Wendy Huang: Good day, and good evening. Thank you for standing by. Welcome to Tencent Holdings Limited 2026 Second Quarter Results Announcement Webinar. I'm Wendy Huang from Tencent IR team. [Operator Instructions] And please be advised that today's webinar is being recorded. Before we start the presentation, we would like to remind you that it includes forward-looking statements, which are underlined by a number of risks and uncertainties and may not be realized in the future for various reasons. Information about general market conditions is coming from a variety of sources outside of Tencent. This presentation also contains some unaudited non-IFRS financial measures that should be considered in addition to, but not as a substitute for measures of the group's financial performance prepared in accordance with IFRS. For a detailed discussion of risk factors and non-IFRS measures, please refer to our disclosure documents on the IR section of our website. Let me now introduce the management team on the webinar tonight. Our Chairman and CEO, Pony Ma, will kick off with a short overview. President, Martin Lau, will provide a strategy review. Chief Strategy Officer, James Mitchell, will provide a business review; and Chief Financial Officer, John Lo, will conclude with financial discussion before we open the floor for questions. I will now pass it to Pony.
Huateng Ma: Thank you, Wendy. Good evening. Thank you, everyone, for joining us. As we enter into the third quarter of the year, we are making substantial progress toward building a new AI-empowered Tencent in terms of intelligence, applications and infrastructure. At the intelligent level, Hunyuan 3's production version provides users a wide use model with a strong cost performance metrics and serve as a stepping stone toward the Hunyuan family of models, attaining state-of-the-art capabilities in the future. At the application level, our WorkBuddy AI office productivity service and CodeBuddy AI coding tool are achieving user growth and are the clear leaders in the field in China today. At the infrastructure level, we substantially stepped up our procurement of compute, which will enable us to convert usage of our applications and models into revenue going forward. At the same time, we continue to enhance our existing services with a sustained marketing service revenue growth, several successful recently released games and rapidly increasing video views on Weixin video accounts. Looking at our financial numbers for the second quarter. Total revenue was RMB 205 billion, up 11% year-on-year. Gross profit was RMB 118 billion, up 13% year-on-year. Non-IFRS operating profit was RMB 76 billion, up 9% year-on-year. Excluding new AI products, non-IFRS operating profit was RMB 86 billion, up 19% year-on-year. And non-IFRS net profit attributable to equity holders was RMB 68 billion, up 9% year-on-year. Turning to our key services. For communication and social networks, combined MAU of Weixin and WeChat grew year-on-year and quarter-on-quarter to 1.4 billion. For digital content, TME's acquisition of Ximalaya strengthened our audio content and unlocked new synergies with the ecosystem IPs, including from China Literature. For games, new game Roco Kingdom World ranked first by average DAU and by gross receipts among all new games released in China industry-wide this year. For cloud, WorkBuddy recently ranked first among productivity AI service in China based on monthly interactions. I will now hand over to Martin for the strategic review.
Chi Ping Lau: Thank you, Pony, and good evening and good morning to everybody. Today, we want to provide you with an update on our overall AI strategy. Tencent's existing businesses are growing solidly due to intrinsic moats and AI enablement. As discussed earlier this year, our moats arise from factors, including network effect, depth and value-added along supply chain, IP, low take rates, regulatory requirements and private data. In addition to these moats, we're further deploying AI to boost returns in areas, including Weixin, games and advertising. As a result, our existing businesses provide a very strong financial support for our new AI initiatives. Regarding our new AI initiatives, we've made significant progress in constructing a robust foundation, including a substantially improved Hunyuan 3 foundation model with leading cost performance, WorkBuddy and CodeBuddy that lead the China market in terms of AI productivity usage and Yuanbao and Xiaowei serving as gateways to drive broader consumer AI adoption. We see increasing potential to generate attractive financial returns from franchise products with differentiated advantages, including Hunyuan, WorkBuddy and Xiaowei over time. We're comfortable in making significant investments in AI because not only there is a substantial upside potential, there is also clear downside protection. The AI investments we're making are mostly in AI infrastructure. And in the worst case, which we do not believe that would happen, we can choose to rent that infrastructure out at cost recovery or even better prices via Tencent Cloud if needed. Now going on to the different components. First, on Hunyuan Foundation model. The release of Hunyuan 3's full production version is very successful, showing a substantial step-up in performance compared to the Hunyuan 3 preview version, leveraging the feedback loop from product teams to improve the quality and diversity of data used for post training and by scaling up reinforcement learning, Hunyuan 3 achieved a notable improvement in task completion rates and meaningful reduction in hallucination and error rates. The improvements in Hunyuan 3's capabilities are most evident in its agent capabilities and product experience. The model's performance step-up across reasoning agentic and long contest tasks delivered clear advantages for use cases such as coding, office work, financial modeling and front-end design. These performance improvements drove accelerated user adoption and growing external customer demand, validating its practical utility in real-world usage as demonstrated by the approximately 6x increase