Tencent Holdings Limited (700) Earnings Call Transcript & Summary
May 13, 2026
Earnings Call Speaker Segments
Wendy Huang
executiveI'm Wendy Huang from the Tencent IR team. Let me now introduce the management team joining us on the webinar tonight. Our Chairman and CEO, Pony Ma, will start us off with a brief overview. President Martin Lau and Chief Strategy Officer James Mitchell will then walk through the business review, and Chief Financial Officer John Lo will wrap up with the financial discussion before we take your questions. I'll hand things over to Pony now.
Wendy Huang
executiveGood day, and good evening, everyone. Thanks for joining us today. Welcome to the Tencent Holdings Limited 2026 First Quarter Results Announcement Webinar. This is Wendy Huang from the Tencent IR team. Right now, all participants are in a listen-only mode. Once management finishes its presentation, we'll hold a question-and-answer session. Those of you who dialed in by phone and would like to ask a question, please press *5 on your telephone to raise your hand. If you're joining via the Tencent Meeting or VooV Meeting application, please click the Raise Hand button at the bottom left. Please note that today's webinar is being recorded. Before the presentation begins, we'd like to remind you that it contains forward-looking statements, which are subject to a number of risks and uncertainties and may not come to pass in the future for various reasons.
Wendy Huang
executiveDetails on overall market circumstances are drawn from multiple sources beyond Tencent. This presentation also includes certain unaudited non-IFRS financial metrics, which ought to be viewed alongside — rather than as a replacement for — the group's financial performance figures prepared under IFRS. A thorough look at risk factors and non-IFRS metrics appears in our disclosure documents located in the IR section of our website. Now I'd like to introduce the management team joining us on the webinar this evening. Our Chairman and CEO, Pony Ma, will begin with a brief overview. President Martin Lau and Chief Strategy Officer James Mitchell will then deliver a business review, after which Chief Financial Officer John Lo will wrap up with a financial discussion before we take questions from the floor. I'll now hand things over to Pony.
Pony Ma
executiveThanks, Wendy. Good evening, and thanks to all of you for joining us. We're beginning 2026 with meaningful early strides on our new AI products, while also continuing to apply AI to expand the core business we already have. Hunyuan3 Preview — built by our rebuilt team of AI researchers on rearchitected AI infrastructure — leads its parameter size class, offers practical utilities along with cost efficiency, and has held the top spot in OpenRouter's token measure since April 28th. Our productivity AI agent solutions have gained early traction, and we believe WorkBuddy is right now the productivity AI agent service with the widest use in China. Engagement, revenue, and profit across our core businesses keep growing, which supplies the cash flow to fund our AI investments and gives us use cases for deploying AI going forward. Turning to our financial numbers for the first quarter: total revenue came in at RMB196 billion, rising 9% year-on-year. If we adjust for how the later Spring Festival this year, versus last year, affected revenue recognition for our value-added services, total revenue would have risen 11% year-on-year. Gross profit reached RMB111 billion, up 11% year-on-year. Non-IFRS operating profit was RMB76 billion, an increase of 9% year-on-year. Leaving out new AI products, non-IFRS operating profit came to RMB84 billion, up 17% year-on-year.
Pony Ma
executiveNon-IFRS net profit attributable to equity holders came in at RMB68 billion, which is an 11% increase year-on-year. Now moving to our key services: in communication and social networks, the combined MAU for Weixin and WeChat rose both year-on-year and quarter-on-quarter, reaching 1.4 billion. On digital content, Tencent Video strengthened its leadership position in animate series, with eight animated titles placing among the top 10 industry-wide during the first quarter of 2026. In games, Honor of Kings and Peacekeeper Elite both set record highs for gross receipts in the first quarter of 2026, and our newly launched creature-collecting title Roco Kingdom: World saw breakout success. Tencent Cloud's AI agent solutions delivered rapid growth along with a healthy retention rate. I'll now pass things over to Martin for the business review.
Martin Lau
analystI appreciate that, Pony, and welcome everyone joining us this evening or this morning. Total revenue for Q1 2026 rose 9% compared with the same quarter a year earlier. As Pony noted, that overall top-line increase was affected by Spring Festival falling later this year than it did last year, which pushed a greater portion of games-related revenue recognition into later periods. If we adjust for when Spring Festival occurred, like-for-like total revenue would have grown 11% versus a year ago. Turning to our segments: VAS accounted for 49% of total revenue, and inside that, Social Networks made up 16%, Domestic Games 23%, and International Games 10%. Marketing Services contributed 19% of total revenue, while FinTech and Business Services was 31%. Gross profit in Q1 climbed 11% year-on-year to RMB111 billion. VAS gross profit rose 9% year-on-year and made up 54% of overall gross profit. Marketing Services gross profit grew 19% year-on-year, accounting for 19% of total gross profit. FinTech and Business Services gross profit was up 13% year-on-year and contributed 28% of total gross profit. Now, a quick update on what we're doing with AI. During the past six months, we've achieved meaningful advances with our Hunyuan large language model. I'd characterize it as merely getting underway.
Martin Lau
analystThe initiative began with a total rebuild of the team working on our foundation models, built around newly recruited top-tier AI researchers and engineers who bring deep knowledge of large language models. This new group is youthful, high-energy, and tightly knit, which lets us move fast in this extremely fast-moving AI period. Back in February, we rebuilt the pre-training and reinforcement learning system and process from scratch. We redesigned the infrastructure so it would be robust, scalable, and efficient across pre-training, data, and reinforcement learning. On the data side, we grew our dataset substantially and made our collection, cleansing, and synthesis capabilities stronger, with quality as the priority. On training, we improved the pre-training and supervised fine-tuning process, and we expanded reinforcement learning. As for evaluation, we're stepping away from pursuing public benchmarks that can be gamed. Rather, we assess our model using the newest exams, human tests, product feedback, and internal tasks, so we can see how it truly performs in real-world conditions. In April, Hunyuan3 Preview was released. The goal when we began building this model was a cost-efficient, solid model for a wide range of applications that would reduce risk in scaling toward larger models. Hunyuan3 Preview's core design principles were to provide comprehensive intelligence and cost efficiency, tuning it for real-world deployment.
