Tencent Holdings Limited (OTCPK:TCEHY) Q2 2025 Earnings Conference Call August 13, 2025 8:00 AM ET
Company Participants
Chi Ping Lau - President
Huateng Ma - Co-Founder, Chairman & CEO
James Gordon Mitchell - Chief Strategy Officer & Senior EVP
Shek Hon Lo - Corporate Participant
Wendy Huang - Investor Relations Officer
Conference Call Participants
Alex C. Yao - JPMorgan Chase & Co, Research Division
Alicia Yap - Citigroup Inc., Research Division
Charlene Liu - HSBC Global Investment Research
Ellie Jiang - Macquarie Research
Kenneth Fong - UBS Investment Bank, Research Division
Robin Zhu - Sanford C. Bernstein & Co., LLC., Research Division
Ronald Keung - Goldman Sachs Group, Inc., Research Division
Thomas Chong - Jefferies LLC, Research Division
William Henry Packer - BNP Paribas Exane, Research Division
Yuan Liao - Citic Securities Co., Ltd., Research Division
Wendy Huang
Good day, and good evening. Thank you for standing by. Welcome to Tencent Holdings Limited 2025 Second Quarter Results Announcement Webinar. I’m Wendy Huang from Tencent IR team.
大家好，早上好/晚上好。感谢大家的耐心等待。欢迎参加腾讯控股有限公司 2025 年第二季度业绩发布网络研讨会。我是来自腾讯投资者关系团队的 Wendy Huang。
[Operator Instructions]
【操作提示】
Please be advised that today’s webinar is being recorded.
请注意，今天的研讨会正在录音。
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.
在开始演示之前，我们要提醒大家，本次内容包含前瞻性陈述，这些陈述存在诸多风险与不确定性，未来可能因各种原因而无法实现。有关一般市场状况的信息来自腾讯以外的多个来源。本次演示还包含一些未经审计的非 IFRS 财务指标，应当与根据 IFRS 编制的集团财务业绩指标一并考虑，而不是替代。如需了解风险因素及非 IFRS 指标的详细讨论，请参阅我们网站投资者关系板块披露的相关文件。
Now let me introduce the management team on the webinar tonight. Our Chairman and CEO, Pony Ma, will kick off with a short overview. President, Martin Lau; and 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.
下面让我介绍今晚参会的管理团队。我们的董事会主席兼首席执行官 Pony Ma 将首先进行简要概述。总裁 Martin Lau 和首席战略官 James Mitchell 将带来业务回顾。最后由首席财务官 John Lo 作财务总结，然后进入问答环节。
I will now pass it to Pony.
接下来我把时间交给 Pony。
Huateng Ma
Thank you, Wendy. Good evening. Thank you, everyone, for joining us. During the second quarter of 2025, we delivered double-digit revenue and non-IFRS operating profit growth as we invest in and also benefit from utilizing AI. Our games performed well in terms of users and revenue. As evergreen games such as Honour of Kings and Peacekeeper Elite evolved into platforms while increasing their usage of AI and as new games such as Delta Force broke out.
谢谢你，Wendy。各位晚上好。感谢大家的参与。在 2025 年第二季度，我们在投资并受益于 AI 的同时，实现了收入和非 IFRS 经营利润的双位数增长。我们的游戏在用户和收入方面表现出色。随着《Honour of Kings》和《Peacekeeper Elite》等长青游戏逐步演化为平台并加大 AI 应用，以及新游戏《Delta Force》的爆发式增长，我们的游戏业务全面向好。
Our marketing services revenue sustained rapid growth as we upgrade our advertising foundation model, leading to a better performance of ads across our traffic platforms. We are striving to bring further benefits of AI to consumers and enterprises through powering more use cases within Weixin and driving usage of our AI native app Yuanbao and upgrading the capabilities of our HunYuan foundation models.
随着我们不断升级广告基础模型，营销服务收入保持快速增长，这带动了广告在各类流量平台上的更佳表现。我们正努力通过在微信生态中推动更多 AI 应用场景、带动 AI 原生应用“元宝”的使用，并持续升级“混元”基础模型的能力，为消费者和企业带来更多 AI 价值。
Looking at our financial numbers for the second quarter. Total revenue was RMB 185 billion, up 15% year-on-year. Gross profit was RMB 105 billion, up 22% year-on-year. Non-IFRS operating profit was RMB 69 billion, up 18% year-on-year. and non-IFRS net profit attributable to equity holders was RMB 63 billion, up 10% year-on-year. If we exclude associated profit contribution from the current quarter and from the same quarter last year, when associated profit was high due to better profitability and catch-up adjustment for a large associate, our underlying net profit would have increased by 20% year-on-year.
从第二季度的财务表现来看，总收入为人民币 1,850 亿元，同比增长 15%；毛利润为人民币 1,050 亿元，同比增长 22%；非 IFRS 经营利润为人民币 690 亿元，同比增长 18%；归属于股东的非 IFRS 净利润为人民币 630 亿元，同比增长 10%。如果剔除本季度及去年同期来自某大型联营企业的利润贡献（该联营企业去年同期因盈利能力改善和一次性调整带来较高利润），我们的核心净利润同比将增长 20%。
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 RMB 1.4 billion. For digital content, TME solidified its leadership position in new stream. For games, we extend our leadership position with the progress in gameplay-centric games such as Peacekeeper Elite and Delta Force as well as in content-centric games such as Naruto and Wuthering Waves. For cloud, Gartner has recognized Tencent Cloud as one of the top communication platform as a service solutions globally for the 3 consecutive years.
再看我们的核心服务。在通讯与社交网络方面，微信及 WeChat 的合并月活跃账户数环比、同比增长至 14 亿。在数字内容方面，TME 巩固了其在新音乐流媒体市场的领导地位。在游戏方面，我们在以玩法为核心的游戏（如《Peacekeeper Elite》和《Delta Force》）以及以内容为核心的游戏（如《Naruto》和《Wuthering Waves》）中都进一步扩大了领先优势。在云业务方面，Gartner 已连续三年将腾讯云评为全球领先的通信平台即服务（CPaaS）解决方案提供商之一。
I will now hand over to Martin for business review.
接下来请 Martin 带来业务回顾。
Chi Ping Lau
Thank you, Pony, and good evening and good morning to everybody. For the second quarter of 2025, our total revenue was up 15% year-on-year. VAS represented 50% of our total revenue, within which Social Networks subsegment was 18%, Domestic Games subsegment was 22% and international games was 10%. Marketing Services was 19% of total revenue, and FinTech and Business Services was 30% of total revenue. In terms of gross profit, our gross profit was up 22% year-on-year in the second quarter to RMB 105 billion.
谢谢 Pony，各位晚上好/早上好。2025 年第二季度，我们的总收入同比增长 15%。增值服务（VAS）占总收入的 50%，其中社交网络板块占 18%，国内游戏板块占 22%，国际游戏板块占 10%。营销服务占 19%，金融科技与企业服务占 30%。在毛利润方面，第二季度毛利润同比增长 22%，达到人民币 1,050 亿元。
Notably, gross profit for each of our 3 reporting segments grew over 20% year-on-year. VAS gross profit increased 23% year-on-year to RMB 55 billion, representing 53% of our total gross profit. Marketing Services gross profit increased 24% year-on-year to RMB 21 billion, contributing 20% of total gross profit. And FinTech and Business Services gross profit increased 21% year-on-year to RMB 29 billion, contributing 28% of our total gross profit.
值得注意的是，我们三个报告分部的毛利润均实现超过 20% 的同比增长。VAS 毛利润同比增长 23% 至人民币 550 亿元，占总毛利润的 53%；营销服务毛利润同比增长 24% 至人民币 210 亿元，占总毛利润的 20%；金融科技与企业服务毛利润同比增长 21% 至人民币 290 亿元，占总毛利润的 28%。
Turning to business segments. Value-added services revenue was RMB 91 billion, up 16% year-on-year. Social networks revenue was up 6% year-on-year, driven by increased revenue from app-based game item sales, video account-like streaming service and music subscriptions. Music subscription revenue increased 17% year-on-year, supported by growth in ARPU as well as subscribers. Overall subscriber grew 6% year-on-year to 124 million. Our Super VIP subscribers exceeded 15 million, benefiting from privileges such as collectible artist cards and expanded early access to merchandise and live events.
分业务来看，增值服务收入为人民币 910 亿元，同比增长 16%。社交网络收入同比增长 6%，主要受益于应用内道具销售、视频号类直播服务及音乐订阅收入的增加。音乐订阅收入同比增长 17%，得益于 ARPU 提升以及订阅用户增长。整体订阅用户数同比增长 6%，达到 1.24 亿。我们的超级 VIP 用户数超过 1,500 万，受益于收藏类艺人卡片、商品与演出提前购等权益。
Long-form video subscription revenue decreased 2% year-on-year. Video subscribers declined 3% year-on-year to 114 million due to fewer scheduled releases of top-tier content. Despite a light release quarter, our self-commissioned drama series, the Prisoner of Beauty was the most watched drama series across all long-form video platforms in May. Domestic games revenue grew by 17% year-on-year, driven by growth contributions from Delta Force and from evergreen games, notably Honour of Kings, VALORANT and Peacekeeper Elite. International games revenue increased by 35% year-on-year or 33% in constant currency terms, driven by Supercell’s games, PUBG MOBILE and the release of Dune: Awakening.