in average daily token usage of Hunyuan 3 compared to the preview version across all channels during the pay period. Additionally, Hunyuan 3 consistently ranks among the top 3 models globally on OpenRouter based on token usage. Hunyuan 3's production version has performed well and will serve as a stepping stone towards the Hunyuan family of models, achieving state-of-the-art capabilities in the future while providing users with the cost performance efficiency that they need today. We have been integrating Hunyuan into our products, making great impact. For WorkBuddy, Hunyuan can facilitate complex agent workflows with higher task success rates and reduced time to completion. For Yuanbao, Hunyuan delivers leading execution quality and information retrieval data processing, document workflows and everyday decision-making. In games, we're leveraging Hunyuan for AI teammate creation and code review for games, including our flagship game, Peacekeeper Elite. And in Weixin, we deployed Hunyuan for powering the AI assistant in official accounts and developer tools for mini programs. At the same time, product integration is making Hunyuan better. By continuously feeding real-world product usage and domain feedback into model training, our model product co-design approach allows Hunyuan to validate model accuracy and identify and work on edge cases, enabling faster model iteration and sustained performance gains. Having now established a new system for fast model iteration and with Hunyuan 3 validating the system, we're accelerating the improvement of our model. We're scaling more powerful reinforcement learning to substantially upgrade models after pretraining is done, and we're in the process of upgrading multimodal capabilities. More importantly, we're training a larger parameter model, Hunyuan 4, which we expect to release later this year. By accelerating the technical iteration and pushing the boundaries of model intelligence, we're confident Hunyuan's capabilities will reach state-of-the-art level. We believe we will generate significant return in building a large and valuable AI native new business for Tencent. The rationale behind investing in our own foundation model is that we can achieve better unit economics, more innovative features and more exposure to the value of intelligence through co-design across our applications, our model and our compute infrastructure, especially at this early stage of AI diffusion. And moving on to the application front. Our AI office productivity workspace, WorkBuddy and coding tool CodeBuddy are achieving breakout success in terms of capability and user growth that the clear leading office productivity service in China based on monthly interactions. WorkBuddy serves as a one-stop shop workspace that orchestrates multiple agents to handle complex work from end to end. Users can remotely control WorkBuddy via Weixin and WeCom as well as PC and access to over 70,000 skills from Tencent Cloud Skill Hub. Besides the rapid user adoption of WorkBuddy, it is also achieving high retention rates and high willingness to pay among users as it directly contribute to users' productivity. It also attracts growing and more vibrant developer community by embedding Skill pay and Weixin Pay inside task flows to enable payouts for developers when their skills are called. This progress supports our view that there are substantial opportunities to be unlocked in the productivity market, including coding and existing office work scenarios. We're currently focused on investing in market education and extending our market leadership position. Over time, product economics will be attractive as enhanced premium benefits accelerate paying user growth while we can reduce token costs through agent efficiency, inference efficiency and model optimization. Given Tencent applications such as Weixin, WeCom and Tencent Meeting are already widely used by enterprises, WorkBuddy provides a new way for us to monetize our enterprise relationships. On the consumer front, we recently released a prototype of Xiaowei, which delivers an embedded and context-aware agentic AI experience within Weixin, leveraging Weixin's social graph, knowledge graph, merchant reach and payment functionality. Xiaowei is powered by the Weixin customized model, WeLM, built with a focus on user privacy, Weixin specific use cases and cost efficiency. Xiaowei can help users navigate and derive insights from Weixin's diverse content universe in a personalized and efficient manner. Xiaowei can also leverage Weixin's unique mini program ecosystem to help users discover products, make purchase decisions and place orders, laying the groundwork for an agent-to-agent transaction loop. While the prototype can technically handle advanced agentic workflows, we currently configure Xiaowei to require user intervention and multiple step confirmations as safety measures. Xiaowei will be rolled out to broader user base in a phased approach as we work on several core initiatives to elevate the user experience. These include upgrading Xiaowei's dialogue, memory and recommendation capabilities, expanding service and content integrations, scaling our AI infrastructure and upgrading our harness to support a significantly larger user base. As we upgrade Weixin for the AI era, we can do it in a cost-efficient way, and we're confident that AI will over time accelerate the growth and thus the monetization of the entire Weixin ecosystem, generating attractive return for us. Turning to Yuanbao. We are focusing on improving its capabilities and user experience, particularly in search, speech recognition and text-to-speech functionality. We're also improving its ability to address broader long-tail AI needs of consumers, including multimodal generation. Yuanbao plays an important role in the co-design flywheel as its conversational use cases generate valuable feedback to help improve our Hunyuan family of models. Over time, functionalities developed and owned by Yuanbao can become atomic capabilities for use in other Tencent products such as WorkBuddy, CodeBuddy, Weixin and QQ Browser. And with that, I pass on to James.
