Kuaishou Technology B

Good day, ladies and gentlemen. Thank you for standing by. Welcome to the Kuaishou Technology Third Quarter 2025 Financial Results Conference Call. Please note that English simultaneous interpretation will be provided for management's prepared remarks. This English line will be in listen-only mode. I will now turn the call over to Mr. Matthew Zhao, VP of Capital Markets and IR at Kuaishou Technology. Thank you, operator. Good evening and good morning to everyone. Welcome to Kuaishou Technology Third Quarter 2025 Financial Results Conference Call. Joining us today are Mr. Cheng Yi Xiao, co-founder, chairman, and CEO, Mr. Jin Bing, our CFO. Before we start, please note that today's discussion may contain forward-looking statements which involve a number of risks and uncertainties. Actual results and outcomes may be different from those discussed. The company does not undertake any obligation to update any forward-looking information except as recorded by law. For important information about this call, including forward-looking statements, please refer to the company's public information or third quarter 2025 results announcement added September 30, 2025, issued earlier today. During today's call, management will also discuss certain non-FRS measures. These are provided for additional information and should not replace FRS-based financial results. For a definition of non-FRS financial measures and reconciliation of RRS to non-FRS financial results and related risk factors, please refer to the third quarter 2025 resource announcement. For today's call, management will use Chinese as the main language. A third-party interpreter will provide a simultaneous English interpretation in the pre-programmed RAC session and a consecutive interpretation during the Q&A session. Please note that English interpretation is for convenience purposes only. In case of any discrepancy, management or original language will prevail. Lastly, unless otherwise stated, all currency units mentioned are in RMB. Now, I'll turn the call over to Yixiao. Hello, everyone. Welcome to Kuaishou's third quarter 2025 earnings conference call. In Q3, we continue to advance our AI strategy, expanding scenario-based AI applications and innovative use cases across our business. These efforts created a tangible business value across all business scenarios, strengthened the quality efficiency for our organizational infrastructure, and fueled the strong operational financial results. Average DA use on the casual app has surpassed $416 million in Q3, marking the third consecutive quarter of record highs. Total revenue for Q3 rose by 14.2% a year over year to Revenue from our core commercial businesses, online marketing services, and other services, primarily e-commerce, increased by 19.2% year-over-year. Adjusted net profit rose 26.3% year-over-year to 5 billion RMB, with an adjusted net margin of 14% per year. We achieved a year-over-year growth in the group's overall profitability while continuing to invest strategically in AI, a catalyst for unlocking deeper value across our content and business ecosystems. First, our AI strategy and the progress of our large video generation model, Cling AI. We continue to refine the foundation models behind Cling AI, developing new features to meet creators' diverse needs and build a one-stop creative productivity platform that empowers everyone. to tell captivating stories with AI. In Q3, we launched a Cling Lab and upgraded the start and end frames function and introduced digital human solution. Notably, at the end of September, we released the Cling AI 2.5 model, achieving substantial advances in prompt adherence, dynamic effects, style consistency, and visual aesthetics. Just 10 days after launch, the model was simultaneously ranked as the world's number one text-to-video and image-to-video a model by artificialanalysis.ai, independent AI benchmarking platform. While maintaining its leading content generation performance, the new model also integrates continuous engineering innovations that lower video inference cost, reducing creators per video generation expense by almost 30%, and further strengthening Cling AI's cost efficiency advantages. Cling AI's innovations and foundation models and product features have provided creators with higher quality video generation solutions establishing a foundation for broader adoption across professional creative fields such as marketing, e-commerce, film, and television, short plays, animation, and gaming. As Cling AI continues to expand its use cases, it has made breakthroughs in monetization and revenue growth. In Q3, revenue from Cling AI exceeded 300 million RMB. Cling AI is committed to empowering global creators and building a premium ecosystem. In September, we launched the Cling AI NextGen Creative Contest, which received over 4,600 entries from 122 countries and regions worldwide covering diverse fields such as history, science fiction, and animation. Outstanding works were screened at international film festivals including Cannes, Tokyo, and Busan for the integrating AI-powered film and TV works with traditional film and TV industries. In Q3, we achieved strong results from integrating AI into diverse internal and external use cases On business empowerment, large AI models have now been integrated across all of Questro's major business areas, driving incremental value across our ecosystem. We iterated our end-to-end generative recommendation large model, One Rec, and extended it beyond a short video recommendation to additional recommendation areas, such as online