长视频订阅收入同比下降 2%，视频订阅用户数同比下降 3% 至 1.14 亿，原因是顶级内容的排播减少。尽管本季度上线较轻，但我们自制的电视剧《Prisoner of Beauty》在 5 月成为所有长视频平台中观看量最高的剧集。国内游戏收入同比增长 17%，主要受《Delta Force》及长青游戏（尤其是《Honour of Kings》、《VALORANT》和《Peacekeeper Elite》）的贡献带动。国际游戏收入同比增长 35%（按不变汇率计算增长 33%），主要受 Supercell 游戏、《PUBG MOBILE》以及新上线的《Dune: Awakening》推动。
Now moving on to communications and social networks. Mini programs increasingly act as a powerful platform for users to connect with merchants and content providers. GMV facilitated by Mini Programs grew by a teens percentage year-on-year in the second quarter, benefiting from improved support for use cases, including financial services, dine-in ordering and transportation. For Mini Games, the total gross receipt increased 20% year-on-year. We upgraded Mini Games technology infrastructure with expanded game engine compatibility, enhanced graphics rendering and reduced load time, which together facilitate developers in porting complex app-based games to mini games, and these games include simulation and MMORPGs.
接下来谈到通信与社交网络。小程序日益成为用户与商家、内容提供者之间的强大连接平台。2025 年第二季度，小程序促成的 GMV 同比增长达到两位数，受益于在金融服务、堂食点餐、出行等场景上的支持提升。小游戏的总流水同比增长 20%。我们升级了小游戏的技术基础设施，扩大了游戏引擎兼容性、增强了图形渲染能力并缩短了加载时间，使开发者能够更便捷地将复杂的 App 游戏移植为小游戏，这其中包括模拟类与 MMORPG 游戏。
For mini shops, we allowed brands and merchants to port their SKU libraries for Mini Programs to mini shops, and to unify loyalty programs across 2 platforms. In addition, we continued to strengthen the social commerce experience, which distinguishes Weixin from traditional e-commerce platforms. First, we extended gifting feature from Mini Shops to Mini Programs, Video Accounts and official accounts. Second, we introduced the shop with friends feature, encouraging users to share products and shops they like with friends via chats and moments and participate together in group discount deals.
在小商店方面，我们允许品牌和商家将小程序的 SKU 库直接移植到小商店，并统一跨平台的会员体系。此外，我们继续强化社交电商体验，这使微信有别于传统电商平台。首先，我们将小商店的送礼功能扩展至小程序、视频号及公众号。其次，我们上线了“和朋友一起逛店”功能，鼓励用户通过聊天和朋友圈分享他们喜欢的商品和店铺，并共同参与团购优惠。
On the AI front, we added AI-powered citation to content so that users reading official accounts articles or video accounts comments can activate contextual AI commentary on related information. We upgraded Mini Shops customer service with large language model capabilities to provide merchants with more intelligent responses to customer inquiries and personalized product recommendations. We enabled Yuanbao as a Weixin contact to interpret and summarize video accounts content. Meanwhile, we are rapidly enhancing the functionalities of our AI native app Yuanbao, and we’ll share more details about how we are growing the DAU later this year.
在 AI 方面，我们为内容新增了 AI 驱动的引用功能，让用户在阅读公众号文章或视频号评论时，可调用上下文相关的 AI 解读。我们为小商店客服引入大语言模型能力，为商家提供更智能的客户咨询回复和个性化产品推荐。我们还将“元宝”作为微信联系人，帮助解读和总结视频号内容。同时，我们正在快速增强 AI 原生应用“元宝”的功能，并将在今年稍晚分享我们如何推动其 DAU 增长的更多细节。
And with that, I’m passing on to James.
下面我把话交给 James。
James Gordon Mitchell
Thank you, Martin. Moving to domestic games. Delta Force has grown to be among the top 5 games by DAU and top 3 games by gross receipts market-wide in China in July. The game’s advanced architecture and modular design enable us to rapidly introduce new features such as underwater combat and a dynamic weather system. Delta Force’s monthly average DAU has been trending up in recent months, reaching a record high of over 20 million in July. Several of our existing evergreen titles demonstrated vitality in terms of user engagement as well as revenues.
谢谢你，Martin。接下来谈到国内游戏。《Delta Force》已在 7 月跻身中国市场 DAU 前五和流水前三的游戏。该游戏先进的架构与模块化设计使我们能够快速上线新特性，如水下作战和动态天气系统。《Delta Force》近几个月的月均 DAU 持续上升，并在 7 月创下超 2,000 万的历史新高。我们的一些长青游戏在用户活跃度和收入方面同样展现了强劲生命力。
For example, Peacekeeper Elite increased average DAU by more than 30% year-on-year in the quarter due to the rising popularity of its extraction shooter mode. VALORANT in China achieved record high average DAU in the second quarter, benefiting from eSports tournaments and a new larger scale map, Corrode. VALORANT Mobile will launch next week in China, extending the game’s presence from PC, PlayStation and Xbox to mobile. Reviewing the progress of our game business domestically and internationally in recent months, AI has become an increasingly important driver of its growth in terms of game content, game engagement and game monetization. We’re increasingly applying AI tools to boost the speed and scale of content production across our major games.
例如，《Peacekeeper Elite》在本季度的平均 DAU 同比增长超过 30%，主要得益于其“撤离射击”模式的流行。在中国，《VALORANT》在第二季度创下平均 DAU 新高，受益于电竞赛事及全新大地图“Corrode”的上线。《VALORANT Mobile》将于下周在中国上线，把该游戏的覆盖从 PC、PlayStation 和 Xbox 扩展到移动端。回顾近几个月国内外游戏业务的进展，AI 已成为推动其增长的关键因素之一，涵盖游戏内容、用户参与度和商业化。我们正越来越多地应用 AI 工具，加速并扩展核心游戏的内容生产。
AI allows us to provide more human-like virtual teammates in our competitive PvP games and to power more realistic nonplayer characters in our story-driven PvE games. And we’re using AI in our game marketing activities to more efficiently target marketing spending towards the users most likely to activate and remain in each game. On international games, for PUBG MOBILE, a rising percentage of DAU are playing user-generated experiences in the World of Wander sandbox. The popularity of the Metro extraction shooter mode and now the World of Wonder user-generated content environment demonstrate our progress in upgrading PUBG MOBILE from a game as a service to a game as a platform.
AI 使我们能够在竞技类 PvP 游戏中提供更接近真人的虚拟队友，并在剧情驱动型 PvE 游戏中赋予 NPC 更逼真的表现。同时，我们也在游戏营销中应用 AI，更高效地将营销资源投入到最有可能激活并长期留存的用户群体。在国际游戏方面，《PUBG MOBILE》有越来越多的 DAU 在“世界创作者”沙盒中体验用户生成内容。“地铁撤离”模式的走红，以及“World of Wonder”UGC 环境的兴起，表明我们正在将《PUBG MOBILE》从一款“游戏即服务”升级为“游戏即平台”。
Monthly gross receipts of PUBG MOBILE reached a record high level in April, driven by the ancient Egyptian-themed outfit. The Clash Royale, Supercell has accelerated the cadence of content updates and of community events and optimized the reward system, resulting in more attention to live streaming influencers and higher DAU counts. Clash Royale monthly gross receipts hit a 7-year high in June. Dune: awakening, an open world survival game developed and published by our Norwegian subsidiary, Funcom, launched in early June. The game was the highest ranked pay-to-own game by revenue on steam worldwide for the first week of its launch. Fans of survival games and of the Dune novels are responding very favorably to the game, and Funcom has a substantial content pipeline to sustain and expand the game world.
《PUBG MOBILE》的月度流水在 4 月创下历史新高，主要受古埃及主题套装推动。在《Clash Royale》方面，Supercell 加快了内容更新和社区活动节奏，并优化了奖励体系，从而吸引了更多直播影响者关注，并提升了 DAU。《Clash Royale》的月度流水在 6 月达到 7 年新高。《Dune: Awakening》是一款由我们挪威子公司 Funcom 开发并发行的开放世界生存游戏，于 6 月初上线。上线首周，该游戏在 Steam 全球付费制游戏收入榜上位居第一。生存游戏玩家和《沙丘》小说粉丝对这款游戏反响热烈，而 Funcom 也有充足的内容储备以维持和扩展这一游戏世界。
For Marketing Services, revenue grew 20% year-on-year to RMB 36 billion in the quarter, benefiting from AI-powered adtech upgrades and from increased closed-loop advertising arising from Weixin’s transactional ecosystem. We expanded AI capabilities in areas, including ad creation, placement, recommendation and performance analysis, which had the effect of boosting click- through rates, conversions and ROI for advertisers. Specifically, we upgraded our ad platform architecture by deploying a scaled-up foundation model, which analyzes advertisement click-through rates and transactions across multiple apps and services as well as user interactions across text, image and video to determine user interest and optimize ad performance in real time.
在营销服务方面，本季度收入同比增长 20%，达到人民币 360 亿元，受益于 AI 驱动的广告技术升级，以及微信交易生态推动的闭环广告增长。我们在广告创意、投放、推荐和效果分析等领域扩展了 AI 能力，从而提升了广告主的点击率、转化率和 ROI。具体来说，我们通过部署规模化基础模型，升级了广告平台架构，该模型可分析跨多应用和服务的广告点击率与交易数据，以及用户在文本、图片和视频中的交互，从而实时判断用户兴趣并优化广告表现。
By property, Video Accounts marketing services revenue rose approximately 50% year-on-year due to more traffic and more transactional activity within Video Accounts. Mini Programs marketing services revenue also increased about 50% year-on-year. Activity within Mini Games and Mini Dramas created a flywheel effect, which drives more developers to use our closed-loop marketing solutions to promote their services. And Weixin Search revenue grew around 60% year-on-year due to more consumer and advertiser interest in Mini Program search results and to enhance ad relevance as we leverage our large language model to deepen understanding of merchandise and of user consumption intent.