James Mitchell: Thank you, Martin. For the quarter, total revenue was up 11% with social networks contributing 16%, domestic games, 23%; international games, 9% marketing services 21%; and FinTech and Business Services 30%. Our gross profit was up 13%, within which VAS gross profit increased 14%, Marketing Services 21% and FinTech and Business Services 9%. Value-added service revenue was RMB 98 billion, up 8% year-on-year. within which the social network revenue was up 1% to RMB 32 billion, driven by increased revenue from app-based game item sales, partially offset by decreased revenue from long-form video subscriptions, where revenue decreased 6% year-on-year. However, our exclusive drama series, The Lead, was the most watched drama series across all video platforms in China in the second quarter. Audio subscription revenue increased 8%, driven by higher music ARPU and enriched content stemming from the inclusion of Ximalaya. We facilitated users discovering new music by enabling one-click access from Video Accounts to QQ Music. In May, we completed the acquisition of Ximalaya. By bringing Ximalaya into Tencent Group, we can enhance Tencent Music's resilience, deepen the content supply relationship between China literature and Ximalaya and provide users with new content formats, including audio books and podcasts. Domestic games revenue grew 17%, primarily driven by Delta Force, VALORANT PC, VALORANT Mobile and Roco Kingdom World. International Games revenue was down 1%, although up 4% in constant currency terms as revenue growth from Wuthering Waves and VALORANT PC was offset by revenue decreases from two Supercell games. For communications and social networks, video accounts total time spent grew over 20% in the second quarter, benefiting from enriched content supply, upgraded interactivity and the introduction of a new multivariable content ranking system. We've added content that appeals to younger users through IP partnerships with game studios, music labels and TV shows. And we've provided new revenue sharing opportunities for creators, expanding the population of creators that generate direct revenue from within the video accounts. Mini Shops GMV increased within which GMV generated from Weixin's centralized e-commerce gateway page grew significantly. For Mini Shops merchants, we introduced marketing tools such as Lucky Draws, helping them to enhance brand awareness and drive product discovery. And for Mini Shops consumers, we enhanced rewards for repeat shoppers to increase customer life cycles and thus customer lifetime value to merchants. On domestic games, Delta Force achieved lifetime high average DAU in the second quarter, driven by the Burst Fest campaign, the game's first professional esports final and a global 20 versus 20 tournament. In terms of production, the Delta Force team have integrated AI across multiple workflows, including using data agents for performance analysis and the Hunyuan 3D model for asset generation. VALORANT PC also achieved lifetime high average DAU in the second quarter, benefiting from the Skirmish Ascension mode with round-progressive weapons and the Summit map with droppable walls. The game expanded its reach by influencer collaborations, on-the-ground city events and promotions in over 10,000 Internet cafes. Among new games, Roco Kingdom World ranked fifth by average DAU and eighth by gross receipts across all mobile games released industry-wide in the second quarter, making it the highest ranked new title released year-to-date. The game has maintained a rapid content delivery cadence since launch, adding 100 creatures and expanding the map with 7 new regions. On July 9, we released Runaway Evolution. This game is adapted from Rust, a survival open world crafting game on PC that has generally ranked among the top 20 games on Steam by concurrent users for the past eight years, thanks to its unique high-risk, high-reward gameplay in which players compete to outlast the other players on their server in one week competitive sprints. Runaway Evolution seeks to tailor this gameplay for China market preferences by availability on mobile as well as PC devices and via a sandbox safe zone for new players. Among our international games, League of Legends DAU increased year-on-year in the second quarter, primarily driven by the ARAM Mayhem mode. We launched League Classic, a nostalgia mode that reengages long-time fans by recreating early era gameplay with pre-rework champions, classic runes and the original Summoner's Rift map layout. Warframe's DAU grew year-on-year and gross receipts achieved a lifetime high in this quarter, benefiting from new Wolf-themed Prime Warframe and a new storyline, Jade Shadows: Constellations. Arrows Puzzle Escape, a maze clearing game developed by Miniclip's subsidiary, Lessmore, was the most downloaded mobile game globally in the second quarter. Arrows' success demonstrates that the innovative capabilities of Miniclip's family of studios, supported by Miniclip's publishing expertise can together pioneer and break out leaders in new genres of casual games. Arrows monetizes via in-app advertising, so we report its revenue in our Marketing Services segment rather than the international gaming subsegment. Adjusted to include Arrows and other in-app advertising game revenue in the prior year and current periods, our international games year-on-year revenue growth would have been 4 percentage points faster than the disclosed figures. For Marketing Services, revenue grew 22% year-on-year to RMB 44 billion, driven by higher eCPM and impressions. Most major categories increased their marketing spending with us, including e-commerce, Internet services and local services. We upgraded AI Marketing Plus end-to-end execution capabilities to better support closed-loop Mini Shop and Mini drama advertisers. For example, AI Marketing Plus now enables Mini Shop owners to automatically select products for promotion, generate product-relevant ad creatives and then run smart bidding to buy inventory for those creatives. We significantly scaled up the parameters of our advertising AI recommendation system to capture user interest with greater granularity and thus improve ad conversion rates. Video accounts ad impressions grew rapidly year-on-year driven by higher video views and ad load, although ad loads remain well below the short video industry average. Mini Programs attracted increasing marketing spend for mini drama and Mini Game studios. For FinTech and Business Services, segment revenue was RMB 60 billion, up 9%. FinTech Services revenue grew year-on-year, driven by increases in commercial payment, wealth management and consumer loan services. For commercial payment, the number of transactions grew year-on-year, while the decline in value per transaction narrowed. For wealth management, aggregated customer assets increased year-on-year, benefiting from the popularity of automated investment strategies and thematic index funds. Within Business Services, while we're still working through capacity constraints, our cloud revenue growth rate accelerated from high teens percentage year-on-year in the first quarter to low 20s percentage in the second quarter, benefiting from AI-related demand, international expansion and increased usage and pricing for general cloud services. AI-related demand translated into increased revenue across GPU rental, Model-as-a-Service and WorkBuddy and CodeBuddy token usage. Our international cloud business expanded rapidly. Using skills developed with CodeBuddy is enabling us to conduct customer cloud migrations over to Tencent Cloud faster than we could in the past, for example, on behalf of the leading telecom company in Indonesia. And now I'll pass to John.