Martin Lau
analystWe've gone past slim specialization into broad intelligence—things like combining reasoning, long-context comprehension, following instructions, dialogue, coding, and tool-use abilities. By co-designing inference alongside the model, we can cut costs substantially, making the intelligence cheap enough for large-scale use. Hunyuan3 Preview has met those expectations. In China, the model has already emerged as a top reasoning model, and it has shown itself effective in real-world software engineering and other agentic productivity tasks. On the internal side, it's been rolled out across 131 heavily used in-house products, among them Yuanbao, QQ, and WorkBuddy, which gives us useful feedback and lets us iterate through the co-design process. On the external side, Hunyuan3 Preview has drawn a warm reception from users and developers in real applications. Since April 28th it has held the number one spot among all models on OpenRouter by token usage, and it kept that lead even after its free period wrapped up on May 8th. Big strides have come with Hunyuan3 Preview, but we see this as only the initial step for Hunyuan large language models. Scaling to bigger models is what comes next. Our Hunyuan team is already at work on a model with more parameters, drawing on our infrastructure and what we learned from Hunyuan3. Through gathering larger and higher-quality datasets and scaling up more powerful reinforcement learning, we can bolster the model's contextual understanding, sharpen its agentic capabilities in areas such as coding, and raise its general intelligence.
Martin Lau
analystBy co-designing and working together with other product teams at Tencent, we are refining how datasets get selected and concentrating reinforcement learning on use cases with high value. Past foundation models, it has grown clearer and clearer that agentic AI is a breakthrough use case now that AI chatbots have gained popularity. AI agents deliver more value in boosting productivity, ranging from the first use cases that helped programmers write code, like with our product CodeBuddy, to now serving a broader set of workloads and occupations, as with Claude and WorkBuddy. More powerful models made these breakthroughs possible, along with the harness infrastructure that lets models use tools and serve as interfaces enabling users to manage agents well. Our platform naturally carries many advantages for hosting AI agents, since users can control AI agents via our communications and browsing interfaces, including Weixin, WeCom, QQ, Yuanbao, and QQ Browser, plus third-party applications. Users may also pick which model to use from a broad range of models according to their own needs and preferences regarding token costs and performance. Going forward, AI agents will be able to reach our Mini Programs ecosystem, using Mini Programs codes as AI skills. Tencent has built an early lead in agentic AI deployment, as shown by the leading DAU of our product WorkBuddy.
Martin Lau
analystEven though we're still at an early stage of adoption, CodeBuddy and WorkBuddy are already posting robust organic growth along with strong retention among both active users and paying users. Among early adopters, the substantial time spent and the frequent interaction with AI agents work as a virtuous feedback loop back to Tencent, letting us spot and deliver complementary software and services, which then fuels greater AI agent usage across a wider enterprise and prosumer user base. As people put more AI agents to work on more complex tasks, paying user conversion rises, producing rapid growth in token usage on Tencent Cloud over recent weeks. With that, I'll hand things over to James.
James Mitchell
executiveThanks, Martin. Now looking at our business segments: revenue for Value-added Services came in at RMB96 billion, which is a 4% increase compared with the same period last year. Social Network revenue fell 2% year-on-year to RMB32 billion, which mirrors lower reported revenue from sales of app-based game items in China stemming from the later Spring Festival. Subscription revenue for long-form video dropped 2% year-on-year as a result of fewer top-tier drama series being released. Even so, we strengthened our leadership position in animated content, since 8 of the series we self-commissioned placed in the top 10 across every video platform in China for the quarter. We think Tencent Video holds competitive advantages when it comes to creating animated series — among them our capacity to cross over IP from China Literature and from our games into animated IP, plus our use of technology tools like Unreal Engine and generative AI for storyboarding and for producing that animated content. At Tencent Music, subscription revenue rose 7% year-on-year, powered by gains in ARPU and subscribers. On domestic games, gross receipts were up at a teens percentage rate year-on-year, mainly because of Delta Force, Peacekeeper Elite, Honor of Kings, and VALORANT Mobile. Revenue growth of 6% year-on-year, though, trailed the growth in gross receipts, since the later timing of the Spring Festival pushed some revenue recognition into the second half of 2026.
James Mitchell
executiveRevenue from international games rose 13% compared with a year earlier, with the main contributors being Clash Royale, Wuthering Waves, and VALORANT. Turning to Communications and Social Networks, transaction volume on Mini Shops kept up a fast pace of growth. To give merchants of branded goods better support, we rolled out incentives that include preferential take rates for brands, product subsidies, and prioritized recommendations. GMV from branded merchants more than tripled versus a year ago in the first quarter, especially in key categories like FMCG and beauty. On the consumer side, frequent buyers can take advantage of our new coupon-sharing feature, sending discount coupons to friends through chats and benefiting from Weixin's virality. For content creators, we upgraded the matchmaking mechanism, so creators can reach relevant merchants more efficiently and promote those merchants' products in the creators' video accounts and official accounts. On the AI side, we've made Weixin and QQ serve as communication interfaces for controlling AI agents, which lets users orchestrate agents from mobile to carry out complicated tasks on PC and cloud. We scaled up the parameter count and improved the algorithm behind the Video Accounts content recommendation model, so it understands users' interests more deeply and recommends content that is more personalized and relevant. Total time spent on Video Accounts climbed over 20% versus a year earlier. As for Mini Programs, we've upgraded the developer toolkit architecture, so users can make better use of AI plugins, including CodeBuddy, to build and debug Mini Programs, and in time we will let Mini Program code evolve into skills for AI agent use within Weixin.