分业务来看，视频号的营销服务收入同比增长约 50%，受益于流量提升及视频号内交易活动增加。小程序的营销服务收入同比同样增长约 50%。小游戏与微短剧的活跃度形成了飞轮效应，推动更多开发者采用我们的闭环营销解决方案推广其服务。微信搜索收入同比增长约 60%，得益于用户和广告主对小程序搜索结果的兴趣提升，同时我们利用大语言模型加深了对商品和用户消费意图的理解，从而增强了广告相关性。
Looking at FinTech and Business Services, segment revenue was RMB 56 billion, up 10% year-on-year. FinTech Services revenue growth increased to a high single-digit percentage year-on-year, primarily driven by commercial payment services and consumer loan services. For Commercial Payment Services, payment volume turned positive year-on-year in the second quarter as the decline in value per transaction narrowed and the number of transactions grew at a faster rate versus prior quarters. Online total payment volume sustained a healthy growth rate, while the previously weak offline total payment volume trend improved due to increased spending in categories such as retail and dining services.
在金融科技与企业服务方面，该板块收入为人民币 560 亿元，同比增长 10%。金融科技服务收入同比增长至高个位数，主要由商业支付服务和消费信贷服务推动。在商业支付服务中，随着单笔交易金额降幅收窄、交易笔数加快增长，第二季度支付总量同比转正。线上支付总量保持健康增速，而此前疲软的线下支付总量趋势因零售、餐饮等消费类别的支出增加而得到改善。
整体增长仍然很有限，自2023年以来的增速比较慢，2023年全年2037.60亿，2025年上半年1104亿。
Business services revenue grew at a teens rate year-on-year. Cloud services revenue growth accelerated versus recent quarters, benefiting from increased revenue from providing GPUs and API tokens for customers’ AI needs. Fees collected on Mini Shops transactions continue to grow at a rapid rate and business services gross margin rose year-on-year due to improved efficiency and positive mix shifts. Our international cloud revenue increased significantly year-on-year. We’re increasingly proficient in helping large international clients migrate to Tencent Cloud and so uplift their IT efficiency. For example, we facilitated Gojek Tokopedia Group’s on-demand service systems migration, one of the most complex cloud migrations in Southeast Asia.
企业服务收入同比增长达到两位数。云服务收入增速较前几个季度加快，受益于向客户提供 GPU 和 API tokens 以满足其 AI 需求所带来的收入增长。小商店交易手续费继续快速增长，企业服务毛利率也因效率改善和业务结构优化而同比提升。我们的国际云收入同比显著增长，帮助大型国际客户迁移至腾讯云的能力日益成熟，从而提升其 IT 效率。例如，我们协助 Gojek Tokopedia Group 完成了按需服务系统的迁移，这是东南亚最复杂的云迁移项目之一。
For HunYuan, we enhanced our data quality and diversity through data augmentation and synthesis and implemented more effective pretraining and post-training scaling. HunYuan 3D model has become the top ranked 3D generative model on Hugging Face due to its geometric precision, texture fidelity and prompt 3D alignment capabilities. Game developers, 3D printing enterprises and design professionals are increasingly using the HunYuan 3D model for their digital asset generation needs.
在“混元”方面，我们通过数据增强和合成提升了数据质量与多样性，并实施了更高效的预训练和后训练扩展。“混元 3D 模型”凭借几何精度、纹理逼真度和 3D 指令对齐能力，成为 Hugging Face 上排名第一的 3D 生成模型。游戏开发者、3D 打印企业和设计专业人士正越来越多地使用“混元 3D 模型”来满足数字资产生成需求。
And I’ll now pass to John to discuss the financial review.
接下来请 John 带来财务回顾。
Shek Hon Lo
Thank you, James. For the second quarter of 2025, total revenue was RMB 184.5 billion, up 15% year-on-year. Gross profit was RMB 105 billion, up 22% Y-o-Y. Other loss was RMB 3.6 billion compared to the gain of RMB 1.5 billion in the same period last year, mainly due to lower subsidies and tax rebates as well as provisions paid for some receivables during the quarter. Operating profit was RMB 60.1 billion, up 18% year-on-year. Interest income was RMB 4.1 billion, up 7% year-on-year. Finance costs were RMB 3.9 billion, up 27% year-on-year due to higher interest expenses and increased foreign exchange losses.
谢谢你，James。2025 年第二季度，总收入为人民币 1,845 亿元，同比增长 15%；毛利润为人民币 1,050 亿元，同比增长 22%。本季度“其他损失”为人民币 36 亿元，而去年同期为收益人民币 15 亿元，主要原因是补贴和税收返还减少，以及部分应收款项的计提准备。经营利润为人民币 601 亿元，同比增长 18%。利息收入为人民币 41 亿元，同比增长 7%。财务费用为人民币 39 亿元，同比增长 27%，主要因利息支出增加及外汇损失上升。
Share of profit of associates and joint venture was RMB 4.5 billion compared to RMB 7.7 billion in the same quarter last year. On a non-IFRS basis, share of profit was RMB 6.3 billion, down from RMB 9.9 billion in the same quarter last year, primarily due to lower estimated associate income from a large associates. Income tax expense increased by 12% year-on-year to RMB 11.4 billion, driven by growth in operating profit. On non-IFRS basis, diluted EPS was RMB 6.793, up 13% year-on-year, outpacing non-IFRS net profit growth due to reduced share count after our share buybacks. Our weighted average number of shares for calculating diluted EPS decreased by 2% year-on-year.
联营公司及合营企业的应占利润为人民币 45 亿元，而去年同期为人民币 77 亿元。按非 IFRS 口径，应占利润为人民币 63 亿元，低于去年同期的人民币 99 亿元，主要原因是对某大型联营企业的预估利润减少。所得税费用同比增长 12%，至人民币 114 亿元，主要受经营利润增长推动。非 IFRS 口径下，每股摊薄收益为人民币 6.793 元，同比增长 13%，高于非 IFRS 净利润增速，原因是回购股份后股本减少。用于计算摊薄 EPS 的加权平均股数同比下降 2%。
On non-IFRS financial figures. Operating profit was RMB 69.2 billion, up 18% year-on-year. Net profit attributable to equity holders was RMB 63.1 billion, up 10% year-on-year. Excluding associate and JV profit contribution in both the current quarter and the same quarter last year, our net profit would have increased by 20% Y-o-Y to RMB 56.8 billion.
在非 IFRS 财务口径下，经营利润为人民币 692 亿元，同比增长 18%；归属于股东的净利润为人民币 631 亿元，同比增长 10%。如果剔除本季度和去年同期联营及合营公司的利润贡献，我们的净利润将同比增长 20%，达到人民币 568 亿元。
Moving on to gross margins. Overall gross margin was 57%, up 4 percentage points year-on-year. By segment, VAS gross margin was 60%, up 3 percentage points year-on-year, primarily driven by a higher mix of higher-margin domestic games revenue alongside margin expansion in our music subscription services driven by growth in ARPU and subscribers, as mentioned earlier. Marketing Services gross margin was 58%, up 2 percentage points year-on-year, supported by growth in high-margin media accounts and Weixin search revenue. FinTech and Business Services gross margin was 52%, up 5 percentage points year-on-year, mainly due to favorable revenue mix shift to wealth management and consumer loan services and improved cost efficiency in our payment and services.
毛利率方面，整体毛利率为 57%，同比提升 4 个百分点。分业务来看，增值服务毛利率为 60%，同比提升 3 个百分点，主要受高毛利国内游戏收入占比提升，以及音乐订阅服务在 ARPU 和订阅用户增长推动下的利润率扩张带动。营销服务毛利率为 58%，同比提升 2 个百分点，受益于高毛利视频号及微信搜索收入的增长。金融科技与企业服务毛利率为 52%，同比提升 5 个百分点，主要由于收入结构向财富管理和消费信贷服务倾斜，以及支付和服务的成本效率改善。
On second quarter operating expenses. Selling and marketing expenses were RMB 9.4 billion, up 3% year-on-year, reflecting promotional efforts to support the growth of our AI native applications. Selling and marketing expenses represented 5% of revenues, down from 6% in the same quarter last year. R&D expenses rose by 17% year-on-year to RMB 20.3 billion, reflecting higher staff and increased investment to support our AI native applications. G&A, excluding R&D expenses increased by 14% year-on-year to RMB 11.6 billion due to higher staff, including performance-based rewards at certain overseas subsidiaries. At quarter end, we had approximately 111,000 employees, up 5% year-on-year or 2% Q-on-Q driven by fresh graduate hires. Our second quarter non-IFRS operating margin was 38%, up 1 percentage point year-on-year.
再看第二季度的运营费用。销售与市场开支为人民币 94 亿元，同比增长 3%，反映了我们为支持 AI 原生应用增长所作的推广投入。销售与市场费用占收入的 5%，低于去年同期的 6%。研发费用同比增长 17%，达到人民币 203 亿元，反映了研发人员增加以及在 AI 原生应用上的更多投资。除研发费用外的一般及行政开支同比增长 14%，至人民币 116 亿元，原因包括员工增加以及部分海外子公司的绩效奖励。在季度末，我们的员工总数约为 111,000 人，同比增加 5%，环比增加 2%，主要由应届毕业生招聘驱动。第二季度非 IFRS 经营利润率为 38%，同比提升 1 个百分点。
To conclude, I will highlight some key cash flow and balance sheet metrics. Operating CapEx was RMB 17.9 billion, up 149% year- on-year, driven by increased investments in GPUs and servers to ramp up our AI capabilities. Nonoperating CapEx was RMB 1.2 billion, down 20% year-on-year. Our total CapEx was RMB 19.1 billion, up 119% year-on-year. Free cash flow was RMB 43 billion, up 7% year-on-year, driven by growth in games, gross receipts. On a Q-on-Q basis, free cash flow was down 9% due to seasonally lower games gross receipt following the spring festival period. Net cash position was RMB 74.6 billion, down 17% quarter-on-quarter or RMB 15.6 billion, primarily due to final dividend payments of RMB 37.5 billion for financial year 2024.