Shek Hon Lo: Thank you, James. For the second quarter of 2026, total revenue was RMB 204.8 billion, up 11% year-on-year. Gross profit was RMB 118.4 billion, up 13% year-on-year. Operating profit was RMB 67.3 billion, up 12% year-on-year. Interest income was RMB 4.2 billion, up 2% year-on-year. Finance costs were RMB 3 billion compared with RMB 3.9 billion in the same period last year, reflecting favorable ForEx movements and lower interest expenses due to lower average interest rates. Our share of losses of associates and joint venture was RMB 10 billion for the second quarter of 2026, primarily reflecting our share of the fair value adjustment recognized by an unlisted investee from revaluation of this issued convertible redeemable preferred shares arising from increased valuation of the investee, which was excluded from our non-IFRS profit. On a non-IFRS basis, our share of profit of associates and joint venture for this quarter was RMB 6.4 billion compared with share of profits of RMB 6.3 billion in the same period last year. Income tax expense increased by 3% year-on-year to RMB 11.7 billion. On non-IFRS financial figures. Operating profit was RMB 75.6 billion, up 9% year-on-year. Operating profit, excluding new AI products was RMB 86.1 billion, up 19% year-on-year. Net profit attributable to equity holders was RMB 68.4 billion, up 9% year-on-year. Diluted EPS was RMB 7.433, up 9% year-on-year. Moving on to gross margins for Q2. Overall gross margin was 58%, up 1 percentage point year-on-year. By segment, VAS gross margin increased by 4 percentage points year-on-year to 64%, driven by a favorable revenue mix shift towards high-margin internally developed games. Marketing Services gross margin was 57%, down 0.3 percentage points year-on-year as higher revenue supported by enhancement to our AI-driven marketing capabilities largely offset higher costs, including depreciation and operating costs associated with expanding our AI infrastructure to improve apps and content recommendation. FinTech and Business Services gross margin was 52%, broadly stable year-on-year. On operating expenses, selling -- selling and marketing expenses were RMB 11.9 billion, up 26% year-on-year due to higher marketing spend to support our games business and to drive adoption of our AI native products. R&D expenses rose by 35% year-on-year to RMB 27.2 billion, primarily reflecting higher R&D spend to support new model enhancements, Weixin AI initiatives and development of AI capabilities across our products and services. G&A, excluding R&D expenses decreased by 1% year-on-year to RMB 11.5 billion. At quarter end, we had approximately 116,000 employees, up 4% year-on-year and 1% quarter-on-quarter, mainly driven by headcount additions to our games and our technology platform, including AI-related headcount. Our second quarter non-IFRS operating margin was 36.9%, down 0.6 percentage points year-on-year. Non-IFRS operating margin, excluding new AI products was 42%, up 2.8 percentage points year-on-year. To conclude, I will highlight some key cash flow and balance sheet metrics. Operating CapEx was RMB 51.8 billion, up 190% year-on-year or 66% quarter-over-quarter as we accelerated investments in AI infrastructure to support model enhancements, WorkBuddy and CodeBuddy inference needs, AI initiatives and development of AI capabilities across our products and services as well as to meet growing external demand for our cloud services. Nonoperating CapEx was RMB 1 billion. Free cash flow was negative RMB 13.8 billion, reflecting large AI infrastructure CapEx and AI-related prepayments as well as seasonally lower games gross receipt. Excluding the prepayments for compute procurement, our free cash flow would have been RMB 37.6 billion. Net cash position was RMB 58.2 billion compared with RMB 146.9 billion as at 31st of March 2026, reflecting capital expenditure payments of RMB 59.3 billion and 2025 dividend payments of RMB 41.6 billion made during the quarter.
Wendy Huang: Thank you, John. We shall now open the floor for questions.
Wendy Huang: [Operator Instructions] The first question comes from Robin Zhu from Bernstein.
Robin Zhu: I guess if we look at your latest quarter's CapEx, RMB 53 billion, it's a step-up from the previous quarter, annualizes over RMB 200 billion, if we just multiply by 4. How should we think about the D&A cost that results from this? To what extent do you think this will be paid off from incremental revenues that come as a result of your investments in AI? Or is this essentially eating into earnings into the next few quarters? And would love to hear your thoughts on the time lags involved when it comes to the payback cycle, especially if we include some of the R&D costs incurred as well?
James Mitchell: Thank you for the question, Robin. So given the surge in demand and therefore, rental pricing for compute, we could recover the depreciation almost immediately by renting the compute out to third parties as many neo-cloud businesses are doing. And we would then achieve a decent return in an immediate time frame. However, in reality, we're playing a different game or executing a larger strategy in that we're allocating a very substantial proportion of the new compute to building our own models to state-of-the-art status and also to deploying, popularizing and bringing our own AI applications to market leadership in China. And our belief is that by providing the superior intelligence that we can achieve through state-of-the-art models through market-leading AI applications, that superior intelligence, we can then convert into superior economic returns over the longer term, for example, by selling tokens through the WorkBuddy application. So that's the path we've chosen.