James Mitchell
executiveWeixin Search total query volume rose more than 25% year-on-year, helped by foundation model-driven ranking and by expanding AI search coverage to take in image-based queries. Turning to Domestic Games, in the first quarter Honor of Kings hit a lifetime quarterly high in gross receipts, helped by top-tier outfits drawing on Chinese ink paintings and ancient Silk Road culture. At Peacekeeper Elite, peak DAU hit a lifetime high of 90 million, while gross receipts climbed more than 30% year-on-year. Growth in users and gross receipts was driven by the game's expanding user-generated content platform, a new class-based extraction mode, and IP collaborations with partners including Ferrari. Delta Force likewise set lifetime highs for the quarter in average DAU and gross receipts, helped by a Cosmic Guardian skin, the Morphosis season with its revamped Space City map, and cross-over promotions. Among the newer titles, VALORANT Mobile has ranked consistently in the top 10 mobile games industry-wide since launching in August last year, and in the first quarter we put out distinctive weapon items, including Ignite Capsule, which drove strong gross receipts. Roco Kingdom: World, a creature-collecting open world title, launched on March the 26th. In its first month on the market, the game has drawn over 13 million average DAUs and is maintaining high user retention, with its mobile version ranked consistently among the top 10 mobile games in China by gross receipts.
James Mitchell
executiveZooming out from the individual titles, AI is giving us more and more useful tools, which help our game developers produce additional content and better experiences. Right now, the biggest wins from AI in games come in areas such as speeding up 3D asset creation and animation, making player experiences richer through smart in-game guides, and producing more lifelike graphics through AI rendering methods. Within our international games, PUBG MOBILE saw its gross receipts rise year-on-year in the first quarter, helped by Eastern mythology-themed outfits, brand collaborations, and anniversary events. At League of Legends, Riot has carried out a major overhaul of team processes and technology over the last two to three years, and those efforts are now together producing a quicker pace of major new content launches, among them the popular ARAM Mayhem constant team fight mode, a warmly received new season in For Demacia, and the Revenant Reign exalted tier outfit for the champion Viego. Because of this, League of Legends is seeing a resurgence in DAU and gross receipts. Wuthering Waves' new season, For You Who Walk in Snow, brought a new map, content, and characters, which contributed to that game's gross receipts growth during the period. Following a consolidation period, Brawl Stars' DAU and gross receipts climbed rapidly year-on-year, helped by new abilities called Buffies, an upgraded Brawl Pass, and a better trophy system.
James Mitchell
executiveTurning to Marketing Services, our revenue climbed 20% year-on-year, reaching RMB38 billion, and growth stood out in the Internet services, e-commerce, and games categories. The pace of revenue growth was better than the 17% year-on-year we posted in the fourth quarter, driven by us going deeper in collaboration with e-commerce platforms and by advertising demand recovering as inventory from Video Accounts increased. AI Marketing Plus, our automated campaign management solution, was behind roughly 30% of all Marketing Services spending by advertisers on our platform during the quarter. We upgraded the runtime advertising recommendation models to a unified transformer-based architecture. That upgrade gives us a deeper grasp of user contexts and intent, while balancing model complexity against system efficiency. Looking at inventory, Video Accounts ad impressions rose quickly year-on-year, helped by higher total time spent, more video views, and greater ad load. We made available additional rewarded ads inventory, which gives advertisers high click-throughs. Mini game and mini drama studios raised their marketing spending inside Mini Programs. For mini games in particular, we launched an instant-play advertising format that lets users play the game right within the ad, with no redirection needed, cutting friction in the process of acquiring new users. On our Mobile Ad Network, we've opened up more rewarded or incentivized ad formats to third-party app developers, which brings in higher-price bidders and therefore more revenue.
James Mitchell
executiveNow within FinTech and Business Services, segment revenue reached RMB60 billion, marking a 9% increase compared with a year earlier. Growth in FinTech services revenue came mainly from commercial payment volume as well as wealth management services. On the commercial payment side, year-on-year growth outpaced what we saw in Q4 of last year, helped by a continuing rise in transaction counts and also by greater value per transaction in areas like retail and dining services. In wealth management, both average assets per user and user numbers rose year-on-year. Moving to Business Services, first-quarter revenue climbed 20% versus a year ago, fueled by stronger demand and an improved pricing environment for our cloud services, together with higher technology service fees stemming from Mini Shops e-commerce. At Tencent Cloud, AI-related demand drove year-on-year revenue increases across GPU, CPU, and storage. We enhanced Tencent Cloud's AI agent solutions with proprietary security infrastructure, skill hubs, and control interfaces, which helped usage grow quickly and brought initial token monetization. Revenue for Tencent Cloud's international business rose more than 40% year-on-year as we broadened our global footprint and met demand for our platform-as-a-service solutions, among them media processing services and the TDSQL cloud database. With that, I'll hand things over to John.
John Lo
executiveThank you, James. Turning to the first quarter of 2026, our total revenue came in at RMB196.5 billion, which represents 9% growth year-on-year. Gross profit reached RMB111.3 billion, an increase of 11% year-on-year. Operating profit stood at RMB67.4 billion, up 17% year-on-year. Interest income amounted to RMB4 billion, 7% higher year-on-year, largely on the back of growth in cash reserves. Finance costs were RMB3 billion, versus RMB3.9 billion in last year's same quarter, mainly reflecting higher forex gains and lower average interest rates. Our share of profit of associates and joint ventures was RMB3.6 billion, compared with RMB4.6 billion in the same quarter a year earlier. On a non-IFRS basis, share of profit came to RMB7.1 billion, compared with RMB7.6 billion in last year's same quarter. Income tax expense rose 6% year-on-year to RMB14.5 billion. On non-IFRS financial figures, operating profit was RMB75.6 billion, up 9% year-on-year. Operating profit excluding new AI products reached RMB84.4 billion, up 17% year-on-year. Net profit attributable to equity holders was RMB67.9 billion, up 11% year-on-year. Diluted EPS came in at RMB7.364, up 12% year-on-year, outpacing non-IFRS net profit growth thanks to a reduced share count under our share buybacks. Now moving on to gross margins for Q1: overall gross margin was 57%, up 1 percentage point year-on-year. Looking at segments, VAS gross margin was 63%, up 3 percentage points year-on-year, primarily driven by increased revenue contributions from internally developed games.