最后，总结一下现金流和资产负债表的关键指标。运营资本支出为人民币 179 亿元，同比增长 149%，主要由于在 GPU 和服务器上的投入增加，以提升我们的 AI 能力。非运营资本支出为人民币 12 亿元，同比下降 20%。总资本支出为人民币 191 亿元，同比增长 119%。自由现金流为人民币 430 亿元，同比增长 7%，主要受游戏流水增长带动。环比来看，自由现金流下降 9%，原因是春节后游戏流水季节性走弱。净现金为人民币 746 亿元，环比下降 17%（即减少人民币 156 亿元），主要由于为 2024 财年支付期末股息人民币 375 亿元。
Thank you.
谢谢。
Wendy Huang
Thank you, John. We shall now open the floor for questions.
谢谢 John。接下来进入问答环节。
Question-and-Answer Session
问答环节
Wendy Huang
We will take the first question from Thomas Chong from Jefferies.
我们先请 Jefferies 的 Thomas Chong 提问。
Thomas Chong
Congratulations on a very strong set of results. My question is about advertising. We have seen Tencent and other overseas peers using AI in driving ad revenue growth. Can management comment about the potential of marketing services over the next few quarters and coming years? In terms of ad growth driver, how should we think about traffic, click-through rate and ad load for video accounts?
恭喜公司取得强劲的业绩。我想问的是关于广告。我们注意到腾讯和其他海外同行都在利用 AI 来推动广告收入增长。管理层能否谈谈未来几个季度乃至未来几年营销服务的潜力？在广告增长驱动因素上，对于视频号的流量、点击率以及广告加载率（ad load），我们应该如何理解？
James Gordon Mitchell
Thank you, Thomas. So on the advertising and the potential, we continue to believe that we enjoy a long and lengthening runway for continuing to grow our advertising revenue at a reasonably healthy rate. And that length of the runway reflects upside in a number of the key variables that determine our marketing services revenue, including the click-through rate where AI delivers better targeting and thus more clicks, including traffic where we see growth in video accounts traffic and search traffic over time in traffic within our AI native experiences, including revenue per click as generative AI used for creating the ads results in more ad demand as well as e-commerce closed-loop transactions resulting in more ad demand.
谢谢你，Thomas。关于广告及其潜力，我们仍然相信广告收入有一条长而且不断延伸的成长跑道，可以维持健康的增长速度。这条跑道的长度反映在多个关键变量上：比如点击率方面，AI 提供更精准的定向，从而带来更多点击；流量方面，我们看到视频号、搜索流量，以及 AI 原生体验中的流量逐步增长；每次点击的收入方面，生成式 AI 用于广告创意，推动了广告需求，同时电商闭环交易也带动了更多广告需求。
And then finally, in ad load, where, as you know, for short video, our ad load is currently in the low to mid-single digits versus our peers who are in the low to mid-teens. So those are the reasons we believe there’s a long runway for growth for our advertising revenue. In terms of prioritizing between the drivers, then in the second quarter, the majority of the advertising revenue growth of 20% year-on-year arose from higher revenue per impression. And that, in turn, was primarily due to a higher click-through rate arising from deploying AI, although also to higher revenue per click arising from more closed-loop activity with mini shops and mini games.
最后在广告加载率方面，目前我们在短视频广告的加载率仍处于低至中个位数，而同行普遍在低至中两位数。这就是我们认为广告收入具备长线增长空间的原因。至于驱动因素的优先级，在第二季度，广告收入同比增长 20% 主要来自单次展示的收入提升，而这主要得益于 AI 部署带来的更高点击率，以及来自小商店和小游戏闭环活动的更高点击收入。
And then a lesser driver was more impressions arising primarily from video accounts traffic growth and search traffic growth, while ad load was not really a driver. So those were the drivers in the second quarter.
此外，次要的驱动因素是视频号和搜索流量增长带来的更多展示次数，而广告加载率并不是主要驱动。这就是第二季度广告增长的具体原因。
Wendy Huang
We will take the next question from Alex C. Yao from JPMorgan.
我们下面请 JPMorgan 的 Alex C. Yao 提问。
Alex C. Yao
The first question is that you guys have been integrating AI – more and more AI features to a large number of mobile Internet applications this year aside from launching native gen AI applications such as Yuanbao. What’s your observation of consumer behavior change when they use these AI embedded mobile Internet applications? And could these consumer behavior change affect the mobile Internet ecosystem? For instance, if people use too much AI search, then they will bypass the website or the public account and that just directly consume the content in the AI summary, which could potentially affect the mobile Internet ecosystem.
我的第一个问题是，你们今年除了推出“元宝”等原生生成式 AI 应用外，还在大量移动互联网应用中集成了越来越多的 AI 功能。你们观察到用户在使用这些嵌入 AI 的应用时，行为有什么变化吗？这些变化是否会影响移动互联网生态？例如，如果人们过度使用 AI 搜索，他们可能会绕过网站或公众号，直接消费 AI 总结的内容，这可能会对移动互联网生态带来影响。
And then my follow-up question is that as you guys continue to offer increasingly more AI features to consumer free of charge, the delivery of these AI features is a lot more expensive than mobile Internet services, which will potentially hurt Tencent’s cost structure. Will management consider to start directly monetizing these consumer-facing AI features in the next 1 or 2 years? If so, what AI features or functions are most monetizable?
我的追问是，随着你们持续免费向用户提供越来越多的 AI 功能，这些功能的交付成本远高于传统移动互联网服务，可能会损害腾讯的成本结构。管理层是否会考虑在未来一到两年内开始直接变现这些面向消费者的 AI 功能？如果是，哪些 AI 功能或特性最具变现潜力？
Chi Ping Lau
Well, first of all, in terms of the AI features, right, I think there is sort of broadly speaking, a number of these features. One is obviously our Yuanbao, which is an AI native app. And then I would say it’s related to search, AI-enabled search. So that lands on our browser that also lands on WeChat search. And then there’s a whole host of different features within even games, right, when we have AI-enabled players or in our productivity tool, for example, summary of meetings in our Tencent Meeting and assistance within our Tencent docs, right, to help people to write.
首先，就 AI 功能而言，大体上可以分为几类。第一类是我们的原生 AI 应用“元宝”。第二类是与搜索相关的 AI 搜索功能，这不仅落地在我们的浏览器，也在微信搜索中应用。除此之外，还有大量功能嵌入在游戏和生产力工具中，比如游戏中的 AI 玩家，或者像腾讯会议里的会议纪要功能，以及腾讯文档中的写作助手。
I would say we’re still at an early stage in observing the user behavior. Obviously, I think when users use these, right, I think they tend to be more satisfied if ‘they tend to have sort of higher efficiency in creating the kind of content and getting to the content that they want. I think it’s an improvement in terms of the user experience. And in terms of whether it’s really going to hurt the search, I think the one sort of negative impact that you are pointing to is when there is AI-assisted search, whether it would just show the content rather than leading people to the pages. We have not seen a very big impact on that. I think overall, people tend to be more satisfied in getting the answer directly. And if they want to explore the topic more, they would click on the different links and articles.
我认为我们仍处在观察用户行为的早期阶段。显然，当用户使用这些功能时，他们往往会更满意，因为效率提高了，无论是获取内容还是生成内容的效率。我认为这提升了用户体验。至于是否真的会损害搜索，您提到的负面影响是 AI 搜索可能直接展示答案，而不是引导用户点击页面。但我们目前没有看到显著影响。总体上，人们往往更满意于直接获得答案，而如果他们想深入探索主题，仍然会点击相关的链接和文章。
So I think overall, it's actually not that much of an impact. And at the same time, if you look at the ecosystem within WeChat, the content ecosystem, right, a lot of the page views is actually from non-search-related origins, right? When people subscribe to a writer's content, then that's sort of most of the viewership is actually based up points. So in terms of the impact on specific authors, I think that impact is minimal at this point in time.
所以我认为总体来看，这方面的影响其实并不大。同时，如果你观察微信的内容生态，很多页面浏览量其实并非来源于搜索。当用户订阅某位作者的内容时，大多数阅读量都是基于这种订阅关系而产生的。因此，就对特定作者的影响而言，目前来看是非常有限的。
In terms of your question about the cost, I think we are actually managing the cost in a relatively granular way, right? I think there are a lot of places in which if we can use smaller models, we'll be using smaller models and the cost will be sort of much lower than using the flagship model. And so in a lot of these use cases, the cost is manageable if we can use smaller models. And at the same time, if we continue to improve the efficiency of inference through software upgrades. And as it relates to whether we would be monetizing eventually -- I think eventually, there should be some monetization.
至于你提到的成本问题，我认为我们实际上是以相对精细化的方式来管理的。在很多场景下，如果可以使用小模型，我们就会采用小模型，其成本远低于使用旗舰模型。因此，在很多应用场景中，成本是可控的。同时，我们也在通过软件升级持续提高推理效率。至于是否会最终实现变现——我认为最终应该会有某种形式的变现。
I think in China, in reality, it's actually very hard to use the user paid model, which now populates the U.S. AI tools. And I think over time, we'll try to figure out whether there will be some ad-supported way of monetizing. But at the same time, I want to point out that AI is already contributing to the growth and monetization of our existing businesses in different ways, right? So somehow we could also fund part of this "subsidy" for AI usage by the users through the growth in our other businesses.