Chi Ping Lau: So just to elaborate a little bit more on that, right. So I think at the time being, you can actually sort of look at the Tencent businesses and break it into two businesses. One is actually the existing core franchises, which actually generates solid growth and also with quite a bit of operating leverage. That's sort of the high-quality growth track that we have been building, and we continue with that. And then there's another new AI native business that we are actually building. And the new AI native business would involve, as James said, our own model as well as new applications that we're building and also sort of a corresponding compute infrastructure. And the financials, you should look at there is the revenue and profit in relation to our core existing business. And we do separately disclose the investment in our AI native business as an operating line. And then when you look at the CapEx, I would say the CapEx will be divided into two parts, too, right? One part is really in relation to our existing business, which you can -- just like in the past, right, you can just say, this is the free cash flow in which we generate operating cash flow and there is a CapEx in relation to that. And that part of the business is still very, very cash flow generative. And then there is another set of CapEx, which is related to the new AI native business, which is essentially a lump sum that we need to invest in order to get our compute for model training in order to prepare for inference needs and in order to also order some more for building our AI compute and AI cloud business. So that's essentially what it is. And the reason we are actually investing in all these compute is that we need that in order to essentially get the business kick started. And at the same time, when we make the investment, there's clear upside that we're seeing because our model is doing well, our new applications is doing well, and we also have a lot of demand for compute. Today, if we can actually allocate the compute towards leasing on the Tencent Cloud, we'd actually sort of generate a lot more revenue and would generate significant return from the CapEx. As a matter of fact, for the prepayment and for some of the compute orders that we have made just a couple of months ago, today, we can actually sell it at more than 30% profit compared to the price that we paid just a few months ago. But we believe if we use this compute for building our own model and building our application and then allocating the compute for rental in that order, over time, we'll build a very significant AI native business, and that will be hugely profitable as well as highly cash as well as return generated for Tencent. So that's the way we think about the business right now.
Robin Zhu: Got it. And if I may have a follow-up just on WorkBuddy. I'd love to hear your thoughts on every AI lab is essentially incentivized to develop their own harness app of some kind. And your thoughts on how the market breaks down between first and third-party harness apps, how you would like to set up WorkBuddy to compete against these first-party harnesses and whether WorkBuddy in your mind, is a piece of enterprise software that sits next to Tencent Meeting Docs? Or is this a new platform play that essentially becomes a marketplace for AI in the future?
Chi Ping Lau: Well, I think it is indeed a new platform that it's a very flexible workspace for agentic AI. The core purpose is actually that it will solve all the productivity needs of office workers and of all kinds of people who engage in their own businesses, right, one person, companies and the like. And below that, there will be harness, which actually helps to -- helps the users to make use of the capability of different models to solve the agentic problems of the users. And over time, there will be many models serving the users through WorkBuddy. There will be many skills developed over time by all kinds of different developers. And the purpose is actually solving productivity problems. And then the platform itself would make use of all kinds of different tools and models available to do that. And of course, we are the orchestrator, right? So we can actually choose the right model and choose the right skills to help users solve the problems. And we choose that to make sure that the work is done perfectly. But at the same time, it will be done also very economically, right? And Hunyuan will be one of the models that will be provided by WorkBuddy. But at the same time, if you can actually solve a lot of the user problems, then and it's quite effective, then Hunyuan would actually be one of the main models within WorkBuddy, but it will not be the only model.
Wendy Huang: We will take the next question from Kenneth Fong from UBS.
Kenneth Fong: I have a question regarding the Xiaowei development. So could management share any preliminary feedback or challenges from the testing phase of Xiaowei? And from a commercial standpoint, how should we evaluate the net monetization potential? Specifically, as agents simplify the transaction path, we worry that it may just be shifting the existing volume away from traditional user self-performed transaction in mini program over to the agents, which carry a higher computing cost without a meaningfully higher net new GTV? And furthermore, would the shortened user transaction journey in Xiaowei also lowering the high-margin ad impression inventory as well?
Chi Ping Lau: Well, I think all the risks that you said would not be relevant because we believe when AI enable the Weixin ecosystem to be more intelligent and it can actually sort of help users to execute transactions, explore content and manage their daily life with a lot of AI, right? Then the AI -- the Weixin ecosystem, which is already very rich and powerful will become even more useful to the users, right? So if you imagine the time when QQ was a communication and social tool in the PC stage. And then when we get into the mobile age, then Weixin appears and Weixin essentially the ecosystem magnified QQ's value by more than 10x, right, because it's enabled in the AI -- in the mobile age and it becomes mobile first. So when we look at AI, we believe there's another huge opportunity for the Weixin ecosystem to be first enabled by AI. And over time, it will be AI-first application and ecosystem. And when that happens, users would have a lot of great experiences like right now, you actually sort of have to type and you have to sort of navigate through clicks. In the future if you just tell Xiaowei one instruction and then Xiaowei can go off and help you execute the transaction and execute your instruction, and that would be an incredible experience for the users. It would also be an incredible empowerment for the entire ecosystem. So we believe if we can deliver that experience, if we can control the cost of that delivery. And if you look at the design of WeLM, it's actually for privacy, for cost efficiency and for making sure that it can execute within the Weixin environment, all the needs of the users, right? And if we can do that, then we can really empower Weixin for the AI age under controllable cost. And when that happens, Weixin's ecosystem would expand and that would translate into a lot of value just based on the current monetization mechanisms within Weixin. And I think that's the future that we're seeing. And with the launch of the prototype, we grow more and more confident about that will be happening.