John Lo
executiveFor Marketing Services, gross margin came in at 55%, a decline of 0.5 percentage point compared with a year earlier. That reflects a higher cost of revenue, which includes depreciation on AI-related equipment and the associated operating costs, as we ramp up AI investments aimed at delivering more relevant content recommendations going forward. Turning to FinTech and Business Services, gross margin reached 52%, which is 2 percentage points higher year-on-year, driven by an improved revenue mix within FinTech services. On Q1 operating expenses: selling and marketing expenses totaled RMB11.3 billion, an increase of 44% year-on-year, reflecting greater promotional efforts supporting growth of our AI-native applications and games, including new games. R&D expenses climbed 19% year-on-year to RMB22.6 billion, mainly because of our stepped-up investment in AI, which pushed up equipment depreciation and associated operating costs, along with higher staff costs. G&A expenses, excluding R&D, fell 24% year-on-year to RMB11.3 billion, reflecting a high base in the prior-year period that included a one-off RMB4 billion share-based compensation expense tied to the restructuring of an existing commercial arrangement at an overseas subsidiary. At quarter end, we had roughly 115,000 employees, up 5% year-on-year, driven by headcount additions to games and our technology platform, including AI-related headcount, and down 1% Q-on-Q, mainly due to lower headcount at our independently operated subsidiaries and in-house tech support subsidiaries.
John Lo
executiveTurning to Q1, our non-IFRS operating margin came in at 38.5%, holding steady compared with a year earlier. When we exclude new AI products, the non-IFRS operating margin reached 43%, which represents a 3.1 percentage point increase year-on-year. To wrap up, let me point to a few important cash flow and balance sheet figures. Operating CapEx totaled RMB31.2 billion, rising 18% year-on-year and 84% quarter-on-quarter, reflecting our stepped-up investment in server infrastructure. Non-operating CapEx amounted to RMB0.7 billion. Free cash flow reached RMB56.7 billion, a 20% year-on-year gain, fueled by higher gross receipts from games and advertising billings, partially offset by increased spending on server infrastructure and compute. Sequentially, free cash flow rose 67%, which reflects seasonally stronger game gross receipts and the timing of certain seasonal accounts payable settlements, partially offset by greater server infrastructure and compute spending. Our net cash position stood at RMB146.9 billion, an increase of 37% quarter-on-quarter, or roughly RMB40 billion, primarily driven by free cash flow generation, partially offset by RMB7.9 billion of share repurchase and RMB7 billion in net cash outflows tied to investments in other corporations. Thank you.
Wendy Huang
executiveThanks, John. At this point we'll begin the Q&A session. Our first question comes from Alicia Yap at Citigroup.
Alicia Yap
analystHi there, is my voice coming through okay?
Martin Lau
analystCorrect.
Alicia Yap
analystGood evening, management, and thanks for taking my questions. There are two of them. First, since Hunyuan3 Preview came out, I believe management noted that the model has been deeply integrated into many internal core products, including Yuanbao, IMA, and WorkBuddy. So can management share some details on the performance improvements you've seen in these workflows since adoption? And beyond that, what's the roadmap for bringing Hunyuan3 more broadly into WeChat workflows? Will enterprises be able to use these agentic workflows to boost their productivity and perhaps future monetization? My second question: as agents increasingly have the potential to replace traditional click-throughs on web pages and also apps, could management share your view on future advertising pricing and the resulting impact on advertiser budget? Which types of digital content or web activities are likely to stay more resilient, continuing to capture more significant user engagement? Are we thinking ahead and strategically perhaps positioning our ad formats to adapt to this potential shift in ad spending? Any thoughts or insights you could share would be helpful. Thank you.
Martin Lau
analystThank you, Alicia. Regarding Hunyuan3, we've provided a fairly thorough overview of it. As the prepared remarks show, it's more intelligent, and despite being a smaller model, it's actually very strong when it comes to reasoning. At the same time, it delivers significant improvement over Hunyuan2 on agentic capabilities. So when we actually integrate Hunyuan3 into the various products, the performance was, I'd say, quite encouraging. All the different products have reported marked improvement in performance, because they're seeing it from the users' end. Total token usage is actually at least 10x compared to Hunyuan2. That's a clear indication that Hunyuan3 is in fact well designed. At the same time, since it benefits from a co-designing process with some of the major products, for example, Yuanbao and WorkBuddy, right? That's why the products have received it well. As for integration into the Weixin workflow, I think it will be a step-by-step process. Weixin itself has actually always been using some part of their product Hunyuan2, and they've already upgraded to Hunyuan3. In some cases, they use different models, and they evaluate the different models and assess what's the best model to use for their users, right?
Martin Lau
analystHunyuan3 keeps improving, so adoption will keep rising. On Mini Programs, one advantage for enterprises is that they'd be using products we've enabled with Hunyuan, like CodeBuddy and WorkBuddy, correct? When these companies or their users are using those products, they'd gain from Hunyuan3's better performance. Also, down the road, once we begin integrating Mini Programs as skills so agents can use them as tools, that would bring more traffic to these Mini Program enterprises. One side is internal: using our own agentic products lets them raise their own productivity. The other side is external: it can help their Mini Programs get used by more users and more agents going forward.
James Mitchell
executiveAlicia, thanks for that question about advertising—it's an interesting one. For e-commerce companies, this could potentially be more of a problem than it is for us, since users actively choose and want to spend their time watching short videos, listening to music, consuming content, or chatting with friends, whereas, generally speaking, when users spend time on e-commerce, it's because they're trying to find the lowest price. It's not because they necessarily enjoy that process. If AI agents play a bigger role in the future in facilitating price comparison, then it's possible that users will spend less time on e-commerce sites and be less exposed to ads than they are today, while the AI agents can scan infinite listings and are therefore not influenced by ads the way that human beings with a finite attention span are influenced. That said, there have been many prior iterations of price comparison services, including search engines, and the big e-commerce companies have generally thrived despite the existence of those price comparison services. I think it's premature for us to have a definitive view at this point on how it'll affect our friends in the e-commerce industry, but we don't see it as a primary risk for Tencent.