我认为在中国，现实情况是用户付费模式（目前在美国 AI 工具中较为常见）很难大规模推广。随着时间推移，我们会探索是否可以通过广告支持的方式来实现变现。但与此同时，我想强调，AI 已经在不同层面推动了我们现有业务的增长与变现。因此，从某种意义上讲，我们也可以通过其他业务增长来部分“补贴”用户在 AI 使用上的成本。
Wendy Huang
Thank you, Alex. We will take the next question from Robin Zhu from Bernstein.
谢谢 Alex。我们下面请 Bernstein 的 Robin Zhu 提问。
Robin Zhu
I guess a couple of questions on gaming. So I think we're seeing the start of -- a very successful start to life for Delta Force. I was wondering if management could share the views on AAA gaming growth in China, including on PC, maybe consoles, whether this is the start of a new market for your business? Or does it replace parts of the current market?
我有几个关于游戏的问题。首先，《Delta Force》上线以来表现非常成功。我想请教管理层怎么看中国 AAA 游戏的增长，包括在 PC 端，甚至在主机端，这是否代表着贵公司进入一个新市场？还是会取代现有市场的部分？
And then a second question on capital allocation on AI. What's the argument against potentially allocating a lot more capital to create new dominant top funnels in Yuanbao, I and some of the other stuff that we're doing as opposed to the kind of more gradual approach that we're taking now.
第二个问题是关于 AI 的资本分配。相比我们现在采取的相对渐进式策略，有什么理由不去投入更多资本，直接打造像“元宝”这样的全新主流入口？
James Gordon Mitchell
Thank you, Robin, for your question. So to confirm, the first question is about AAA games, not about Delta Force. Is that right?
谢谢你，Robin 的问题。确认一下，你的第一个问题是关于 AAA 游戏，而不是专门问《Delta Force》，对吗？
Robin Zhu
Yes about the market opportunity, I guess, illustrated by Delta Force growing a lot.
是的，我主要是想借《Delta Force》的增长来说明 AAA 游戏市场的机会。
James Gordon Mitchell
I mean we would define AAA games as pay-to-own games such as Black Myth: Wukong versus we would define Delta Force as more of a live game or game as a service. So we're optimistic about both, but we think the majority of the market in China today and for the future is live service games such as Delta Force. And Delta Force is gratifying for us, not so much because it's top 3 this and top 5 that, but more because having launched at a certain scale, it's actually been progressively growing in the 9 or so months since its launch in terms of both users and revenue, which is exactly what we want to see from our evergreen game strategy.
我会将 AAA 游戏定义为买断制游戏，比如《Black Myth: Wukong》，而《Delta Force》则更属于长线运营型或“游戏即服务”模式。我们对两者都持乐观态度，但认为中国市场当下和未来的主流还是像《Delta Force》这样的长线服务型游戏。《Delta Force》令我们满意的地方，并不仅仅是跻身市场前三、前五，而是它在上线达到一定规模后，在过去约九个月里用户数和收入持续增长，这正符合我们“长青游戏”战略的目标。
And that persistent growth, we think, demonstrates our ability to produce a high-quality game content at scale, which spans 3 modes within Delta Force, 2 platforms, 3 platforms as of tomorrow when it moves to console as well as both China and international markets. And we're able to do that because of the quality of our developers, the capabilities of our tools and our production process, which includes AI. So very happy with Delta Force.
这种持续增长，证明了我们具备规模化生产高质量游戏内容的能力。《Delta Force》涵盖三种模式、运行在两个平台（明天将扩展到三平台，登陆主机），并覆盖中国和海外市场。我们之所以能够做到这一点，得益于开发团队的实力、工具的能力和生产流程的完善，其中包括 AI 的应用。因此，我们对《Delta Force》的表现非常满意。
In terms of AAA or pay-to-own games, then there is an increasing audience for such games in China. And we think that where people release excellent AAA games, they have the potential to be solidly profitable. The bar to being excellent is very high. And we think AAA games or pay-to-own games will remain a minority of the China market given gamers are accustomed to free-to-play games. But the emergence of AAA in China, it provides a really good opportunity for passion projects that would otherwise be not economical or to become economically feasible without needing to provide live content and ongoing services for 10 years or more.
至于 AAA 或买断制游戏，中国的受众正在逐渐增加。我们认为，一旦推出优秀的 AAA 游戏，它就有可能实现稳健盈利。当然，“优秀”的门槛极高。考虑到中国玩家已经习惯了免费游戏模式，AAA 或买断制游戏仍会是市场的少数。但 AAA 游戏在中国的出现，为那些以往缺乏经济可行性的“热情项目”提供了机会，不必依赖十年以上的内容运营和持续服务，也能实现商业化。
In terms of whether those AAA games growing is cannibalistic of the live service games, we're very confident that that's not the case, and they're complementary. I think that if you look at Tencent Music as an analogy, then there's 100-something million people subscribing to the core music streaming service and then some subset of them will periodically buy digital albums. And that spending on digital albums is supplementary. It's not cannibalistic to the original subscription, the mainstream subscription monetization.
至于 AAA 游戏的增长是否会蚕食长线服务型游戏，我们非常有信心不会，相反它们是互补的。我举一个类比，比如腾讯音乐，有一亿多用户订阅核心流媒体音乐服务，其中一部分用户会不时购买数字专辑。这类消费是补充性的，并不会蚕食主流的订阅变现。
So we think the same thing will be true here that the people who purchased the AAA games will overwhelmingly be people who are already playing live service games, who will continue to play and spend money on live service games. But once every so often, if there's an excellent product like Black Myth: Wukong, then they'll purchase those AAA games as well. So that's on the game question.
所以我们认为，情况在这里也是类似的：购买 AAA 游戏的玩家，大多数本身就是长线服务型游戏的用户，他们会继续玩并在这些游戏中消费。但如果偶尔出现像《Black Myth: Wukong》这样优秀的产品，他们也会选择购买 AAA 游戏。因此，这就是我对游戏问题的回答。
Chi Ping Lau
Well, in terms of the AI spending right now, I would have to say we have been spending quite a bit, and we are also increasing our spending and we'll be spending more as we go along. So that's sort of already happening. But at the same time, we also have to be spending smartly rather than just sort of saying, oh, we're going to go all in and spend on buying a lot of chips and sort of hiring a lot of people and doing a lot of marketing, right? I think we have to sort of spend in the right tempo.
关于 AI 投入，目前我们确实已经花费了不少资金，并且还在持续增加，未来投入还会进一步提升。这种趋势已经在发生。但与此同时，我们必须“聪明地花钱”，而不是一股脑地全盘投入，比如大量采购芯片、大规模招人、疯狂砸营销。我认为我们需要保持合适的节奏去投入。
And in particular, when you look at the 2 products you mentioned, right, for example, \[ IMA ] is right now an innovative product, and there's not that much of an example of this product. And so it's right now in the process of figuring out the optimal product offering and use case. So before we do that, I don't think we should be piling in on marketing expenses of that product.
尤其是你提到的两个产品，比如 \[IMA]，目前它还属于创新型产品，市场上并没有太多同类参考案例。现在更处在探索最佳产品形态和应用场景的阶段。在这之前，我认为没有必要对该产品投入过多的市场营销费用。
Now Yuanbao is a more mature product, there's a very clear product use case. And so we have been spending quite a bit in promoting the product, especially in the first quarter, right? And we -- on the back of that spending, we have actually grown the user base quite a bit. I think the second quarter is really about improving the product so that we can actually retain the large user base that we actually have acquired and give them a great user experience. And as we have been able to do that, then likely, we're going to be ramping up the promotion again.
相比之下，“元宝”已经是一个较为成熟的产品，有非常清晰的应用场景。因此我们在推广上确实投入了不少，尤其是一季度。在这些投入的支持下，我们的用户规模有了明显增长。第二季度的重点更多是优化产品，以便真正留住已获得的大量用户，并为他们提供良好的体验。一旦这一点得到落实，我们很可能会再次加大推广力度。
But the promotion is not going to be all just spending money in acquiring users in the market because bear in mind, we actually have a lot of existing platforms that we can leverage. And I think integrating ramp up with our existing platforms is actually a very important advantage and leverage that we have. And we're going to be doing a lot more of that in the coming months in addition to starting to ramp up the promotion of this product.
但推广不会只是依靠市场上花钱买量。要知道，我们自身已经拥有许多现有平台可以利用。把新产品推广与现有平台紧密结合，这是我们非常重要的优势和杠杆。未来几个月，我们将在 ramp up 产品推广的同时，更加注重这种平台整合式的放大效应。
Wendy Huang
We will take the next question from Kenneth Fong from UBS.
我们下面请 UBS 的 Kenneth Fong 提问。
Kenneth Fong
I have a question about the impact on the new advertising law for gaming company sales and marketing. Under the new ad regulation effective in July, sales and marketing spending in excess of 15% of revenue will need to pay an additional 25% tax. So how do you expect this to affect our advertising income, especially for mini games, which heavily rely on traffic acquisitions, i.e., the sales and marketing could easily surpass this 15% revenue threshold?