marketing services and e-commerce shopping malls. This expansion has generated meaningful incremental benefits. In Q3, large AI models demonstrated notable effects, especially in online marketing services. We pioneered a generative reinforcement learning-based bidding model that integrates sequence modeling with a goal optimization. This innovation transformed advertising from a single-step decision-making to long-term strategic planning, significantly enhancing bidding capabilities and ROI for clients, especially for small and medium-sized ones. Meanwhile, we explored using end-to-end generative recommendation in online marketing services scenarios through OneRack. Tailored to the characteristics of online marketing services, we introduced the client marketing expression and marketing commercial value perception mechanism to achieve bidirectional matching between users' interests and clients' demands, enhancing personalization and matching efficiency. Large AI model technologies, especially One Rack, drove roughly 4% to 5% growth in domestic online marketing services revenue in Q3. In terms of online marketing material generation, Cling AI's large model has significantly reduced video production costs for clients. Meanwhile, advanced digital human technology has also opened up new operational scenarios in live streaming for both online marketing clients and e-commerce merchants. Consequently, the total spending from online marketing services driven by AIGC marketing materials exceeded 3 billion RMB in Q3. For e-commerce, we launched OneSearch, an end-to-end generative retrieval architecture. It enables more precise product matching and optimizes the user experience, driving nearly 5% growth in shopping mall search order volume. The adoption of OneRack in e-commerce also contributed to high single-digit GMV growth in the shopping mall feed in Q3. For entertainment live streaming, we leveraged CoinAI to introduce the AI Universe gift customization feature which generates highly personalized avatar-based personal gifts, increasing both user engagement and willingness to pay. Second, user growth and content ecosystem. In Q3, average DAUs on the Kuaishou app reached 416 million and MAUs reached 731 million. This is the third consecutive quarter that average DAUs reached a record high. The sustained and steady traffic growth reflects the Crashers community's unique appeal to users. By refining our user growth strategies, offering distinctive and diverse content, optimizing our traffic allocation mechanism, and enhancing community engagement, we continue to reinforce Crashers' identity as a heartwarming, diversified, informative, and engaging online community. In Q3, average daily time spent per DAU on the Crashers app was 134.1 minutes, while total user time spent rose by 3.6% year-over-year. Our refined user growth strategies leverage smart marketing material placement on acquisition efficiency, lowering the acquisition cost per new user year over year. In traffic allocation, by modeling users' long-term user interaction patterns, we improve it about the user satisfaction and retention. We also continue to upgrade users' sharing experience within private messaging and iterated on social interaction features. As a result, the daily average penetration rate of private messages among users with mutual followers increased by more than 3 percentage points year-over-year. We also elevated the user product experience through a series of device-level intelligent optimizations. In content operations, we partnered with the Beijing Radio and Television Station to launch the 2025 Kuaishou Super Summer Gala, where celebrities and everyday users come together and celebrate. The live stream session attracted a peak over 5.4 million concurrent users. To cater to young audiences, we hosted an online concert featuring DNA, which drew 980 million live streaming views. In the Pan-Knowledge category, we created the Liyuan Music Festival Summer Tour Series, showcasing offline tours across diverse traditional art forms, such as Qinjiang and also Xianbei Shushu. By bringing these live performances to audiences, we help to benchmark creators like I want to achieve cumulative creative breakthroughs and gain recognition. Third, online marketing services. In Q3, revenue from our online marketing services reached 20.1 billion RMB, up 14% over year. With the growth rate accelerating quarter over quarter, we continuously iterated and upgraded our online marketing placement and products with AI models. Drawing our unique traffic dynamics, we cater to the needs of more marketing customers through our smart placement capabilities, achieving more precise targeting and higher conversion rates. This drove strong year-over-year growth in both external and closed-loop marketing services and revenue. In Q3, our UX solutions accounted for over 70% of external marketing spending. Ongoing innovations, iterations, particularly with our generative and reinforcement learning-based feeding model and generative A recommendation large model further improved marketing recommendation efficiency and enhanced management of marketing variety and value. The combination of our three key AIGC commercialization tools, AIGC short video, digital human, and digital employee has empowered our customers with an end-to-end AI solution covering marketing material creation, live streaming operations, and user engagement. In Q3, for a closed-loop e-commerce marketing services, we upgraded the product and content optimization capabilities of our Omni domain. platform marketing solutions to