James Mitchell
executiveAppreciate it.
Wendy Huang
executiveThanks. Our next question comes from Kenneth Fong at UBS.
Kenneth Fong
analystGood evening, management, and thanks for taking my question. I’d like to ask about the ROI on AI investment. When we look at global peers, they’ve been putting 80 to even 100% of their operating cash flow into AI CapEx, while we invested roughly 35% in the last quarter. That said, many of them have seen free cash flow and ROE decline as their businesses became more asset-heavy. Could management share or give more quantifiable guidance on AI-related CapEx for this year, and which KPIs are used to assess value creation and the return on this investment? Thank you.
James Mitchell
executiveHi, Kenneth. Demand is rising from both internal products and external users of our model for our AI-related services. Earlier, we guided that CapEx would go up this year compared with last year, and our confidence in that guidance has now grown. Expect a substantial CapEx increase, particularly in the second half of this year, as more China-designed ASICs reach us month by month throughout the year. On the KPIs, they vary product by product and business by business. At a high level, for existing activities like advertising and games, the KPIs would relate more to revenue and profit. For our new AI products, the KPIs would center more on capabilities — how intelligent our foundation model is — and usage — how much token consumption is happening on WorkBuddy related. Then for Tencent Cloud, where until now we actually haven't had sufficient GPUs to begin servicing external demand, the KPIs will relate more to revenue and market share. Thank you.
Kenneth Fong
analystThanks.
Wendy Huang
executiveThanks, Kenneth. Now we'll go to Alex Yao at JPMorgan for the next question.
Alex Yao
analystThanks, management, for taking my question. I want to build on what Kenneth asked, though from a somewhat different angle. As you direct financial resources toward CapEx and AI investments, in what way do you assess the spending on AI infrastructure internally? What ROI framework or payback period are you using to underwrite this investment, and across what time horizon? Thank you.
James Mitchell
executiveI believe we shared some perspective in the earlier response on how we think about returns on investments, whether measured quantitatively or qualitatively. For certain products, we adopt a very strict financial lens, while for others we place more emphasis on the long-term benefits to our franchise. I'm not sure whether you'd like to restate the question, since I may not have fully grasped what you were getting at.
Alex Yao
analystThanks, James. The key thing the investment community has to grasp is this: who ultimately foots the bill for this AI CapEx, which budget does it come out of, and across what timeframe — over what horizon — do we anticipate being able to access those new commercial opportunities?
James Mitchell
executiveLooking back across Tencent's history, I'd say we've generally delivered solid returns, and you can measure that through our return on equity across the past 20 years or so, which has stayed consistently high. We didn't achieve that by holding every new product and every new service to very short-term quantitative ROI targets. What got us there was running the portfolio as a portfolio and managing products across their entire life cycle rather than over a particular quarter or 12-month window. Many offerings at Tencent — our push into games, the release of the Weixin app, our move into payments — passed through long incubation phases with no return on investment, yet we had confidence in the franchise value being built. Later, over time, came longer harvesting phases in which we generated very healthy returns on that sunk investment. AI covers a mix of shorter-cycle and longer-cycle investments. When we purchase GPUs and put them to work in our ad tech, that's a comparatively short-cycle investment. Those GPUs deliver improved targeting, higher click-through rates, and greater revenue and profit on a fairly accelerated timeline.
James Mitchell
executiveConversely, putting GPUs to work in our Hunyuan foundation model is a move we see as valuable for our franchise, and there we're adopting a longer-term perspective. To repeat, we don't run any individual product on a quarterly basis. What we run is the portfolio, and we run the products across their entire life cycle.
Martin Lau
analystTo expand a bit on what James said, there's kind of a range of ways to think about this. Looking at model training, it's essentially an investment in the future. You're probably not going to get much immediate return, but as time goes on, the capability builds up, and it really kind of helps unlock plenty of different business opportunities. Then, when you look at the products we're launching — whether that's Yuanbao, WorkBuddy, CodeBuddy — there's a process where you have free services, and over time, you may generate revenue. Business-oriented revenue can arrive faster than consumer-oriented revenue, and there would be a return cycle on that. Then, if you look at business revenue, or cloud revenue in the sense of MaaS or compute rental, there's a clear ROI. You have a depreciation cost, then you typically add some kind of margin, and then you rent it out. That would have a clearer return. Then with advertising, as James pointed out, we usually see very good returns from those investments. I think, for different computes, there are different ways to look at it.
Wendy Huang
executiveThanks, Alex. We’ll go next to William Packer at BNP Paribas. William, your line is open.
William Packer
analystHello, management. Thanks a lot for fielding my questions. First, since domestic gaming is your largest source of revenue and free cash flow, can you walk us through how generative AI is affecting that business right now? Is it producing extra monetization through quicker content creation cycles, or is that a benefit that comes later? Are cost efficiencies helping margins at present, or must you invest the initial safe and that limits any gain? As a brief follow-up, I appreciate the commentary on the CapEx outlook. Could you break down any consequences for the share buyback in the second half of the year? Thank you very much.
James Mitchell
executiveLet me go ahead and take those. Just as you suggested, in our game business, generative AI lets us create content at a quicker pace, and in certain cases that content improves the player experience overall, while in other cases it leads directly to monetization — a virtual outfit would be one example. That is what we're doing, and that is what we're seeing. We believe we lead in China on this, and to some degree we're even more of a global leader when it comes to putting that capability to work and realizing that benefit. Right now the goal really is quicker content creation and generating incremental revenue. Margin expansion per se is not what we're prioritizing. Rather, as we deliver the revenue uplift we're seeing, and if we can hold headcount fairly steady, then I suppose mathematically that combination would tend to produce higher margins over time — but that's more of a happy outcome than the intention behind the process. Then, on your question regarding capital returns, we will be increasing our AI investments in response to the greater demand we're seeing. That said, we do have a very cash-generative business, as the first quarter results show.
