Tencent (00700) Q3 Earnings Call: WorkBuddy Revenue Grows Over 5x, Cloud Services Growth Jumps to 35%, AI Dividends Realized

date
06:34 19/11/2026
avatar
GMT Eight
WorkBuddy's daily active users exceeded 10 million, with annualized revenue over the past seven days surpassing US$500 million, and Q3 revenue more than 5 times that of Q2;
Tencent (00700) delivered a quarterly report card reshaped by AI: cloud business growth jumped from 21% to 35%, WorkBuddy surpassed 10 million DAU with annualized revenue exceeding US$500 million, and management set the tone"AI allows existing businesses to operate with fewer people, and the freed-up people go build new businesses." On Wednesday evening, Tencent's Q3 2026 earnings call was held. Founder and CEO Pony Ma, President Martin Lau, Senior Executive Vice President and Chief Strategy Officer James Mitchell, and CFO John Lo participated in the call. Overall, management set the tone on this call: AI investment is generating visible returns, but compute constraints are creating real pressure on next-generation model training. Tencent said WorkBuddy's daily active users (DAU) exceed 10 million, consuming 100 trillion tokens per day, with revenue over the most recent seven days annualizing to more than US$500 millionmore than double that of Tencent Meeting. Q3 revenue was more than 5x that of Q2, and 60% of China's top 100 enterprises have already become its customers. According to Blue Whale News, regarding the impact of compute-related restrictions on Tencent's AI business, Tencent President Martin Lau said that both leased and self-built compute face constraints, and model training will inevitably be affected; in the short term, it may take longer to launch the next version of the Hunyuan model. WorkBuddy: Seven-day annualized revenue exceeds US$500 million, enterprise momentum fierce On the AI monetization front, the number that drew the most market attention on this call came from WorkBuddy. Tencent President Martin Lau disclosed that WorkBuddy's annualized revenue over the past seven days has exceeded US$500 million, more than double Tencent Meeting's revenue. Q3 standalone revenue grew more than 5x quarter-over-quarter. On the user side, WorkBuddy's DAU surpassed 10 million, consuming more than 100 trillion tokens per day. Lau said it "has become the No. 1 AI application in China by large language model token consumption." Penetration in the enterprise market is equally impressive. Since launching the enterprise edition in June this year, 60% of China's top 100 enterprises have become WorkBuddy customers. Lau said, "This strong early adoption is encouraging, and we believe this is just the tip of the iceberg." On the business model, Lau compared WorkBuddy's monetization logic to the evolution path of China's online gamesAI services naturally avoid software piracy, and the hybrid subscription + token usage payment model allows users to "enter the paid experience at a low threshold and increase spending as they derive more value." He also likened WorkBuddy's growth driver to the item-based payment model of online games: "When consumption can scale up along the demand curve, it unlocks a much larger revenue pool." Regarding market concerns that the Flash lightweight model lowers overall token pricing, Lau responded that the Flash model is essentially a customer acquisition tool, not a profit killer: "The Flash model is a great way to acquire customers, while more powerful models monetize more by delivering more value to users... This perfectly fits WorkBuddy's logic of monetizing at every point along the demand curve." Chip restrictions: Inference unaffected, but training under pressure This was another core topic of the call. Facing analysts' follow-up questions, Lau directly addressed the impact of compute chip-related restrictions on Tencent's AI business. According to Blue Whale News, Tencent President Martin Lau said that both leased and self-built compute face constraints, and model training will inevitably be affected; in the short term, it may take longer to launch the next version of the Hunyuan model. Lau noted that in the model training and inference process, Tencent has found that computational efficiency is in fact improving rapidly year by year, and the approach previously adopted in Hy4 model training was already highly efficient. Over the long term, the domestic supply chain will gradually participate in the model inference segment; the domestic industry is currently developing very fast, and this localization process is expected to accelerate further. Xiaowei: Will be as powerful as Muse in the future WeChat's embedded AI assistant Xiaowei has reached 30 million DAU. Lau said the next step is to continue expanding the testing scale and connect more than 30 touchpoints, including Moments, Official Accounts, and Channels. On cost control, he gave a persuasive figure: even if Xiaowei is opened to all 1.4 billion WeChat users, its quarterly operating cost would still be lower than Yuanbao's quarterly cost at the beginning of this year, because Xiaowei uses a lightweight WLM