scenario integration. Currently, the emergence of independent apps, Dolphin Research believes, is likely just a transitional choice.

To achieve a truly practical agent, a lot of integration of backend content and third-party applications is needed, as well as defining the task nodes that super agents and vertical agents can complete. This requires continuous multi-party testing and discussion processes. Therefore, immediately integrating and launching it may harm users' initial impressions of the product and may also affect the original user experience.

Thus, the final super entry point is more likely to be attached to existing products. This requires each company to showcase its flagship products for comparison. From the perspective of traffic distribution based on functional calls, it is clear that WeChat > Douyin > Baidu > Quark. However, the stickiness of traffic and product strength determines that in this final match, ByteDance also can catch up and win.

Once a super agent is formed, it not only signifies the disappearance of traditional search but also implies that application stores have no meaningful existence. This agent can exist on any physical device, replacing millions of applications to become the most frequent entry point for one-on-one interaction with users.

It shortens the jump chain between users and the realized functions (terminal merchants) and holds the most concentrated traffic distribution rights—tasks it cannot complete will also be determined by which vertical agent to call. Once users become accustomed to an AI assistant that knows all their preferences, the migration cost will be enormous. As the supply chain becomes shorter, the value becomes more concentrated at the end closest to the user.

In summary, the AI agent of 2025 is the industry opportunity with the greatest potential for Tencent to gain, but if it fails, it will also face the greatest threats. Therefore, it is willing to invest the highest level of resources. Using Tencent's usual formula, compared to direct repurchase dividends, investing in itself now will yield a higher ROI.

In 2025, do not expect Tencent to take a conservative approach, but rather anticipate Tencent to play more cards in AI. Of course, the reason for holding back also reflects the management's crisis awareness. Unless extreme situations occur during the trade war, such as the recent threats of delisting made by Treasury Secretary Basant or investment restrictions, which may put pressure on stock prices in the short term due to liquidity, the management may choose to temporarily repurchase shares to save the situation.

As a precaution, Dolphin Research will specifically sort out Chinese concept assets shortly, using unconventional thinking during extraordinary times to interpret investment opportunities in the worst-case scenario. An updated and clearer value judgment on Tencent will also be provided, so please stay tuned.

Dolphin Research "Tencent Holdings" related articles:

Earnings Season (Past Year)

March 19, 2025 Conference Call: Tencent (Minutes): Adjusting AI Investment Dynamically Based on Demand, Balancing Capital Allocation

March 19, 2025 Earnings Review: Tencent: Crazy for "AI"! Is this stock king loved or hated?

November 13, 2024 Conference Call: Tencent: Increased R&D Investment Mainly Used for AI (3Q24 Conference Call Minutes)

November 13, 2024 Earnings Review: Tencent: Accelerating Investment, Slowing Profit, Can the Stock King Still Be "Both Offense and Defense"?

August 14, 2024 Conference Call: Macro Pressure Exists, but Tencent Does Not Lack Profit Adjustment Points (2Q24 Conference Call Minutes)

August 14, 2024 Earnings Review: Tencent: A 360-Degree Detailed Breakdown of the "Impressive" Performance

May 15, 2024 Conference Call: Tencent: Significant Solutions to Game Challenges (1Q24 Conference Call Minutes)

May 15, 2024 Earnings Review: [Tencent: Video Accounts Transform into Cash Cows, Stock King Takes Off Gracefully](https://longportapp.cn/zh-CN/topics/21083884? invite-code=032064)》

March 21, 2024 Conference Call: Tencent: Games undergoing internal self-transformation, AI is first effectively applied to advertising business (4Q23 Conference Call)》

March 21, 2024, Financial Report Commentary: Tencent: Games lost their soul, the stock king "spends money" to regain respect》

November 16, 2023 Conference Call: Fully shifting to "high-quality growth" (Tencent 3Q23 Performance Conference Call Minutes)》

November 15, 2023, Financial Report Commentary: Saving money "middle-aged" Tencent: Can't run anymore, just have to save, save, save》

In-depth

July 23, 2024 Industry Overview: Summer Game Chaos, can Tencent still sit firmly in the high seat?》

January 6, 2023: Pan-entertainment "good start", whose rebound is more sustainable, Tencent or Bilibili?》

September 28, 2022: Revisiting Tencent, exploring the "bottom" of the stock king》

January 5, 2022: Is "Little Tencent" scaring Tencent's little brothers? The significance of Sea's situation is different》

Hot Topics

September 23, 2024: Reflections after the explosion of "Black Monkey": Can AAA save the valuation of domestic games?》

June 21, 2024: Holding the new "King", Tencent is at odds with the channels again》 On July 19, 2023, "Tencent: With major shareholders selling off, can the stock king still have faith?"