James Mitchell
executiveBeyond that, we hold a sizable investment portfolio, and we're speeding up efforts to liquidate portions of it, which will let us keep up buyback activity for the remainder of this year. Right now, we feel our share price is a bit dislocated, so this is a good moment for buybacks — an especially good moment for buybacks. Thank you.
William Packer
analystAppreciate that context.
Wendy Huang
executiveThanks. Our next question is from Robin at Bernstein.
Robin Zhu
analystThanks, management, and thanks for taking my question. I have a couple if I could. First, regarding the changes we've made to our data pipeline and training, RL, inference and so on, do you feel that now, having seen the release of Hunyuan3, we're on a more sustainable upward trajectory where we can deliver major updates — say, one major update a year, with several smaller ones in between, like some of the other labs? Are we there, or is that still more of a work in progress? Then second, just on the philosophy behind some of the AI product investments. I think one of the tensions in the US has been that OpenAI has been the big DAU player, whereas Anthropic has kind of gone after a small pool of high-intent power users, highly willing payers, effectively. Tencent obviously historically has been the former, but I'm just curious your thoughts on the relative merits of going down one path versus the other. To what extent does the current token-maxing phenomenon play into that thinking? Thank you.
Martin Lau
analystOn the production pipeline side, I believe we’ve already discussed at considerable length how we see it as genuinely making strong progress. If you look at Hunyuan3, we have overhauled the whole team, the production process, the infra, and every one of the major modules involved in producing great models, right—including the data pipeline, as you point out, the pre-training, the post-training, the RL, as well as eval. We deliberately built a smaller model in order to validate all of these different points, right? The combination of this, right, once it’s all integrated into Hunyuan3 Preview, is that it yields a pretty competent model at its size. We clearly see in each one of these modules that there’s a lot of work we could be doing. I think we’re happy with the result. To some extent, we are actually surprised by the speed at which it’s done and the fact that it’s actually proved to be useful, right? For a long time, I think a lot of models would actually come up pretty high in terms of the benchmarks, but then when it’s actually rolled out to different products, people complain, and when you put the model to the developers and the user, right, people won’t use it.
Martin Lau
analystWhen you consider how this is being received in real-world use cases, it's actually outperforming what we expected by a fair margin. That, I think, gives us a really solid foundation to scale the model up to the next level. As for how we view the various products, we felt this is really quite an early stage when it comes to AI diffusion, right? We'd expect many different products to emerge as we go forward. At first it was chatbot, and everyone felt chatbot was the king of products, right? Then all of a sudden, coding showed up, and that becomes an even more eye-catching and significant use case, because it carries very high value, right? Now we're seeing agentic capability spreading, right? I think that will actually let AI diffuse into different industries, and you'll have many different agents emerging that can help you get work done, right? New products are going to keep coming. I think that will continue to propagate. To some extent, I think, in the AI world you actually have to find the high-value use case, rather than purely focusing on DAU, because the difference between the AI revolution and the Internet is that this is about intelligence, and intelligence shows its value in how much people are willing to pay for it.
Martin Lau
analystAnd at the same time, intelligence isn't free — correct? In the Internet world, you essentially have mostly information that already exists, and then you also generate some new information and content, but that ends up being a fixed cost. After that, the variable cost of delivery is actually quite small — right? All you need to cover is bandwidth, and the compute sits on people's devices — right? Because of that, you can almost achieve infinite scaling. In this case — right? — each individual delivery to a DAU actually costs you a fair amount — right? Consequently, you can't simply take the same logic as the Internet and apply it to AI. What I'd say is that the ability to identify high-value use cases is going to matter as much as — if not more than — just blindly chasing a lot of DAU and user time. I think that's one key distinction. When we think about how to deploy our product and how to co-design our product alongside the models, these are the kinds of new considerations we need to bring into play.
Wendy Huang
executiveThanks. Thanks. We’ll go on to the following one, Alex Liu with Bank of America.
Alex Liu
analystOh, thank you, management, for fielding my questions. There are two things I'd like to ask. I appreciate you sharing the details on the agentic AI strategy. What particularly draws my attention is the idea that AI agents from Tencent might be able to reach the Mini Program ecosystem and employ Mini Program code as AI skills. On that point, I'm just wondering whether there's a timeline we ought to expect for it to become a reality. My follow-up ties to the previous questions. With computing resources in such tight supply at present, I was wondering how management is weighing the speed of rolling out more AI features, like Weixin Agent, against the rather constrained compute capacity currently available on hand? Thank you.
Martin Lau
analystRegarding Mini Program, I believe this is something that's on the way. I think we need to work out what the best approach is for presenting these, and how Mini Program owners can genuinely and actively engage with this. We do have a timeline. We won't be able to give you a definitive answer, because a great deal of design work still has to fall into place, correct? At some point, I think this is a concept where our ecosystem can truly help one another, and that's one unique advantage we hold. Over time, many potential ecosystem resources can in fact be converted into skills for agents. Over time, agents would in fact have their own identity and be able to obtain accounts across some of the services as well, right? I think that's a unique advantage we hold, and we'll be offering that over time.
James Mitchell
executiveRegarding your question about how we balance the different AI products we want to build in-house, the truth is we've already made that choice and borne the cost, because we put a range of internal services ahead of Tencent Cloud. I think most large hyperscale tech companies that run cloud businesses have a single flagship internal use case to which they devote a big share of GPUs. We have several flagships. There's the Hunyuan foundation model, there are agentic developments inside Weixin, there's Yuanbao support, and there's AI deployment for advertising, for games, and now also for the WorkBuddy and CodeBuddy use cases. The reason we've managed to support all of them at the same time is that we haven't been actively leasing out GPU capacity in Tencent Cloud. Now, as we look across the remainder of this year, once supply of China-designed GPUs gradually ramps up, we'll be fixing that situation, and we'll make more capacity available in Tencent Cloud, which will in turn push up Tencent Cloud's rate of expansion. That's where the trade-off was made: we have deliberately been slow to monetize the AI opportunity through Tencent Cloud, because we've been supporting a number of AI initiatives internally at the same time.