Kenneth Fong: I have a follow-up question on the AI cloud with domestic API token prices for very rapid commoditization. So -- and also China cloud market remains structurally price sensitive. So how should we think about the margin profile of Tencent AI Cloud currently compared to, say, IaaS and PaaS offering? And as this gradually scale up as AI adoption scale, so how should we also think about the margin progression going forward?
James Mitchell: It is true that domestic token prices are low, but the domestic token manufacturing costs are also extremely low, much lower than widely perceived or externally estimated. So the token business, it can be positive gross margin at these low token prices because the cost is low. And if you look at the gross margin for the paying users of WorkBuddy or you look at the gross margin for our Model-as-a Service, then the gross margins today are already comparable to the gross margins for Tencent Cloud overall. Of course, WorkBuddy in aggregate has a lower gross margin because there's a proportion of free users whom we're subsidizing to drive market share and market growth. But on the paying users, we're generating a pretty good gross margin right now. And on your broader concern, it is true also that the China cloud market is price competitive, but that environment has changed greatly in the last several months as the input costs, particularly for memory, have gone up, -- so we have been increasing the prices we charge to our customers. We have increased prices across the board in May for Tencent Cloud. And beyond those headline price increases, we've also been more substantially reducing discounts. So the overall pricing environment in cloud in China is not as difficult as it's been in the past.
Wendy Huang: We will take the next question from Ronald Keung from Goldman Sachs.
Ronald Keung: So two questions. I think first on the Hunyuan model. Just want to hear after the progress of 3, which is on cost efficiency, I would say and very good agents, where will Hunyuan 4 differentiate itself as we look into a, let's say, 3 trillion parameter size class looking more crowded in the next few months? So which category are we looking or which segment or differentiation are we thinking for? And then a second question is on the CapEx and the focus on our AI initiatives. But looking at some of the U.S. peers where there has been a shift in strategy on the hyperscaler business. I just want to hear what stage or time line that we think we may focus more on cloud as a potential high ROI business that is worth prioritizing more CapEx on and some similarities and differences that we see Tencent Cloud versus what how U.S. peers have shifted that focus more from applications to cloud for some of our peers, So two questions.
Chi Ping Lau: If you look at the Hunyuan 3, right, Hunyuan 3 is a very small model even in today's terms, but it's actually very widely used, right? So I think there are a number of characteristics of Hunyuan 3, which is it actually has the capability of beating -- matching or beating much larger models. That's one. And two is it's actually focused on use cases rather than just benchmark beating. And as a result, right, in real life, it has become much more useful than a lot of models of the same size or even bigger size. We believe that's a principle that we will be applying to Hunyuan 4 as well. So when Hunyuan 4 comes out, it will be a bigger model, and it will be able to beat models of bigger size. And it would also be extremely useful and more useful than Hunyuan 3. And we believe that would actually take us into the next stage of being able to provide much better intelligence to a lot of our users. And bear in mind that we also have products which are codesigning with the model. So when Hunyuan 4 comes around, the products that would be using Hunyuan 4 would actually become even more powerful and even more useful than what they are today, and that would actually provide a very significant lift for the products that is powering. So I think that's sort of the path. And Hunyuan 4 is only sort of another stop, right, and that will be sort of upgrading to Hunyuan 5. So as we continue to progress, we will be approaching SOTA. And at some point in time, we'll definitely sort of be able to reach SOTA. And once we are there, we would also have a lot of models of different sizes that will be able to solve different kinds of user problems at a different level of model and cost efficiency. And at the same time, we have multiple models that can be used for co-design with our different products, and that would help us to make the products feature-rich and help to make the models -- the products powerful as well as the speed of execution will be fast. So I think that's what we envision Hunyuan 4 and then subsequently Hunyuan 5 to be like.
James Mitchell: And in terms of your second question about allocating CapEx between different use cases, including Tencent Cloud. So the immediate primary use case for the CapEx is for training bigger and better Hunyuan models in the coming months, as Martin discussed. But an important secondary use case is providing inference for the use of Hunyuan models as well as DeepSeek and other models behind WorkBuddy. And so the intention of that WorkBuddy initiative is primarily to drive adoption of what we think is a strategically important application with critical feedback loops back to our model and our broader ecosystem. But it also has the happy effect of generating revenue upfront. Now from an accounting perspective, the majority of the WorkBuddy spending by users is on subscriptions. And so similar to games in some of our other businesses, there's a lengthy time lag between the cash receipts coming to us from the users and those cash receipts translating into reported revenue. But we are seeing a substantial ramp in the cash receipts today. And that will translate into reported revenue growth for Tencent Cloud as we move through the year. And then towards the end of the year and into next year, we'll also have sufficient GPU ASIC capacity to step up in terms of Tencent Cloud, renting out bare metal GPU or providing Model-as-a-Service. But within those opportunities, renting out GPU, Model-as-a-Service and then token production for WorkBuddy, we think that it is token production for WorkBuddy that carries the most enduring economic value to us, and that's why we're prioritizing it today.
Wendy Huang: We will take the next question from Alicia Yap from Citigroup.