我想问的是新广告法对游戏公司销售和营销的影响。根据 7 月起生效的新规定，若销售与营销支出超过收入的 15%，需额外缴纳 25% 税。那么管理层如何预期这会影响我们的广告收入，特别是高度依赖买量获客的小游戏领域？因为这类业务的销售与营销支出很容易超过收入的 15% 门槛。
超过15%就不允许税前扣除，防止内卷的政策，总是倾向于多管一点，多找一些可以管的事。
James Gordon Mitchell
Kenneth, so we don't expect a meaningful impact. Our advertising business has become quite broad-based over time. And if you look at the second quarter, there was an adverse impact from the food delivery companies and some of the e-commerce companies ramping up in food delivery, reducing their advertising spend as they invested more in subsidies. But despite that, our advertising revenue grew 20% year-on-year. So in our view, there's always going to be individual blips up and down in terms of individual categories. But what we're doing in terms of deploying AI within advertising is a much more important variable.
Kenneth，我们认为不会产生实质性影响。随着时间推移，我们的广告业务已经相当多元化。如果你看第二季度，一些外卖公司和电商公司加大了补贴力度，从而减少了广告投放，对我们的广告收入造成了一定冲击。但即便如此，我们的广告收入仍同比增长 20%。因此在我们看来，各个细分行业总会有起伏波动，但真正重要的变量是我们在广告中部署 AI 所带来的效果。
Wendy Huang
We will take the next question from Alicia Yap from Citi.
我们下面请 Citi 的 Alicia Yap 提问。
Alicia Yap
I have 2 questions. First, regarding management comment on the business services revenue growth, see acceleration this quarter, benefiting from the increasing enterprise demand for the GPU rental and also the API token usage. Now that we actually have started to see fruitions from our internal utilization of the AI in our internal application, will Tencent start to allocate more GPU resources to support the growing demand of the external enterprise customer? And if so, could we actually see potential continued acceleration of the business services revenue line in the coming quarters?
我有两个问题。第一个关于企业服务收入的增长。本季度增长加速，受益于企业对 GPU 租赁和 API token 使用需求的上升。随着我们在内部应用中对 AI 的利用已经开始产生成果，腾讯是否会分配更多 GPU 资源来支持外部企业客户的增长需求？如果是的话，我们是否可以在接下来的几个季度里看到企业服务收入线的持续加速？
Second question is a follow-up on the advertising. Obviously, with the upgrade on the foundation model, and also the expansions of the AI capabilities. I recall management mentioned about last quarter, when we achieved that 20% year-over-year growth in the marketing service line is more the -- reaching the upper end of the growth range. Could -- because of this whole AI enhancement, could that 20% becoming the new base and being the lower band of the growth range in the coming quarters if the AI technology continue to drive better conversion and the eCPM improvement, especially we actually have a relatively lower ad loads versus the peers?
第二个问题是关于广告的跟进。显然，随着基础模型的升级以及 AI 能力的扩展，我记得管理层在上个季度提到过，当时营销服务线实现了 20% 的同比增长，已经接近增长区间的上限。那么，在 AI 技术持续推动转化率和 eCPM 提升的情况下，这个 20% 是否可能成为新的基准，成为未来几个季度增长区间的下限？尤其是考虑到我们的广告加载率相对同行仍然较低。
James Gordon Mitchell
Alicia, so let me take a stab at both of those. With our business services revenue growth, we did see more revenue related to GPI (sic) \[ GPU ] and API token rental. But we also saw broad-based growth. We feel that in the last 2 to 3 years, we've effectively debugged our cloud business by scaling back and sometimes exiting from low margin and low value-add activities. And now that we've put our cloud business onto a more sustainable base as well as improve the cost competitiveness of the supply chain for our cloud business, we can -- we are refocusing on growing revenue at an accelerated rate versus the prior rate without depending too much on the vagaries of the GPU supply situation.
Alicia，我来回答这两个问题。关于企业服务收入增长，我们确实看到 GPU 和 API token 租赁带来的更多收入，但也有广泛的增长。在过去两到三年里，我们通过缩减甚至退出一些低利润、低附加值业务，实际上对云业务进行了“调试”。现在，我们已经把云业务放在一个更可持续的基础上，并改善了供应链的成本竞争力，因此我们可以重新聚焦，实现比以往更快的收入增长，而不需要过度依赖 GPU 供应的不确定性。
So if we do have sufficient GPUs that we can rent out more in the cloud, then we'll do so. But our cloud strategy is not dependent on the GPUs. We're also growing in CPU, in storage, in database, in CDN and so forth. So that's on the cloud side.
如果我们有足够的 GPU，我们当然会在云业务中租赁更多 GPU。但我们的云战略并不依赖于 GPU。我们在 CPU、存储、数据库、CDN 等方面也在增长。这是关于云业务的部分。
In terms of your question around advertising, I'd say that we're comfortable with the growth rates that we're delivering. There's no change in our philosophy at this point in time, and we're focused on really elongating the runway for growth. Now of course, if the cost of deploying AI, including GPU depreciation was suddenly to step up and become very burdensome, we could accelerate the advertising monetization, but we don't see the need to do that right now. Thank you.
至于广告的问题，我认为我们对当前的增长率感到满意，目前我们的理念没有变化，我们关注的重点是延长广告收入的增长跑道。当然，如果 AI 部署的成本（包括 GPU 折旧）突然上升，变得非常沉重，我们可能会加快广告变现，但目前我们认为没有必要这样做。谢谢。
比较现实又留有余力，在短视频领域已经在争抢几个流量的明星，有点像以前的电影产业。
Wendy Huang
We will take the next question from Ronald Keung from Goldman.
我们下面请 Goldman 的 Ronald Keung 提问。
Ronald Keung
Maybe a first question on games. With this expanding evergreen game titles, and we also see a very strong pipeline. So is it fair to say we are moving towards an overall platform now with increased stability, decreased single title volatility. So how should we think about a normalized growth rate for the overall gaming business from here, whether it's from a market share or number of gamers or ARPU potentials?
第一个问题是关于游戏的。随着越来越多的长青游戏，以及强劲的产品管线，是否可以认为我们正在逐渐迈向一个整体平台，具备更高稳定性，单一游戏波动性降低？那么我们该如何看待游戏业务未来的正常化增长率？无论是从市场份额、玩家数量，还是 ARPU 潜力角度。
And then a second question on just AI models. I want to hear on what are the next priorities for the HunYuan model family, especially on -- we're seeing some closing of gaps between Chinese models versus state-of-the-art global leading models. So with this improvement, we see kind of increased emphasis now from kind of multimodal, we mentioned about 3D and agent capabilities. I want to hear what are we thinking about the agent potentials, particularly when you enter into WeChat.
第二个问题是关于 AI 模型。我想了解“混元”模型家族的下一步重点，特别是考虑到中国模型和全球领先的尖端模型之间的差距正在缩小。在这种提升下，多模态、3D、Agent 能力正受到更多关注。那么当“混元”进入微信时，我们怎么看 Agent 的潜力？
James Gordon Mitchell
Ronald, thank you for the question. So on games, I think your broad observation is correct that as our game portfolio broadens out and becomes more platform-like in nature, then we should expect less volatility in the overall game revenue growth. Now of course, each game is still a creative endeavor and individual games may have stronger seasons and weaker seasons. But what we can see is that as long as the games are sort of best-in-class and we nurture and support them, then they can tolerate a weaker season, the development team makes some changes and then the users and the revenue bounce back in a stronger season.
谢谢你，Ronald 的问题。关于游戏，我认为你的整体观察是正确的。随着我们的游戏组合不断扩大，并越来越具有平台化特征，整体游戏收入的波动性应当会下降。当然，每款游戏仍然是创意驱动的作品，个别游戏可能会有强势和弱势的周期。但我们看到，只要这些游戏本身是同类最佳，并且我们不断培育和支持它们，它们就能够度过低谷期，开发团队做出一些调整，用户和收入便会在下一个旺季反弹。
And sometimes with a game like Fight for the Golden Spatula, that quieter season might be 2 months. But other times, if you look at the Supercell games, then Brawl Stars went somewhat quiet for 4 years before having a big recovery 18 months ago. And Clash Royale actually was rather quiet for 7 years and now is performing very strongly. So in each of those cases, we feel that each of those games is the best game in the market at what it does and we provided it with the best support.
有时候，像《Fight for the Golden Spatula》这样的游戏，淡季可能只有两个月。但也有像 Supercell 的《Brawl Stars》，它沉寂了 4 年，直到 18 个月前迎来强劲复苏。《Clash Royale》实际上安静了 7 年，如今表现也非常强劲。在这些案例中，我们始终认为这些游戏在各自领域都是最优秀的，并且我们也给予了最好的支持。
And so even if there's a sort of creative misfire for whatever reason and the revenue is weaker for a season or weaker for a year or weaker for 7 years, we have the potential to reinvigorate them and bring them back to growth. And of course, as you observed, we periodically release new games, some of which may themselves become evergreen in time, which broadens out the portfolio and strengthens the platform.
因此，即便某款游戏因为创意上的失误，导致一个季度、一年甚至七年的收入表现疲弱，我们依然有潜力让它重新焕发生机，恢复增长。当然，正如你所观察到的，我们也会定期推出新游戏，其中一些最终可能会成为新的长青作品，从而拓宽组合，强化平台。
Chi Ping Lau
In terms of the model, I would say there's actually a lot to be done, right? And I would say sort of in the broad bucket, there is the large language model itself, and we want to keep improving the LLM itself. And that actually involves improvement along a number of different dimensions, including making sort of the data sort of higher quality and more comprehensive. That includes making the pretraining more efficient and more effective and improving the pretraining model that includes improving the post-training and reinforced learning processes in basically extracting the capability of the pretrained model and that includes improving our infrastructure so that we can actually train more efficiently as well as inference more efficiently, right?