James Mitchell
executiveAppreciate it.
Wendy Huang
executiveThanks, and we’ll go ahead and take the next question from Charlene Liu at HSBC.
Charlene Liu
analystThanks, Wendy, and I appreciate the chance to ask. There are 2 questions I have. The first concerns monetization. Doubao, as we can see, is beginning to test subscription offerings aimed at its 2C users. What I'd like to grasp is how large management believes that 2C subscription market is in China. Beyond subscriptions, I'm wondering whether advertising and Mini Programs or Shop represent the main monetization channels for 2C AI, or whether there are additional ones. What amount of upside do you anticipate from ads and Mini Programs? That's my first question. My second question relates to the compute power bottleneck. Clearly, the US players have kind of progressed to tackling bottleneck problems or shortages in CPU and networking chips past GPU. Have such issues started to emerge? Are we expecting them as well? What does management intend to do to address them? Many thanks.
Martin Lau
analystOn 2C monetization, I’d say it really isn’t easy, right? If you look at global standards, in the Western market, where paid service penetration is actually very strong and the living standard is very high, the subscription price there is multiple times what an equivalent service in China costs, whether it’s a music service or a video service. Paying penetration is probably in the single digit, right? When you sort of apply that to China, I think the subscription model won’t be that big for the China market. If you use that as the standard and then start to read across for China, then it wouldn’t be that much. At the same time, it’s necessary, right, for the reason I talked about earlier. It’s not like Internet services, where you can have a very low cost of scaling your services, right? Every single user actually sort of costs you something in terms of variable cost. I think the more important implication is that when you have to have payment to support a service, then most likely the service is not going to be a winner-take-all business.
Martin Lau
analystSo essentially, it would be a situation where multiple players are supported, each holding a portion of the market, and every one of them would have certain users as well as some share of subscriptions. Beyond that, when we consider e-commerce or advertising as monetization routes, I think it's still very early as well. Even for the US players, where the eCPM is in fact significantly higher, the leading player hasn't managed to launch a particularly robust advertising model. I think that will come over the longer term, and it will be a supplement to, probably, a subscription model. That's why I said that in the world of AI, when you apply the compute and apply the model to different use cases and different applications, you actually need to think about which is truly the high-value use case, so that you get the best return on your limited compute in order to achieve the best results.
James Mitchell
executiveTo address what you asked regarding the different bottlenecks across GPU, CPU, networking and so on — just to recap: the GPU bottleneck has been far more pronounced in China than elsewhere, and that stems from two things combined. First, policy restrictions affect certain foreign-designed GPUs being imported into China; second, China-designed GPUs face limited fab capacity inside China. Consequently, the country has really experienced a shortage of GPU or ASIC capacity. That is now being resolved, since China-designed ASICs are getting more supply from fabs within China and also more supply from fabs in neighboring countries. By contrast, we haven't encountered those kinds of artificial extra constraints on CPU or networking chips. We were a major purchaser of CPU and networking chips for many years before GPUs became such a big presence in data centers. We hold very long-term relationships with the companies that supply the CPUs and supply the networking chips. From their side, while one might think these suppliers would be sitting back and simply selling at the highest possible price into the spot market, that isn't actually the reality. The smart suppliers are taking very conscious 3 to 5-year forward views and negotiating long-term agreements in order to give them certainty of their revenue outlook over the next 3 to 5 years.
James Mitchell
executiveAs they make the call on who gets those long-term agreements, they want several partners involved rather than relying on one, and they want partners with a long track record of being there and who will still be there well into the future, and ideally partners whose demand they expect to rise substantially over time. Fortunately, we meet every one of those criteria. For many years we've been a major customer of Intel and AMD and so forth. Our volume with them has been growing steadily across many years, and they expect that we'll keep steadily increasing that volume for many years ahead. On procurement, I'd say the difficulties center more on GPU, and those difficulties are now being addressed. Beyond that, we've managed to lock in solid supply of CPU and networking chips. Thank you.
Wendy Huang
executiveThanks. The next question comes from Ronald Keung at Goldman Sachs.
Ronald Keung
analystThanks, Pony, Martin, James, John, and Wendy. I have two questions. The first concerns consumer AI agents. We've been discussing a possible Weixin Agent, which operates at the app level, and we know Weixin is a super-app — but how does management see the long-term potential, or the possible disruption, from agents sitting at the operating-system level, meaning agents from iOS or Android or ones built by handset makers? How should we view that potential, disruption, or threat? Second, on advertising, we've seen quite a good reacceleration, and we've been very patient on ad load relative to peers. With a slight edge-up in ad load during the first quarter, I'm wondering whether there's any room for fresh thinking or a change that could bring further revenue acceleration in ads — and the reason I ask is that I think bigger advertising profits could fuel more reinvestment into AI. I'd love to hear your thoughts. Thank you.
Martin Lau
analystFrom an operating system standpoint, I think you've blended several different things together, right? There's a genuine operating system — iOS and Android, right? Then you throw in other kinds of applications that attempt to act as if they were operating systems. If you're an operating system such as iOS or Android, then you genuinely want to guarantee that the ecosystem is properly protected and properly curated and, considering whatever you permit applications to do, right, you really want to maintain that balance, right? An agent can try to offer services to your users, but then you need to obtain the permission, right, of the various applications. If not, then as an operating system you're basically stealing from different apps, and that isn't the best approach to running an operating system. Operating systems have been around for a long time, I think, and the principle behind an operating system is that it's truly neutral, and it genuinely offers a level playing field for all the apps, and in the future, all the agents to work with the operating system. If you say, oh, there's another app that kind of attempts to turn into an operating system-like service and tries to kind of encroach on other apps, I think that's genuine competition, and that's not something any app would truly permit.
Martin Lau
analystI think the operating system itself ought to attempt to prevent that from occurring too. If we're discussing agents that will essentially be an app competing against other apps, I'd say that's one kind of problem, correct? The operating system would, in my view, always aim to remain fairly neutral and to preserve a healthy ecosystem in which everyone can take part, since that's what makes an ecosystem or an operating system successful.