model custom-built for low cost and high efficiency. Lau also said that because computational efficiency is improving significantly, each new version of Xiaowei will see its capabilities multiply upon release. Its feature completion requires more ecosystem connections, which takes time. But at some point, it will become as powerful as Muse, which will also greatly enhance the value within the WeChat ecosystem, and revenue brought by the WeChat ecosystem will also increase substantially. On the long-term monetization logic, Lau described Xiaowei's "multiplier effect" on WeChat's existing revenue system: advertising and transaction revenue are near-term opportunities, while agent-to-agent service fees and merchant revenue sharing are medium-term paths. He expects that "adding 10% to 20% on top of WeChat's existing revenue would be a very considerable number," and judged that "we will see this very clearly in the next six to twelve months." Cloud services: Growth jumps to 35%, AI dividends realized Tencent CFO John Lo disclosed that Q3 cloud service revenue growth accelerated to 35% year-on-year from 21% in the previous quarter, mainly driven by AI-related demand, including WorkBuddy, model-as-a-service, and GPU leasing. Among these, model-as-a-service revenue grew more than threefold quarter-over-quarter; GPU leasing revenue grew more than 30% year-on-year. Management expects this higher growth rate to persist throughout 2025 (the original report refers to full-year 2026). On capital expenditure, Q3 operating capital expenditure was RMB63.6 billion, up 257% year-on-year, mainly for AI infrastructure construction. As a result, free cash flow was negative RMB69.1 billion. Excluding compute prepayments, free cash flow would be positive RMB16.9 billion. AI reshapes organizational structure, still at a "very early stage" On the strategic level, both Pony Ma and Martin Lau mentioned the ongoing organizational restructuringusing AI to reshape business processes and free up manpower to migrate toward new business directions. Pony Ma acknowledged that it is still at a "very early stage," but the core logic is clear: "AI can really be applied across various businesses to improve operational efficiency. But to do that, you actually need to undertake fairly large-scale organizational restructuring, and you also need to redesign many workflows. If we can do that, existing businesses can operate with fewer people." "At the same time, AI enables small teams to leverage AI to explore new ideas, build new businesses, and create new growth engines. If we can make existing businesses run with fewer people and redeploy those people to build new growth drivers, that is the ideal state." He cited the Hunyuan model team and the WorkBuddy team as examples: "The model team was rebuilt completely from scratch and within a year pushed our models to China's top levelthat's one example. The WorkBuddy team had almost nothing six or seven months ago, but they built such an exciting businessthat's another example. We hope to see more and more such cases in the future." Lau said: "If we can operate existing businesses with fewer people and redeploy those people to new growth engines, that is the ideal state." He said the Hunyuan model team was built from scratch and pushed the model to the domestic forefront within a year, while the WorkBuddy team went from "nobodies" to building an "exciting business" in six or seven months. Below is the transcript of the earnings call: Wendy Huang, Head of Tencent Investor Relations: Good morning and good evening, everyone. Thank you for waiting. Welcome to Tencent Holdings Limited's Q3 2026 results presentation. Currently, all participants are in listen-only mode. After management's presentation, we will conduct a Q&A session. Please note that today's meeting will be recorded. Before we begin, we remind you that this meeting contains forward-looking statements, which are subject to various risks and uncertainties and may not materialize in the future for various reasons. Information about overall market conditions comes from multiple sources outside Tencent. This presentation also contains certain unaudited non-IFRS financial measures, which should be considered in addition to, not as a substitute for, the group's financial results prepared under IFRS. For a detailed discussion of risk factors and non-IFRS measures, please refer to the disclosure documents in the Investor Relations section of the company's website. Management participating in tonight's call includes: Chairman and CEO Pony Ma, who will begin with brief remarks; President Martin Lau, who will discuss corporate strategy; Chief Strategy Officer James Mitchell, who will discuss business; and CFO John Lo, who will discuss financial performance. Before we enter the Q&A session, let me first invite Pony Ma to speak. Pony Ma, Co-Founder, Chairman and CEO of Tencent: Good evening, everyone, and thank you for joining today's call. Recently, we have made important progress in models and applications, accelerating Tencent's pace toward a new AI era. Hunyuan Hy4 has reached