mean a steady supply of premium marketing materials. By integrating multi-content reimbursement and ROI bidding recommendation tools, we help e-commerce merchants improve traffic and sales conversions thereby enhancing their willingness to invest in marketing placement. In Q3, total marketing spending from Omni platform marketing solution accounted for over 65% of its closed-loop marketing spending. Additionally, we established a bidding agent based on AI capability to replace mutual menu adjustment decisions, enabling more consistent conversions and allowing greater economies to go. On the traffic side, by enhancing the synergy between e-commerce and commercial value, we released more traffic capacity to merchants with long-term operations, helping more brand e-commerce merchants achieve a scaled expansion and stable conversion improvements. From a scenario perspective, in Q3, closed-loop e-commerce marketing services in pen-shelf-based scenarios also realized a solid growth. We optimized the people-to-goods matching in pen-shelf scenarios, and we used large models to better meet a user's needs and improve efficiency. These efforts increased marketing placement and penetration and drove stronger merchant participation. In Q3, for the lifestyle service sector, where clients mainly operated on a lead-based model, we upgraded our private messaging products and optimized vertical-oriented products. These improvements helped clients reach users more efficiently and achieve higher user conversion rates across various conversion goals. In lifestyle services, particularly among our small and medium-sized customers, we improved private messenger response rates with AI-powered customer service. In Q3, we combined our local services with a lead-based marketing business to form our lifestyle service segment, iterating teams, product lines, and traffic distribution. This unification strengthens our ability to support merchants pursuing sustainable operations and help build a more diversified collaborative ecosystem with local merchants. These three The content consumption sector led by Shorter Plays was another key revenue driver for our external marketing services in Q3. We continued to enhance content supply and product innovation across Shorter Plays minigames and novels while capturing incremental growth opportunities from the rapid rise of comic-style Shorter Plays, further expanding external marketing services revenue. Comic-style short plays combine features of comics, short plays, and audio dramas, typically featuring vertical screen episodes to one to three minutes long. This new genre has recently gained an investor attraction among the broader market. Cling AI has significantly lowered the barrier to creating comic-style short plays while elevating overall content quality. In addition, through a mix of marketing placement, revenue sharing, IA, and IAP models, we created multiple monetization pathways for high-quality short play content, expanding reach on both the supply and demand side. Fourth, our e-commerce business. In Q3, our e-commerce GMV grew 15.2% a year to 385 billion RMB. Through a mix of merchant incentive programs, Omni Domain traffic support, and intelligent tool empowerment, we helped merchants build Omni Domain operations ecosystems, continuously elevating user experience and driving high quality supply and demand growth to support emerging sustainable growth we encourage them to adopt an efficient conversion path that integrates public and private domains using public domains to acquire customers in private domains to strengthen retention in q3 the mix of our e-commerce monthly average paying users to show healthy trends active e-commerce users repeat purchase frequency increase every year and user stickiness continue to improve In Q3, in e-commerce supply, building on our platform's traffic and content-based e-commerce advantages, we continue to attract new merchants organically and onboard merchants through diverse channels. We introduced a range of incentives to lower onboarding costs and entry barriers for new merchants. In addition, we continue to launch initiatives to empower new merchants to navigate early growth stages and ramp up operations more efficiently. Driven by a growing number of small and medium-sized merchants, together with our targeted support for high-quality existing merchants, our average monthly active merchant base continues to grow. We also broaden the range of products. The number of Level 3 product categories per store among our average monthly active merchants is increasing by nearly 30% year-over-year. To empower merchants of KOLs, in Q3, we launched a series of initiatives to unlock greater value creation within their private domains, supporting their ability to build a dual growth engine of exceptional content and superior products. We launched the pop-up follower rewards product to accelerate follower growth and empower merchants of KOLs from traffic generation to follower conversion, ultimately to sales. With a stronger control over merchandise selection and supply, we expanded our product portfolio for high-quality platform offerings. We focused on the premium brands through our KOL blockbuster initiative, leveraging the traffic pool of hit product to spotlight dedicated live streaming sessions for treasury brands, supported by improved KOL product matching, KOL targeted vertical outreach, and platform incentives. We expanded the KOL engagement, enhanced brand performance, and empowered KOLs to address product selection and assortment expansion challenges. In Q3, the average