Alicis a Yap: On the solid results. First question is on the Xiaowei. Could management elaborate on your comment on the agent-to-agent transaction loop? Will this concept lead to the long-term vision for a fully autonomous agent ecosystem within the Weixin? And then management also highlight that you will explore the on-device inference for Xiaowei. So what are the challenges and the benefits of this approach? And then is this on-device inference approach another reason why the proprietary WeLM model is more suitable empowering the Xiaowei rather than the external model? And then a quick follow-up is on your marketing service revenue. So this quarter, the growth rate accelerated to 22%. Should we expect this ongoing upgrade of the ad tech and also this automated campaign to further support this growth momentum? So any further future benefit that you would anticipate from the deeper integrations into your Hunyuan 3 model?
Chi Ping Lau: Yes. So on the agent-to-agent transaction, I think we are envisioning a future in which a lot of users would be executing their instructions and over time transactions via Xiaowei and via agents, right? And in the past, if you think about the Weixin ecosystem is users interacting with content, interacting with mini programs themselves. And in the future, if they can actually send a complex instruction to an agent and the agent can actually sort of start helping the user to execute transactions. And a lot of the mini programs, a lot of the merchants would actually also have agents, which over time can interact with the agent of the users. And longer term, there will be even user -- each user has got an agent, and they can actually interact with each other to execute a transaction. So I think that essentially is what's possible for the future, and we're building the architecture for making that possible on a step-by-step basis. And in terms of on-device inference, I think it would, number one, be happening maybe step by step, and it will be only over the long run that most of the inference will be happening on device, right? But I think at some point in time, it not hard to imagine some kind of inference will be actually happening on device and some inference will be happening in the cloud. And over time, as the on-device compute becomes more and more powerful and as the model becomes more and more efficient, you have more inference happening on people's devices. And I think that would be going back to the normal state of the computer industry. If you think about the compute industry as well as the smartphone industry, right, most of the compute, right, which is CPU actually happens on device. And the cloud actually sort of only is responsible for a small part of the compute. But in this initial phase of AI infrastructure, most of the compute because it has to be sort of very powerful, right? And the problem of getting enough compute on device, getting it cheap enough and also getting it power efficient enough has not happened yet. So that's why everything happens on the cloud. But there will be a time in which more and more GPU capability will be put into everybody's phone and computer. And when that happens, then more and more inference will be happening on the device, and it will be sort of going back to the time when it's actually the software, it's actually the model that becomes much more important and the return for running models and the return on running applications will be higher because the compute CapEx will be not just borne by the model company, but it will be borne across the ecosystem. And I think that would definitely happen at some point in time when we're building and preparing for that.
James Mitchell: And on your marketing services question, our advertising revenue growth has ticked up and ticked down in the past, and it will continue to tick up and tick down in the future. I wouldn't sort of straight line extrapolate anything. And there's a number of reasons. One is that the sort of obverse of the comments I made about in-app advertising games being a drag on the International Games segment revenue growth versus where it would otherwise have been is that they did contribute about 2 percentage points to the advertising segment revenue growth this quarter. And these in-app advertising games are sort of new product for Tencent, to some extent, a new product for the world. And so we don't have the same degree of clarity on what the growth trajectory will be for the in-app advertising game contribution as for our sort of conventional marketing services revenue. In addition, the China consumer and therefore, advertising market remains choppy and there are some economic or consumption headwinds that may have an impact on advertising trends. That said, we have been outperforming the overall China advertising market, and we're confident we'll continue to do so by a substantial margin, given the upside to us from deploying AI ad targeting, given the fact that engagement, especially for our key video accounts inventory is increasing at a good rate. And given we're early in the evolution towards more closed-loop advertising that drives much higher ad pricing.
Wendy Huang: We will take the next question from Alex Liu from Bank of America.
Alex Liu: I have only one question. So we noted that Tencent has recently increased the buyback. The buyback activity started from May, while at the same time, the CapEx has been accelerated meaningfully as well. So we understand it's still in the relatively early stage in AI investment cycle. But with that in mind, how should investors think about Tencent's capital allocation priority into the next 12 to 24 months?
James Mitchell: I think that -- I know that our capital allocation will be dynamic and reflective of the environment that we see. And so if we identify that there's superior returns from capital expenditure from increasing our compute and then using that compute to build the model, renting out that compute for WorkBuddy tokens, renting out that compute for Model-as-a-Service, then we'll steer more cash toward the capital expenditures than we had in the past and therefore, potentially less cash towards buybacks, but it will be a dynamic situation.
Chi Ping Lau: And the other thing I do want to stress is that when we look at the CapEx that we allocate for building the AI native business, it is more of sort of a lump sum that we're going to be investing this year and next year. And then I think one should not assume that it will be sort of every year because sort of the model building part is more of a fixed cost that you actually sort of you have to get enough compute, but it will not be sort of every year, you have to invest more. And in terms of the inference compute, yes, we need to have enough so that we can generate the tokens and we can sort of build a compute business, right? And -- but we will only keep on investing if it generates a great return, right? If not, then this is actually sort of the amount that we're going to be investing and then sort of the additional investment in CapEx would actually be tied to sort of what kind of returns that we'll be generating from that business. And so in order to pay for this lump sum, then it should not be just sort of measured against our operating cash flow. It should be measured against how much cash we have on our balance sheet, how much investment portfolio we have on our balance sheet and then the operating cash flow and then a prudent level of debt capacity. So all these would come into play in terms of paying for this initial part of compute investment.