the level of China's leading AI models, while Hy4 Flash delivers excellent cost-performance in real-world application scenarios at a lower cost. We believe that measured by total large language model tokens consumed by users per day, WorkBuddy has grown into China's most-used AI application. Driven by WorkBuddy and by our allocation of more compute to model-as-a-service and GPU leasing, Tencent's cloud service revenue growth has accelerated. Core businesses remain resilient and vibrant, with several evergreen games hitting record-high DAU, and Channels user engagement rising rapidly, especially among younger users. Channels has further strengthened its appeal to younger users by enriching content supply. Looking at Q3 financials, total revenue was RMB207 billion, up 8% year-on-year; gross profit was RMB116 billion, up 7% year-on-year. Excluding new AI products, gross profit was RMB118 billion, up 8% year-on-year. Non-IFRS operating profit was RMB68 billion, down 6% year-on-year; excluding new AI products, non-IFRS operating profit was RMB83 billion, up 8% year-on-year. Non-IFRS net profit attributable to equity holders was RMB67 billion, down 6% year-on-year. In communications and social networking services, combined MAU of Weixin and WeChat grew both year-on-year and quarter-on-quarter, reaching 1.4 billion. In gaming, Delta Force's peak DAU in October exceeded 58 million, demonstrating our ability to continue expanding the user base of new games years after launch. In productivity tools, measured by large language model token usage, WorkBuddy has become China's most-used AI product. On foundation models, we released the official version of Hunyuan 4 earlier this month. At launch, it ranked No. 1 among Chinese models on the Agent Arena leaderboard, reaching China's leading level. Next, let me invite Martin Lau to discuss corporate strategy. Martin Lau, President of Tencent: Thank you, Pony. Good morning and good evening, everyone. Over the past three months, we have made breakthrough progress in AI-native businesses. Today, I would like to discuss some important progress in intelligence, applications, and commercialization. First, intelligencefoundation models. We believe the latest generation Hunyuan Hy4 has reached a leading level among Chinese foundation models and offers excellent practical value in real productivity scenarios. Hy4 leads among models of comparable scale, and its capabilities can even reach those of larger models. For example, at launch, Hy4 ranked No. 1 among Chinese models on the Agent Arena leaderboard and achieved high scores on numerous benchmarks without being specifically optimized for those tests, unlike many other models. Our team significantly scaled up Hy4's pre-training and post-training and added vision and computer operation capabilities. As a result, compared with previous models, Hy4 has significantly improved in instruction following, long-context understanding, long-horizon tasks, and deep research. To further reduce inference costs and drive model adoption in more real-world scenarios, we also released the customized lightweight model Hy4 Flash. Hy4 Flash is priced at only one-twentieth of Hy4, and actual usage has far exceeded expectations; among all models released this year, its daily token consumption ranks No. 1 on the OpenRouter platform. We are pleased with Hy4's capabilities and rapid iteration progress, while also believing that Tencent's intelligence process is still at an early stage. By deploying high-performance models into core products such as WorkBuddy, we can obtain high-value user feedback, thereby driving continuous co-design and iteration between models and products, especially forming a virtuous cycle in real-world application scenarios. Recently, we also unified multimodal R&D under the Hunyuan team. In 3D generation, we are industry-leading, while also continuously improving image generation capabilities and developing video generation capabilities. Next, applications. With WorkBuddy, we have also achieved breakthrough progress in productivity AI. WorkBuddy's user growth is very rapid, with DAU now exceeding 10 million and daily token consumption exceeding 100 trillion. Measured by large language model token usage, we believe WorkBuddy has become China's No. 1 AI application. In productivity AI, WorkBuddy is also far ahead. Based on external services and internal data, its gap with competitors has further widened in recent months. In the past, we had been cautious about the pace of development of China's enterprise software market, but now we believe that the addition of AI capabilities will drive rapid and substantial growth in productivity software spending, and WorkBuddy will be an important driver and beneficiary of this trend. There are two main reasons. First, AI services can avoid software piracy. Second, WorkBuddy covers a large number of valuable application scenarios within enterprises, such as document preparation, data analysis, project management, research, and software development, which makes the whole suite worth paying for. In addition, the subscription