daily number of active merchandise items increased by over 30% every year, We provided a guaranteed resources such as traffic support and a product supply to onboard a small and medium sized KOLs and establish long-term growth mechanisms. These efforts to strengthen the KOL content ecosystem in Q3 driving a 14.8% a year over year increase in the number of average daily active streamers hosting live sessions with over 10,000 followers. In Q3, in terms of operating across diverse scenarios. PenShelf for the e-commerce GMV continued to outpace overall GMV growth, contributing over 32% of the total e-commerce GMV. We continue to enhance our infrastructure and supply ecosystem, driving a 13% year-over-year increase in average daily active merchants for PenShelf-based e-commerce. We built on the diverse engagement features, strategy tools from Q2, including Super Links, the official channel platform recommending products. These tools help the merchants quickly boost product exposure and sales conversion. The marketing hosted tool we introduced for merchants in content-based scenarios effectively lowered their operational barriers and drove steady quarter-over-quarter growth in merchant adoption. In Q3, we've maximized the synergies between short videos and live streaming. We helped merchants integrate traffic from content-based scenarios through a seamless loop from product recommendations via short videos to rapid conversion in live streaming rooms and back to user engagement via short videos. This strategy steadily expanded the merchant's customer base, supported by more short videos with embedded shopping links and our customized funnels, short video e-commerce chain will maintain healthy growth. In Q3, in terms of integrating AI into our e-commerce business, we focus on empowering merchants across our e-commerce business chain with three core areas. AIG's content and production, merchant efficiency improvement, and product matching efficiency optimization are AIG's capabilities for generating and optimizing materials continue to deliver strong results, helping merchants improve conversion efficiency across both image and video formats in diverse scenarios. Penetration of the smart live streaming highlights and AI live stream scenarios also steadily increased. Concurrently, our AI product management assistant is providing comprehensive Omni scenario support. It helps merchants reduce costs, increase efficiency, and strengthen their operational capabilities while also generating high-quality data. On the matching front, our extensible recommendations powered by our e-commerce knowledge graph predict the user's potential long-term interest. This boosts conversion rates and also strengthens the user trust and effectiveness with our recommendations. We believe these AI capabilities will automatically power growth by view of data infrastructure, Precise matching and emerging efficiency empowerment driving the healthy and sustainable development of our e-commerce ecosystem. Next, regarding our live streaming business, Q3 live streaming revenue grew by 2.5% a year to 9.6 billion RMB. Growth was driven by high quality content, expanding live streaming scenarios, and AI-powered product innovations for live streaming supply. The healthy development of our talent agency ecosystem provided robust support. By NFQ3, our partner talent agencies increased by more than 17% and talent agency managed streamers grew by over 20% of both year-over-year. We focused on categories such as group live streamings by supporting premium benchmark groups, guiding content optimization to achieve high-quality development and steady revenue growth. Innovative NGC applications also injected momentum into our business growth, leveraging AI cleaning AI capabilities. In late September, we rolled out the AI Universe gift series with a customizable special effects platform-wide, effectively diversifying options for personalized interactions in live streaming rooms. On launch day alone, users pay to create and send over 100,000 personalized virtual gifts. In Q3, for entertainment live streaming operations, we launched a Super Grand Stage 2.0, organized as five regional contests nationwide to further integrate online live streaming and offline scenarios. Targeting the summer season and demand from young users, we hosted the Summer Gaming Music Festival in Chengdu and offline that event that blended gaming music and interactive experiences, depending on our partnerships with game developers. The event attracted 672 million live stream views and over 550 thousand participants on site. Moreover, our live streaming plus strategy continue to empower traditional industries further validating its commercial value. In Q3, average daily number of users submitting resumes and client hire increased by over 20% year-over-year. In ideal housing, monthly number of paying clients increased by over 90% year-over-year. Finally, our overseas business. In Q3, we continue to strengthen our foothold in overseas markets, focusing on high-quality growth. On the traffic front, we optimized Customer acquisition efficiency to precisely reach high-value demographics. By prioritizing operations for core category creators, we foster stronger connections between our high-quality characteristic content and a core user base. Brazil, our core international market, maintains stable SEUs while reducing user acquisition costs year-over-year, delivering consistent year-over-year growth in average daily time spent per DEU. For online marketing services, we bolstered business resilience to diversify our