James Mitchell
executiveOn the advertising revenue side, honestly, Video Accounts still carries the lowest ad load in the industry, at 4 to 5%. So there's obviously considerable headroom for us to raise that ad load. As for whether we'll feel the need to do that, and to what extent, we're in a position where advertising isn't our only revenue growth driver — we have several. In gaming, the deferred revenue jump this quarter gives us a tailwind for reported revenue across the coming three quarters. In cloud, we've discussed how adding more GPUs on stream during the second half of the year ought to help cloud revenue growth trends. In FinTech, for many quarters we dealt with a situation where volume growth was positive while pricing was negative. Today volume growth is still positive, and pricing has shifted to a more neutral stance. Across our portfolio of businesses, we believe certain positive things are playing out. We'll keep managing it as a broad portfolio, and we'll keep managing ad load within Video Accounts in a way we believe is conducive both to supporting overall growth of our business metrics and to supporting growth in time spent and engagement inside the Video Accounts product itself.
Wendy Huang
executiveThanks. Our next question comes from Ellie Jiang at Macquarie.
Ellie Jiang
analystThanks a lot for taking my question. Good evening, management. Two questions from me. First, when you look at how WorkBuddy is currently performing, it really does appear that we've built some very early leadership, particularly in the productivity agent space. Of the 20% growth in Business Services this quarter, could you give us some color on what share would be recurring agentic revenue — meaning MaaS- or SaaS-related — versus the traditional cloud revenue and the other pieces? And specifically, do you have any internal ARR target for these agentic workflow products by the end of fiscal '26, and over the next coming two to three years? Thank you.
James Mitchell
executiveEllie, that productivity jump from AI is genuinely something that has come about not over the past few quarters or even the past few months, but over the past few weeks. Honestly, I believe that holds true on a global basis — really, it's only since the close of the first quarter that agentic AI has made its breakthrough, both in its capacity to generate code and in its capacity to help people get more done. During the first quarter, growth in Business Services did not stem from that token consumption. Token consumption came later than the end of the first quarter. As for ARR targets and similar matters, by extension, everything is shifting so rapidly that we might set a target today and find ourselves off by an order of magnitude 12 months from now, in either direction, since usage demand is so dynamic. Right now, we're less focused on reaching particular dollar ARR figures and more focused on getting the products right. Getting the products right covers having the right product at the model level, where Hunyuan3 performs very well across many agentic capabilities, and the next iteration will be substantially better later this year, and also at the product level, with CodeBuddy and increasingly WorkBuddy serving as the right interface for users to access and draw the most intelligence out of these foundation models.
James Mitchell
executiveAt this moment, that is what we are focused on. Thanks.
Wendy Huang
executiveThanks. The final question comes from Thomas Chong at Jefferies.
Thomas Chong
analystHello there, and good evening. I appreciate the management team taking the time for my questions. My first one concerns AI on the content side. Since we've witnessed certain disruptions originating from AI within online video, is it fair to anticipate that Tencent Video will likewise see AI drama become blockbuster content over the next few years? And how ought we to consider the content costs and the business model going forward? My second question relates to AI within the FinTech segment. Because risk management has been substantially strengthened thanks to AI, how should we view the outlook for online lending and wealth management with AI? Thank you.
James Mitchell
executiveOnce more, because the AI landscape has shifted so rapidly and our own standing has evolved enormously over the past few weeks and months, it's hard to make very definitive statements. I think when you mention AI disruption in content creation, you might be referring to short videos or short drama series, not the long-form drama series that has traditionally been Tencent Video's strong suit. On the long-form side, what we observe is that a certain double-digit percentage of the market enjoys animated content. Now, thanks to the combination of tools like Unreal Engine and capabilities like generative AI for video, it's growing increasingly practical to produce the same 3D assets for games and for animated linear video content, and each becomes a truly best-in-class product within its own category. That's an area where Tencent is a natural leader, since we possess our large content IP operations, our gaming business, our AI technology capabilities, and particular strength in certain multi-modalities within AI. As Pony noted during the opening remarks, we have in fact now become a very clear industry leader in the business of making animated TV series, and AI lets us do so faster, cheaper, and better.
James Mitchell
executiveThat also lets us move a lot more IP into linear video format that before was kind of trapped in novel format or in game format, and therefore widen the funnel and widen the audience. So that's on AI for content. Next, on AI for FinTech, financial services makes up a very large share of global GDP. It makes up a very large share of our revenue. It's a sector that is by nature extremely data-heavy, and therefore it's naturally a sector that should and will be suited over time to a productivity uplift from AI. If you consider the sectors already being uplifted by AI, like coding, like advertising, then financial services is in fact very logical to be uplifted in the near future as well, since it shares certain characteristics with coding, with advertising, and so on. As one example, on the lending side, credit scoring has historically been more of an art than a science, and there has been this universe of data available, but only a small subset of that universe is actually fed into the model effectively, whereas now, with transformer-based models, you can kind of take the totality of data available and see what is predictive and improve your loan extension on the basis of that uplift in predictability.
James Mitchell
executiveThis is something I believe a lot of companies are going to pour substantial time and effort into, and it's a space where we'll be taking part too.
Wendy Huang
executiveThanks. That brings us to the close of the webinar. Appreciate everyone being with us on today’s results call. If you’d like to review the press release and the rest of our financial materials, they’re in the IR section on our corporate site, www.tencent.com. A replay of this webinar will be posted shortly as well. Thanks again, and we’ll see you next quarter.
Read the full transcript via the API
You're viewing the first half of this call. Get the complete Tencent Holdings Limited transcript — plus 132,000+ transcripts from 7,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.
Get the API View API docs →This call discussed
For developers and AI pipelines
Programmatic access to Tencent Holdings Limited earnings transcripts and 132,000+ others is available through the
EarningsCalls.dev REST API. Plans from $39.99/month — full transcripts, speaker segments,
full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.
More tools for stock research
OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in Tencent Holdings Limited or any ticker on your watchlist.
StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.