fee plus token usage business model allows customers to obtain paid services at a low fixed entry price and increase spending further as they derive more value. Over the past seven days, WorkBuddy's annualized total revenue has exceeded US$500 million, more than double that of Tencent Meeting, which launched seven years ago. To some extent, this development path is similar to the evolution of China's online game industry: cloud connectivity can curb piracy, while an item-based monetization model can unlock greater revenue space as user demand and willingness to spend increase. We have already seen that when on-site engineers show enterprises specific use cases and their value, enterprises significantly increase spending. Therefore, we are working with partners to provide frontline deployment support for WorkBuddy customers. Since launching the enterprise edition in June this year, 60% of China's top 100 enterprises have become WorkBuddy customers. Early adoption is encouraging, and we believe this is just the beginning of the market potential. At present, we are still investing resources to drive WorkBuddy adoption, including temporarily providing free services to some user groups such as school teachers. We believe that as the proportion of paying users rises, token spending per paying user increases, and token costs decline through inference optimization, model customization, and compute efficiency improvements, WorkBuddy's unit economics will gradually improve. By integrating WorkBuddy with Tencent and external ecosystems, we believe we can also improve user retention, user lifetime value, and overall unit economics. If WorkBuddy drives the depth of AI usage, then Xiaowei, the personal assistant embedded in WeChat, will drive the breadth of AI usage, making AI accessible to every internet user in China. Since testing began in June, we have continued to expand Xiaowei's test user base, and DAU has now reached 30 million. We have added more Xiaowei entry points in scenarios such as Moments, Official Accounts, and Channels, enabling it to cover a broader range of use cases, provide more practical features, and increase interactions per DAU. We are connecting large mini programs and merchants and integrating with some stores to support transactions. Xiaowei is powered by a purpose-built vision-language model (VLM) customized for cost efficiency and optimized around the WeChat ecosystem. Xiaowei's inference cost is very low and will continue to decline in the future. In fact, even if Xiaowei is opened to all 1.4 billion WeChat users, we expect its quarterly cost to remain lower than the quarterly cost incurred by Yuanbao earlier this year. As compute efficiency improves year by year, Xiaowei's capabilities will also be significantly enhanced. Over the long term, some inference tasks can be completed on user devices, further reducing costs. Leveraging the existing merchant network in WeChat Pay and mini program services, Xiaowei can achieve deep integration with merchants and service providers at low cost, forming synergy rather than competition with them, thereby establishing a mutually beneficial cooperation model. On commercialization, we expect Xiaowei to become an amplifier of WeChat's existing revenue sources, further enhancing the scale and growth rate of these revenues. Advertising and transaction-related revenue are near-term opportunities, and new revenue sources can be added in the future, such as charging merchants and service providers for agent-to-agent collaboration services. We are currently at an early stage of accelerating compute additions, while also opening some compute to external paying customers through Tencent Cloud. Therefore, the cloud services business has begun to generate financial returns from AI investment, with initial revenue sources including WorkBuddy, model-as-a-service, and GPU leasing. In Q3, cloud service revenue growth accelerated to 35% year-on-year, up from 21% in Q2, while full-year 2025 year-on-year growth was in the teens. In Q3, WorkBuddy revenue reached more than five times that of the previous quarter, mainly driven by an increase in paying users and initial adoption by enterprise customers. Model-as-a-service revenue grew more than twofold quarter-on-quarter, mainly due to improved inference efficiency and our strengthened enterprise operational controls and security safeguards, enabling enterprises to better oversee AI deployments. GPU leasing service revenue grew more than 30% quarter-on-quarter, mainly driven by enterprise inference demand and improved compute scheduling efficiency. Next, let me invite James to discuss the business. James Mitchell, Chief Strategy Officer of Tencent: Thank you, Martin. In this quarter, total revenue grew 8% year-on-year. By segment contribution to revenue growth, social networks contributed about 15%, domestic games 23%, international games 9%, marketing services 21%, and fintech and business services (FBS) 31%. Gross profit grew 7% year-on-year, with VAS gross profit up 7%, marketing services gross profit up 10%, and fintech and business