关于模型，我认为还有很多工作要做。总体来看，第一类是大语言模型本身，我们希望不断提升 LLM。这涉及多个维度的改进，包括提升数据的质量与全面性；让预训练更高效、更有效；改进预训练模型；优化后训练和强化学习过程，以更好地挖掘预训练模型的能力；以及改进基础设施，使我们能够更高效地完成训练与推理。
And then when we can inference more efficiently, then we can make use of the GPUs better. We can lower the cost of providing the services to our users and customers. And when we have an improved LLM, it's actually sort of the foundation for all our AI services. And in particular, it would improve our search and productivity-related services. And then, of course, we also want to improve the multimodal capability of our model so that we can actually provide more customized functions for the users in Yuanbao, right? Within Yuanbao, it's not -- people are not just using it for search and productivity-related activities. They are using it for all kinds of different multimodal activities.
当推理效率提升后，我们就能更好地利用 GPU，降低向用户和客户提供服务的成本。更优的 LLM 实际上是所有 AI 服务的基础，尤其能改进搜索和生产力相关的服务。当然，我们还希望提升模型的多模态能力，以便在“元宝”中为用户提供更多定制化功能。毕竟用户在“元宝”中不仅仅用它来搜索或做生产力相关的事情，他们还会进行各种多模态活动。
They may want to speak, they may want to turn text into pictures, turning pictures into text and there are a lot of multimodal conversions within Yuanbao, which we actually need to have very strong capability for.
用户可能希望语音输入，或是将文字转换为图像、将图像转换为文字。在“元宝”中存在大量此类多模态转换，我们需要具备非常强大的能力来支持。
And then I think the third broad category is actually coding and agents, right? So that if we can sort of keep improving, then basically, we can provide much better coding environment for both ourselves as well as our enterprise customers. And at the same time, that would enable a better agent and instruction follow capability for our agent. I think that's particularly important for Weixin going forward and as we build an agent for Weixin that can be personalized assistant to the Weixin users in a personalized way. So we are actually improving our capability across all these different dimensions.
第三类是编码和 Agent。如果我们能持续改进，就能为我们自己和企业客户提供更好的编程环境。同时，这也将提升 Agent 的能力和对指令的执行效果。我认为这对微信的未来尤为重要，因为我们正在为微信打造一个 Agent，使其能以个性化方式成为用户的专属助手。所以，我们正在各个维度全面提升能力。
Wendy Huang
We will take the next question from Charlene Liu from HSBC.
我们下面请汇丰银行的 Charlene Liu 提问。
Charlene Liu
I wanted to ask about the government's anti-involution efforts. In fact, I mean, these efforts should reduce competition, but probably accelerate consolidation of certain sectors. What are your observations so far? Do you expect these changes to impact or how do you expect these changes to impact your main business lines in the short and long term? Also separately, we see commercial payment volume growth turned positive in the second quarter. Can we expect that growth to sustain or even accelerate?
我想问一下关于政府“反内卷”的举措。理论上，这些措施应当减少竞争，但可能加速部分行业的整合。到目前为止，你们的观察是什么？你们预期这些变化会对主营业务在短期和长期产生怎样的影响？另外，我们看到二季度商业支付交易量增长转正，这一趋势是否能够持续甚至加快？
Chi Ping Lau
So on that question, I think your 2 questions are somewhat related, right? The biggest impact that we have seen is that our commercial payment actually sort of has seen better trends and the better trend is actually driven by the less decrease in the check size. We have always said the commercial payment is a function of number of transactions times the per transaction ticket size and the number of transactions have been increasing healthily, but then the ticket size has been decreasing. And the improvement that we have seen is actually when the ticket size actually decreased less. And I think part of the driver of that is actually the anti-involution initiative by the government.
关于这个问题，我认为你的两个问题其实是相关联的。我们观察到的最大影响是在商业支付方面，趋势有所改善，而这种改善主要是因为单笔交易金额的下降幅度减小。我们一直强调，商业支付 = 交易笔数 × 单笔交易金额。交易笔数在健康增长，但单笔金额此前一直在下降。我们看到的改善，是单笔金额下降得更少了。我认为部分原因正是政府的“反内卷”举措所带来的。
Wendy Huang
We will take the next question from Yuan Liao from CITIC Securities.
我们下面请中信证券的廖远提问。
Yuan Liao
My question is about your gaming business. And we see that FPS has always been Tencent's strength. And recently, we see great success across many titles such as Delta Force, Peacekeeper and VALORANT. And we would like to ask management what you believe are the key factors behind this success.
我的问题是关于你们的游戏业务。我们看到 FPS 一直是腾讯的强项，最近《Delta Force》、《Peacekeeper》和《VALORANT》等作品都取得了巨大成功。管理层认为这些成功背后的关键因素是什么？
Additionally, looking ahead, we see some upcoming release of the FPS gaming such as VALORANT mobile games and also some FPS title, Rainbow Six Siege and Crossfire: Rainbow. So how does management view the differentiation among these titles? And could they compete each other?
另外，展望未来，我们看到一些即将上线的 FPS 游戏，比如《VALORANT Mobile》、以及《Rainbow Six Siege》和《Crossfire: Rainbow》等。管理层如何看待这些作品之间的差异化？它们之间会不会相互竞争？
And lastly, besides FPS, so what other game categories you will invest more in the future?
最后，除了 FPS，你们未来还会在哪些游戏品类加大投资？
James Gordon Mitchell
So thank you for the questions. And yes, you're right. It is not necessarily intuitive that VALORANT, Delta Force, Peacekeeper Elite would all be driving at the same time. And in fact, Call of Duty: Mobile, Crossfire, Arena Breakout are also doing well despite all of them fitting within the broad tagline of first or third-person shooter games. And we believe that the key factor is that this category is the dominant gaming category in the Western world, but it was historically underrepresented in the China market.
感谢你的问题。是的，你说得对，像《VALORANT》、《Delta Force》和《Peacekeeper Elite》同时都表现强劲，这并非直觉上的结果。事实上，《Call of Duty: Mobile》、《Crossfire》和《Arena Breakout》也同样表现良好，尽管它们都属于第一人称或第三人称射击类的“大类”。我们认为关键因素在于，这一品类在西方世界是主导性的游戏类别，而在中国市场却在历史上一直供给不足。
But now every year, there's another 10 million students going to college in China, receiving laptops as part of their college materials, then some proportion of them discover that they enjoy playing first-person shooter games. And you can see that in the statistics for sales of special keyboard and mice for those games in China that have been surging. And so in that sense, there's a sort of a structural demographic shift working in favor of these games as technology brings down latency as we get better at combating cheating and as the graphics quality of these games improves.
但现在，中国每年有约 1,000 万学生进入大学，他们会收到笔记本电脑作为学习配备，其中一部分人逐渐发现自己喜欢玩第一人称射击游戏。中国市场上专用键盘和鼠标的销量激增也印证了这一点。从这个角度看，人口结构的变化正在推动这一类游戏的发展；再加上技术进步降低了延迟、反作弊手段更强、画面质量不断提升，都使得这些游戏更受欢迎。
In terms of how the games differentiate, then in the Western world, Call of Duty, Halo, Fortnite are all shooter games, but in each of them is very successful. None of them cannibalize upon each other. And that's what we've seen so far, and we expect to continue to see in China because there actually is a great deal of differentiation within the sort of super genre. There's differentiation by mode, obviously, between Battle Royale versus extraction shooter versus tactical versus arcade. There's differentiation by character between hero-based games versus class-based games. There's differentiation based on time to eliminate if you have a fast time to eliminate game and positioning is key. If you have a slow time to eliminate game, then team play is key. There's differentiation in graphics between realistic versus animated graphics and so forth.
至于这些游戏之间如何实现差异化，在西方世界，《Call of Duty》、《Halo》、《Fortnite》都是射击类游戏，但它们各自都非常成功，互不蚕食。这也是我们至今在中国所看到的情况，并且我们预计这种情况会持续下去。因为在这一“超级品类”内部存在大量差异化：在模式上有“大逃杀”、“撤离射击”、“战术射击”、“街机射击”等区别；在角色设定上有基于英雄的和基于职业的不同；在“击杀所需时间”上，有的游戏节奏快，位置操作是关键；有的游戏节奏慢，团队配合是关键；在画风上，有写实与动画的差异，等等。
So we think that this is a very large and rapidly growing genre. It's one where we're the clear market leader with the most experience, the most skills in China. And we look forward to bringing Rainbow Six and VALORANT Mobile and other games within the genre to market and complementing the games that are already there, just as VALORANT PC and Delta Force complement Peacekeeper Elite and the rest.
因此，我们认为这是一个非常庞大且快速增长的品类。在中国，我们是拥有最多经验和技能的市场领导者。我们期待将《Rainbow Six》、《VALORANT Mobile》等更多同类游戏推向市场，以补充现有作品，就像《VALORANT》PC 版和《Delta Force》补充了《Peacekeeper Elite》等一样。
In terms of other categories of games, we're investing on a broad basis for games. I would say that historically, we've been somewhat underrepresented in content-centric games. But today, we have Naruto, we have Wuthering Waves, both internal studio games. We also have Nikke. Each of them are 1 or 2 in their respective category in China. Wuthering Waves and Nikke are performing very well internationally as well. So that's an important growth area that targets a very different user base from the classic user base for the first- person shooter games.