Wendy Huang: Thanks, Martin, for your supplement on CapEx and return consideration. We will move on to the next question from Alex C. Yao from JPMorgan.
Alex Yao: My first question is on Hunyuan flagship strategy. The Hunyuan 3 competes on cost efficiency rather than raw capacity -- capability. If you succeeded in building a truly frontier level model, which should be larger and more expensive to run, what specific business value would that create that the current Hunyuan 3 cannot deliver today, whether a more capable Weixin agent or stronger advertising performance or enterprise customers? How does that opportunity justify a major increase in training spend over the next 12 months?
Chi Ping Lau: Well, let's be very clear. So Weixin model and the strategy -- the position is very different, right? Weixin agent doesn't really require or depend on Hunyuan's sort of new, reach SOTA status. Weixin's design, as we have said a few times, is actually centered around user privacy and focusing on solving all the necessary interactions and agent needs within the Weixin environment and also for cost efficiency. So that's its positioning. Now the SOTA status would actually allow us to be able to build a very significant token business -- and at the same time, it will also empower WorkBuddy to be able to complete even more challenging and more value-added services and operations for the users. And one of the things that we actually sort of focused on WorkBuddy is actually not just saying, oh, it's an enterprise software and we would just do all the things that people can do today. It's actually sort of constantly looking for value-added use cases so that we can really deliver additional value and return for the users. And in some cases, even help the users to make more money, right? And if we can do that, then there will be a lot of business models that we can unlock, right? So I think that's what we can also achieve with SOTA model. And at the same time, once we reach SOTA, we can actually start creating a lot of the other models, which can perform specific tasks for users at different levels of cost efficiency curve still at the frontier curve, right? And that would actually help us to cater to the many different needs of intelligence for users. And that at each level, the cost will be different, but we will be able to generate a margin because we control the model, we control the inference cost, we control the compute. And that's, I think, what we envision for our future generations of models to be able to achieve.
Alex Yao: My follow-up question is on AI product economics. The new AI product drag rose from roughly RMB 8.8 billion in first quarter to about RMB 10.5 billion this quarter. Can you walk us through how you manage that investment? Do you run these products to spending envelope or to a return threshold or to a strategic position? And what signals, whether usage, revenue traction or unit economics would cause you to step investment up further or begin shifting product from investment mode to harvesting mode?
Chi Ping Lau: Well, at this stage, it's actually very dynamic. And I think we would be investing prudently until the point that we actually see breakout opportunity, then we may step up the investment. So I think that is essentially the way we look at it. So it will be a certain percentage of our profit. But if clearly, we see that if we step on the pedal, it would actually generate a lot of returns, then we may step the pedal. But the overall, we do believe that it is a business that has to run for a long time. So we will be investing for the long run. And over time, we believe the economics would actually start coming in. And at some point in time, it would actually be able to turn into profit. And I think more importantly is that today, if we just switch the model to just renting out compute, it will be actually not loss-making, it will be profitable. So I think we always have that fallback option. So I think that's sort of something that we -- that's why we feel comfortable.
James Mitchell: It's also the case that we dynamically reprioritize the spend within the budget or within the envelope. And so if you look at where the RMB 8 billion in the first quarter flowed in terms of user acquisition spending and so forth and which products it supported versus where the RMB 10.5 billion in the second quarter flowed, there was actually a very big change because we identified that WorkBuddy was breaking out. And therefore, we aggressively prioritized WorkBuddy while deprioritizing some of the other products in that new AI product portfolio.
Chi Ping Lau: Yes. And you can assume there is an envelope at the back of our mind.
Wendy Huang: We will take the last question from Gary from Morgan Stanley.
Gary Yu: I have one follow-up on the AI investment. I understand the priority on model training, WorkBuddy, and maybe cloud. Where does Xiaowei fit in terms of inferencing capacity required to support Xiaowei when it's launched? So that's my first question. My second question is on management view about timing and visibility of monetization and ROIC for these new AI initiatives. And particularly, should we expect close to negligible earnings growth in the near term -- and then when should we expect the operating profit, including AI investment to grow even faster than excluding the AI investment?
Chi Ping Lau: I think for Xiaowei, the envelope of investment on the cost side will be less than what we actually invested in Yuanbao on an ongoing basis in the past, let's say, year. So I think that is the way we think about it. So the cost would be quite manageable. But as the experience keeps getting better and better, the return will actually start flowing in and it would actually sort of outweigh that investment pretty quickly. And in terms of guidance, I don't think we are in the business of actually providing sort of that specific guidance, right? I think we have talked a lot about how we think about the business and how we think about there is an envelope of investment that we will be adhering to. It will be sort of kind of disciplined in the same way as Tencent has always managed our business. But then if we clearly see great opportunities to build a very significant and profitable business for the future, then we would actually make the investment. And we also sort of have the comfort that if we actually just move more compute into the compute, right, we can generate revenue, profit and return very quickly. So that actually sort of is a fallback decision any time that we choose to do that.
Wendy Huang: Thank you. We are now concluding the webinar. Thank you all for joining our results today. If you wish to check out our press release and other financial information, please visit the IR section of our company website at www.tencent.com. The replay of this webinar will also be available soon. Thank you, and see you next quarter.