services gross profit up 8%. VAS revenue was RMB98 billion, up 3% year-on-year. Among this, social networks revenue declined 1% year-on-year to RMB32 billion, mainly due to a lower number of top content launches and a decline in long-form video subscription revenue; meanwhile, in-app game item sales and audio subscription revenue grew. Domestic games revenue grew 11% year-on-year, mainly driven by Delta Force, Valorant Mobile, and Honor of Kings. International games revenue declined 9% year-on-year, or 4% at constant currency, mainly due to a high base last year for certain Supercell games and the launch of Dying Light: The Beast in Q3 last year. In WeChat's communications and social networking services, our short drama entry achieved rapid user growth, offering users free-to-watch micro short dramas. This content benefits from WeChat's social recommendations and from high-quality content including adaptations of Yuewen novels. Developers can now use WorkBuddy to automatically complete mini program development and publishing, lowering the development threshold and shortening development cycles. On Channels, we improved younger user engagement through deeper content cooperation with internal games, music, and TV services. We also launched AI tools for creators that can convert Official Account articles into videos, automatically completing storyboard design, script writing, and media content generation. Total time spent on Channels grew more than 20% year-on-year, and we believe there is still further growth space. Notably, our time spent per DAU remains less than half that of the industry-leading platform. On domestic games, Delta Force's DAU grew more than 50% year-on-year this quarter, reaching a record high since launch. We deepened the game's core extraction gameplay with new maps and added a brand-new 5v5 tactical shooter mode. Valorant Mobile's DAU and gross billings both hit record highs, benefiting from popular floral-themed and Chinese folk-themed weapons. Honor of Kings' average DAU also achieved year-on-year growth, mainly benefiting from the new anomaly system that reinterprets classic heroes and a faster-paced arcade mode. Game billings also grew, mainly driven by successful IP collaborations and dark fantasy-themed skins. In September, we launched Honor of Kings: King's Chess, a hero card battle game set in the Honor of Kings universe, available on mobile and PC. Since launch, the game's average DAU has ranked among the top ten mobile games industry-wide. Martin Lau, President of Tencent: On international games, Valorant revenue grew year-on-year in Q3, mainly driven by increased player engagement and content from the 2026 global championship series. At the Valorant global championship finals held in Shanghai, we hosted a pop culture festival around gaming, fashion, and outdoor experiences. We also partnered with the developer of Steal a Brainrot to bring the Brainrot IP into PUBG Mobile's World of Wonder creative workshop. The resulting map, Brainrot Royale, drove the largest increase in DAU for the creative workshop mode since launch, with 30% of PUBG Mobile DAU now participating in this mode. On October 29, we launched the mobile action RPG Monster Hunter: Traveler. The game features impressive monster design, the series' iconic combat experience, and a brand-new adventure mode. In its first week, the game ranked among the top ten mobile games globally by downloads. On marketing services, despite overall industry pressure, revenue grew 19% year-on-year to RMB43 billion, with both impressions and ad pricing increasing. Among this, the consumer goods and local life services categories contributed notably to revenue growth. We launched AI advertising solutions for content publishers. For example, we help micro short drama providers automatically insert ads of appropriate duration at the right time, improving monetization efficiency while minimizing disruption to the user viewing experience. Currently, for many short drama publishers, these AI solutions already support about 80% of advertising monetization. Channels ad impressions grew more than 50% year-on-year, mainly driven by increased video views and ad load. We now place ads based on a content ranking system, improving ad relevance and monetization efficiency. The short drama entry contributed importantly to mini program marketing services revenue growth. Fintech and business services revenue was RMB64 billion, up 9% year-on-year. Fintech services revenue achieved low single-digit year-on-year growth, mainly driven by increased revenue from consumer loans and wealth management services. Transaction volume grew slightly year-on-year, with transaction count increasing, but overall weak consumption led to a decline in average transaction value, partly offsetting the impact of transaction volume growth. On business services, we launched TenPayGo, a payment app for foreign tourists visiting China, integrating overseas bank card and digital wallet support, transit QR codes, tax refunds, and AI translation services. James Mitchell, Chief Strategy Officer of Tencent: As Martin just introduced, driven by growing demand for AI-related services and increased overall cloud service demand, our cloud service revenue growth accelerated to 35% year-on-year. We have integrated WorkBuddy into WeCom, enabling agents to use authorized workplace information to complete tasks, share results within the organization, and turn organizational knowledge into reusable capabilities. Next, let me invite John Lo to discuss the financials. John Lo, CFO and Senior Vice President of Tencent: Thank you, James. In Q3 2026, total revenue was RMB207.3 billion, up 8% year-on-year; gross profit was RMB115.9 billion, up 7% year-on-year; operating profit was RMB59.1 billion, down 7% year-on-year. Interest income was RMB4 billion, down 6% year-on-year, due to lower average interest yields. Finance costs were RMB3.5 billion, compared with RMB3.8 billion in the same period last year, mainly reflecting favorable foreign exchange movements, partly offset by higher interest expenses due to increased debt. Our share of profits of associates and joint ventures was RMB4.5 billion, compared with RMB7.8 billion in the same period last year. On a non-IFRS basis, our share of profits of associates and joint ventures was RMB8.8 billion, compared with RMB10.3 billion in the same period last year, mainly due to lower profits at certain associates in China's internet industry, also affected by market consensus expectations and adjustment factors. Income tax expense declined 7% year-on-year to RMB9.1 billion, mainly due to lower related tax expenses at certain overseas subsidiaries. Excluding this impact, our income tax expense would have increased year-on-year. On a non-IFRS basis, operating profit was RMB68.1 billion, down 6% year-on-year; excluding new AI products, operating profit was RMB83.3 billion, up 8% year-on-year. Net profit attributable to equity holders was RMB66.5 billion, down 6% year-on-year. Diluted EPS was RMB7.227, down 5% year-on-year. Overall gross margin in Q3 was 56%, broadly stable year-on-year. By segment, VAS gross margin improved 3 percentage points year-on-year to 64%, mainly benefiting from a revenue mix shift toward higher-margin self-developed games. Marketing services gross margin was 52%, down 4 percentage points year-on-year, mainly due to AI-related operating costs and increased content costs from building the micro short drama service. Fintech and business services gross margin was 50%, broadly stable year-on-year. In Q3, selling and marketing expenses grew 6% year-on-year to RMB12.1 billion, mainly due to increased promotional efforts to drive adoption of AI-native products, including WorkBuddy. R&D expenses grew 42% year-on-year to RMB32.4 billion, mainly reflecting increased R&D investment to enhance the Hunyuan model series, advance WeChat-related projects, and develop AI capabilities in products and services. Excluding R&D expenses, administrative expenses grew 2% year-on-year to RMB11.7 billion, reflecting our control over non-AI-related expenses. Non-IFRS operating margin in Q3 was 33%, down 5 percentage points year-on-year. Excluding new AI products, non-IFRS operating margin was 40%, broadly stable year-on-year. Finally, a few important indicators on cash flow and the balance sheet. Operating capital expenditure payments were RMB63.6 billion, up 257% year-on-year. We accelerated AI infrastructure investment to support capability improvements of the Hunyuan model series and WorkBuddy, inference demand, and the development of AI capabilities in products and services, while also meeting external customer demand for cloud services. Non-operating capital expenditure payments were RMB1.5 billion. Free cash flow was negative RMB69.1 billion, mainly due to large-scale compute procurement. If compute procurement prepayments are excluded, free cash flow would be positive RMB16.9 billion. The company had net debt of RMB13.7 billion, compared with net cash of RMB58.2 billion as of June 30, 2026. The change in net debt position mainly reflected the prepayment spending mentioned earlier, partly offset by RMB8.4 billion of net cash inflow from net disposal activities in the investment portfolio. As of the end of September 2026, the fair value of listed equity investments excluding subsidiaries was RMB505.4 billion, and the carrying value of unlisted investee equity excluding subsidiaries was RMB397.2 billion. Thank you. Q&A Session Wendy Huang, Head of Investor Relations: We will now open the Q&A session. (Operator instructions) If you are joining via Tencent Meeting or the Group Meeting app, please click the raise-hand button at the bottom. We will take one question and one follow-up per person. The first question comes from Robin Zhu of Bernstein. Robin Zhu, Analyst: Hi, thank you Wendy. Thanks to management for answering my questions. I have two questions. First, we have now had two quarters of very high capex and prepayment spending. I'm curious where Tencent is in this spending process to acquire compute. How far do you think we are from reaching a good compute level? And then, how do the tightened restrictions