在其他游戏品类方面，我们的投资布局是广泛的。历史上，我们在“内容驱动型游戏”上的代表性不够强。但现在，我们拥有《Naruto》、《Wuthering Waves》这类内部工作室开发的作品，还有《Nikke》。它们在中国各自所属品类中都位居前二，《Wuthering Waves》和《Nikke》在海外市场也表现出色。这是一个重要的增长领域，目标用户群体与传统射击类游戏的核心用户群体完全不同。
And I think one not intuitive aspect -- or another not intuitive aspect of the development of the game industry in China is with most industries, you start with your highest-yielding users and then over time, you expand the audience into lower revenue per user audience. But actually, the game market in China is not like that. If you look at some of these new cutting-edge tactical FPS games like VALORANT, then they actually deliver substantially higher revenue per DAU than our big incumbent games. Similarly, if you look at the big new content games, whether it's a Wuthering Waves or Love and Deepspace, they deliver substantially higher, multiple times higher revenue per DAU than the big existing games. So as we sort of develop these new emerging audiences, we also have the benefit of higher revenue per time in some of these audiences.
我认为中国游戏产业发展中还有一个“非直观”的现象。在大多数行业，企业往往从高价值用户开始，逐渐扩展到低客单价用户。但中国的游戏市场并非如此。如果你看一些新兴的战术类 FPS 游戏，比如《VALORANT》，它们的 DAU 收入实际上远高于我们的大型存量游戏。同样，在一些新兴内容游戏中，无论是《Wuthering Waves》还是《Love and Deepspace》，它们的 DAU 收入也是现有大型游戏的数倍。因此，当我们拓展这些新兴用户群体时，我们不仅扩大了受众规模，同时还享有更高的用户单次价值。
Wendy Huang
We will take the next question from William Packer from BNP.
我们下面请 BNP 的 William Packer 提问。
William Henry Packer
Congrats on the strong quarter. At your last earnings update, you shared some expected levels of CapEx spend with chip spending a clear strategic priority in the context of the AI opportunity. Since May, we've had a huge amount of news flow around chip restrictions. Could you help us think about your chip sourcing options in the context of new rules and the consequent impacts on your CapEx spend and follow through on group margins?
祝贺公司本季度取得强劲表现。在上次财报中，你们分享了预计的资本开支水平，其中芯片投入被明确列为 AI 机会背景下的战略重点。但自 5 月以来，关于芯片限制的新闻很多。能否请管理层谈谈在新规之下芯片采购的选择，以及这对资本开支和集团利润率的影响？
And then my follow-up is around the impressive quarter for gross margins in the VAS segment following the recent Epic decisions. Should we think that those changes in App store take rates in your international business are a positive driver for Q2? Or is that a source of upside in the future? And any progress you're having with App store partners in China would be interesting, too.
我的追问是关于 VAS 板块的毛利率。在 Epic 最新判决之后，本季度毛利率表现突出。我们是否应当认为国际业务中应用商店抽成比例的变化是二季度的正向驱动？还是未来的潜在利好？在中国与应用商店合作伙伴的进展也请分享一下。
Chi Ping Lau
Yes, I'll answer the first question. Well, with respect to the acquisition of chips, especially the U.S. chips, right, the answer is that we don't really have a definitive answer on the import situation yet. I think there's a lot of discussion between the 2 governments, right, and waiting to see what exactly come out of that. But from our own perspective, we do have enough chips for training and continuous upgrade of our existing models. And we also have many options for inference chips. And we are also executing a lot of software improvement and upgrade in order to drive efficiency gain in inference so that we can actually put more workload on the same number of chips.
我来回答第一个问题。关于芯片采购，特别是美国芯片，目前我们对进口情况还没有明确答案。两国政府之间仍有很多讨论，我们也在等待最终结果。但从我们自身情况看，我们拥有足够的芯片来支撑训练和现有模型的持续升级。同时，我们在推理芯片方面有很多选择。我们也在进行大量软件改进与升级，以提升推理效率，从而让同样数量的芯片承载更多工作负载。
So with respect to our CapEx target, we have not revised our full year CapEx target yet. And when it talks about sort of what's the impact on group margins, I would say the depreciation cost related to AI will definitely continue to go up. But at the same time, we also see that we continue to reap the benefits of AI. And the issue is that these 2 may not match each other completely, but I think both of them will be moving in the same general direction.
因此，在资本开支目标方面，我们目前尚未调整全年目标。至于对集团利润率的影响，我认为与 AI 相关的折旧成本肯定会继续上升。但与此同时，我们也在不断收获 AI 带来的效益。这两者未必完全匹配，但我认为它们总体上是朝着相同方向发展的。
James Gordon Mitchell
And in terms of your second question, Will, then the impact of some of the court rulings around App store fees was not really a driver for our VAS gross margin in the second quarter. We have been ramping up the usage of link out payments in the United States, where the court order took effect relatively sooner. But in a number of other jurisdictions, those court orders have only been flowing through in the third quarter, including actually the Australian jurisdiction overnight, I think. And also, we have the accrual accounting for our game revenue and cost of revenue, which means it would take time to flow through the P\&L.
至于你的第二个问题，Will，关于应用商店抽成的法院裁决，其实并不是推动二季度 VAS 毛利率的主要因素。在美国，这些裁决较早生效，我们已经在加快“跳转支付”的应用。但在其他司法辖区，裁决大多要到三季度才开始逐步落地，包括澳大利亚刚刚在昨晚才生效。另外，由于我们在游戏收入和成本上采用权责发生制，会计确认需要时间才能在损益表中完全体现。
In China, then our Android App Store revenue splits have been moving in what we think is a much more rational direction and one that favors developers and therefore, overall usage of the software and the hardware over time. So that has been a gentle tailwind for our China business. But I think if you look at the second quarter specifically, then the mix shift from licensed games towards self- developed games, in particular Delta Force was a bigger positive contributor to the VAS gross margin improvement.
在中国，我们与 Android 应用商店的分成安排正在朝着更加合理、对开发者更有利的方向发展，这从长期来看也会推动软件和硬件的整体使用。这对我们的中国业务来说是个温和的顺风因素。但如果单看二季度，VAS 毛利率改善的更大推动力来自业务结构变化——尤其是《Delta Force》在内的自研游戏收入占比上升，而非分成政策的变化。
Wendy Huang
We will take the last question from Ellie Jiang from Macquarie.
我们下面请 Macquarie 的 Ellie Jiang 提最后一个问题。
Ellie Jiang
So I have a question on the overall AI strategy. The management has shared a lot of the insights into us kind of exploring into different dimensions of AI integrations. Just in terms of the KPIs we're utilizing to track our AI development progress from the model layer into the application, Agentic AI layer, how do we really tackle the existing challenges, especially on the supply side? I guess that's the first part of the question.
我的问题是关于整体 AI 战略的。管理层已经分享了很多在不同维度整合 AI 的探索。那么从 KPI 的角度，我们是如何跟踪 AI 从模型层到应用层、再到 Agentic AI 层的发展进展？尤其是如何应对供应端的挑战？这是第一个问题。
And then as a follow-up to the prior question on the AI investment front, how would we really assess the potential M\&A opportunities in the market right now, considering there are actually a surge in volume of smaller kind of AI -- Agentic AI ventures that's been happening. So I guess that's the second question.
第二个问题，延续之前关于 AI 投资的提问，现在市场上涌现出大量较小型的 AI 或 Agentic AI 创业公司。在这种背景下，你们会如何评估潜在的并购机会？
Chi Ping Lau
In terms of the first question, I would say we do track our AI development progress very closely. And I think there are a number of indicators that we use right in tracking the progress. And the first one is that we focus on tracking how AI is actually helping our existing businesses such as ads, such as games, such as FinTech. And I think that's one area. And when we see that AI is actually being applied in driving the efficiency gain as well as the growth of these businesses, then that's good. Secondly, we focus on tracking the performance and quality of our large language model, HunYuan. And I think there's a lot of metrics that we actually have to use in order to track the capability as well as the quality of the model.
关于第一个问题，我要说我们确实非常密切地跟踪 AI 的发展进展。我们有几个主要的衡量指标。首先，我们关注 AI 如何在广告、游戏、金融科技等现有业务中发挥作用。当我们看到 AI 推动效率提升并促进这些业务增长时，这就是积极的信号。其次，我们关注自家大语言模型“混元”的性能与质量，这需要通过大量指标来衡量其能力和效果。
The third one is we do track how our AI app is actually growing. How many users are using our AI app. And that would include users of our Yuanbao and users of our browser and user of our AI-powered search. And finally, I would say we do track what's the progress in the design of other AI-related innovative products within our entire ecosystem. And that would include, for example, the AI agent for WeChat that would include agents within our productivity tools. And these are the metrics that I think we will use in terms of tracking the progress of our AI development.
第三，我们跟踪 AI 应用的用户增长情况，包括“元宝”、浏览器以及 AI 搜索的用户数。最后，我们也会跟踪整个生态体系内其他 AI 创新产品的设计进展，例如微信的 AI Agent，以及我们生产力工具中的 Agent。这些指标共同构成了我们衡量 AI 发展进展的体系。
And when you talk about sort of challenges from supply, I think I've answered largely the question that we do have enough chips for doing the training and continuous upgrade of our foundation model. And at the same time, we are actually using all sorts of different ways to both get inference done using different chips. And at the same time, we are also improving our entire infra and software capability in order to squeeze more performance out of the AI chips that we have for inference.
至于供应端的挑战，我之前也提到过，我们目前有足够的芯片来进行训练和基础模型的持续升级。同时，我们在推理上尝试使用不同类型的芯片，并通过改进基础设施和软件能力，从现有 AI 芯片中榨取更多性能，以提升推理效率。
Wendy Huang
Thanks you. We are now ending the webinar. Thank you all for joining our results webinar. 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.