affect existing and future compute transactions? Will this affect the acceleration we are seeing in cloud and other businesses? Second, WorkBuddy is performing very well. Curious whether management can size the opportunity here from a user traction perspective? Is it mainly that heavy users are the addressable market and the revenue profit potential? Given chip restrictions, is this at risk? Thank you. Management: Robin, let me first address the first set of questions on prepayment capex and the impact of advanced semiconductor restrictions on cloud revenue growth. On prepayments, Q2 and Q3 were at abnormally high levels, and we have now signed the long-term contracts we wanted to sign. So our prepayment outflows will decrease. In fact, in certain quarters prepayments may reverse due to cancellation of contracts we previously prepaid. But due to recent changes in capex, we are now exiting. Then for the current run rate, we would look at the cash capex basis, which I think is RMB65 billion, rather than RMB81 billion of accrued capex. Because if you look at prepayments and capex together, then you should look at prepayments and cash capex. Otherwise, looking at prepayments and accrued capex would be double counting. Anyway, looking at the cash capex level, we think this is at a peak level, and cash capex may fluctuate around that peak level quarter by quarter for the remainder of 2026 and in 2027, and then we expect it to decline thereafter. As for how restrictions affect our cloud business, we actually don't see an impact, because our cloud business today is mainly running inference, especially inference for efficient models such as Hy4 Flash and other Flash models. We are now at a stage where inference can run very effectively on a surprisingly limited number of domestic chips. So, I think there has been some shift in inference productivity over the past 12 months. As a result, we are able to drive far more tokens on domestic chips than imagined a year ago. And it is inference that drives cloud revenue. So actually, we do not believe our cloud revenue will be materially adversely affected by these changes. Management: Yes, let me add a few points. First, on capex, I think to reiterate, the capex we are spending now is actually infrastructure spending to kick-start our AI-native businesses. It is somewhat like building a factory so that you can conduct training and token production. Its nature is that once you have this facility, you can continue to use that facility to train new generations of models and also continue to generate tokens. So that is why once we spend in 2026 and 2027, the trend into 2028 will be better, because you can continue to use the infrastructure we build in 2026 and 2027. Second, the prices we are seeing in 2026 and 2027 are actually on the high side. Going into 2028, even if you need to buy the same things, you should expect prices to fall significantly. Third, and I think most importantly, after completing the training and inference work for our models, we see that computational efficiency is actually multiplying year-on-year. This means that over time, the same facility will be able to achieve higher training efficiency and more tokens. With these factors, we actually think 2028 and beyond will be much better. Now, regarding the impact of the latest tighter restrictions on our AI business, I would like to say a few things. First, everyone has less advanced compute available, regardless of when you start leasing or building capacity outside China. Second, training will be affected. But our Hy4 training efficiency is actually very high. Looking at the next-generation version, in the short term we may need longer to launch the next version, but over time the domestic supply chain will catch up. Looking at inference, as James said, the models we build are actually very efficient, so the domestic supply chain should be able to cover it, and this should not be a problem. Looking at WorkBuddy as a product, WorkBuddy first uses very efficient models, but it also uses a range of different models and obtains a very diverse token supply. Some are provided by us, and some are actually provided by various model companies hosting WorkBuddy. In fact, WorkBuddy has become the largest customer of almost all open-weight models in China. In the long run, we believe all Chinese players will actually rely on the domestic supply chain and domestic chips, and we do see the industry progressing very fast; with the latest restrictions, we expect it to accelerate further, because all development efforts will have to focus on accelerating the domestic supply chain. Now on WorkBuddy, we are very, very excited about the recent growth. When we look at the AI business, the characteristic of the AI business compared with the internet era is that in the internet era, if your business doubled, that was considered a very good result. But in the AI era, it is often a successful product or model that achieves 10x growth, and I think WorkBuddy actually falls into this category. So we are very pleased to see this growth traction. WorkBuddy has reached scale in terms of DAU, and DAU continues to grow very rapidly. If you look at the AI business from the perspective of delivering intelligence, it is already China's largest AI application, and tokens are a good proxy indicator. It is already by far China's largest AI application. Token growth is actually accelerating as well. We do see it as a great new model in the enterprise market. We explained there is no piracy. You actually cannot have a small group of people build an organization that mixes use of WorkBuddy. WorkBuddy basically provides a whole suite of different functions. So it is not just one piece of software; it is almost a whole suite of software functionality delivered to you. I think perhaps most importantly, it is a great business model, because as we begin to drive monetization, the free tier actually allows you to acquire users very well, and then the paid tier actually gets some people to pay, and you give them quite a lot of tokens. Then for the truly heavy users, they actually continue to increase token spending. You see users spending up to RMB40,000 per month on tokens. So when you look at thisthis is actually very different from the SaaS business in China in the past, where you could only charge a fixed fee based on subscription. So we feel very good and excited about WorkBuddy, and over time, I think, first, it has an excellent market position. Everyone thinks of WorkBuddy when talking about productivity AI, and as it collects more and more data, it actually makes our models and the models we work with better. At the same time, it lets users have more and more skills, and more and more work products are stored in WorkBuddy's memory. As they integrate into enterprise workflows, we think it will be very sticky. So we are really enjoying WorkBuddy's amazing growth. Wendy Huang, Head of Investor Relations: Thank you. Robin Zhu, Analyst: Thank you. Wendy Huang, Head of Investor Relations: Thank you, Robin. The next question comes from Alex Liu of Bank of America. Alex Liu, Analyst: Thank you for the opportunity. I have two questions. First, on the recent Tencent 28th anniversary, I think management talked about organizational change. Can you share more about this strategic thinking and how broad the changes in different business units and resource allocation should be? Where should we expect the earliest financial impact to appear? Management: Well, we are at a very early stage. But the insight we see is, first, AI can indeed be applied across various businesses to improve operational efficiency. But to be able to do that, you actually need to undertake fairly significant organizational restructuring and also redesign many workflows. If we can do that, then existing businesses can actually operate with fewer people. But at the same time, what we see is that AI actually allows small teams to leverage AI to explore new ideas, build new businesses, and establish new growth drivers. If we can actually operate existing businesses with fewer people and redeploy those people to build new growth drivers for existing businesses as well as new businesses, then I think that would be the ideal state. That is the whole idea behind this. As we build AI-native businesses, we are increasingly excited about this opportunity, because we have seen it in the model team. We have seen it in the WorkBuddy team. We have also seen it when WorkBuddy is applied to our customers and actually helps substantially improve efficiency. So that is why we basically say that now is the time for us to make bigger changes to the organization. If you ask what we have already seen? I think our model team actually rebuilt completely from scratch and within a year pushed our models to China's top position. That is one example. The WorkBuddy team was unknown six or seven months ago, but then they built such an exciting business. I think that is another example, and we hope to see more and more such examples over time. Wendy Huang, Head of Investor Relations: Thank you, Alex. The next question comes from Alex Yao of JPMorgan. Alex Yao, Analyst: Thank you, management, for answering my questions. I have two questions. First, on global AI deployment, as you bring Hunyuan and WorkBuddy to users outside China and grow the international cloud business. How do you think about where AI workloads are best hosted between domestic infrastructure and international regions? As market conditions change in different markets, how much flexibility do you have to adjust this mix? The second question is about WorkBuddy's economics. On one hand, we are pleased to see very rapid growth in users and token consumption, but on the other hand, we also see many very efficient yet very cheap Flash models being introduced into WorkBuddy, which arguably lowers the overall blended pricing of token consumption in WorkBuddy. So can you help understand the economic relationship between volume growth and price declineeven if cost per workload or price declines, do you still see a large opportunity for monetization uplift? Thank you. Management: On where hosting should