marketing client base across industries. Through an updated product capabilities and placement strategies, we improved overall conversion efficiency across our marketing funnel, unlocking more money-sufficient potential for diverse user groups, and earning sustained client recommendation. Concurrently, our e-commerce business in Brazil improved both specificity and operating sufficiency. While maintaining disciplined ROI management, we achieved a healthy year-over-year growth in GMV transaction scale and order volume in Q3. Looking ahead to Q4 and into 2026, we will continue investing in our AI strategy, exploring efficient gates that empower users, video creators, marketing clients, and e-commerce merchants through clean AI and other large AI model technology. At the same time, Guided by our development philosophy and AI strategy, we will comprehensively transform and upgrade our organization structure, talent and deployment, product design and features. We will persistently uphold and accentuate Kuaishou's technology innovation ethos, maintaining and deepening our long-term competitive advantages in the era of AI. That concludes my preparatory remarks. Next, our CFO, Bing, will review the company's financial update for Q3 2021. Thank you, Yixiao, and hello, everyone. In Q3, we continue to strengthen our core advantages, leveraging our large AI model capabilities. We further empowered our content and business ecosystems. With our rich content supply and optimized Omni domain operations ecosystem, we continuously enhance the experience for users and creators while helping merchants and KOLs improve their operational capabilities and support sustainable growth. During the quarter, we achieved solid operational and financial results with our total revenue increasing 14.2% year-over-year to 35.6 billion RMB. This included a 19.2% year-over-year increase in revenue from our core commercial business, which includes our online marketing services and other services, primarily commerce. With our steady revenue growth and improved operating efficiency, we improved our overall profitability. Operating profit increased 69.9% a year to 5.3 billion RMB. Adjusted net profit grew 26.3% a year to 5 billion RMB with a healthy adjusted net margin of 14%. Now let's take a closer look. Our total revenue grew 14.2% year-over-year to 35.6 billion RMB in Q3. The increase was mainly driven by growth across each of our business, including online marketing services, live streaming, e-commerce, and client AI. In Q3, online marketing services revenue increased 14% to 20.1 billion RMB from 17.6 billion in the same period of last year. The growth was primarily attributable to the use of AI technology to continuously upgrade online marketing product solutions that improve the conversion efficiency, which drove higher client spending from our marketing clients. Revenue from other services, including e-commerce and clean AI businesses, reached $5.9 billion in Q3, up 41.3% from $4.2 billion R&D in the same period last year. The increase was mainly driven by growth in e-commerce GMV, which boosted e-commerce commission income, as well as the expansion of our Cling AI business. We have continuously refined Cling AI's foundation models, developed more innovative features. Its application coverage has expanded, driving further breakthroughs in commercialization. In Q3, our live streaming revenue was $9.6 billion, RMB up 2.5% from $9.3 billion in the same period last year. We consistently cultivated high-quality content, expanded live streaming scenarios, and leveraged AI-empowered product innovations to build a diverse and healthy live streaming ecosystem. These steps drove greater user engagement with high-quality live streaming content. Cost of revenues increased 13.4% year-over-year in Q3 to 16.1 billion RMB, accounting for 45.3% of total revenue. The increase was mainly due to increased revenue sharing costs and related taxes in line with our revenue growth, partially offset by decreases in depreciation of property and equipment and right of use of assets and amortization of intangible assets. Q3, our gross profit grew 14.9% year-over-year to 19.4 billion RMB. Gross profit margin was 54.7% up 0.4 percentage points year-over-year. Moving to expenses, selling and marketing expenses were 10.4 billion RMB, roughly flat year-over-year, and accounted for 29.3% of total revenue, down from 33.3 in Q3 last year, reflecting our refined efforts and improved operating efficiency. R&D expenses were 3.7 billion RMB, up 17.7% year-over-year, accounting for 10.3% of total revenue. The increase was mainly due to higher employee benefit expenses, including share-based compensation expenses, and increased investments in AI. Administrative expenses decreased 13.6% year-over-year to $688 million, or 1.9% of total revenue, mainly due to poorer employee benefit expenses, including share-based compensation expenses. level operating profit for Q3 increased 69.9% a year over year to 5.3 billion RMB. Net profit for Q3 was 4.5 billion RMB. Adjusted net profit rose 26.3% a year to 5 billion with an adjusted net margin of 14%. Our balance sheet is quite robust with Cash and cash equivalents, time deposits restricted to cash and wealth management products totaling 106.6 billion RMB as of September 30, 2025. We generated positive operating net cash flow of 7.7 billion in Q3. Additionally, we actively deliver on our commitment to shareholder returns based on marketing conditions. As of September 30th, we had to repurchase an aggregate of promisably 2.17 billion Hong Kong dollars, or around 42.25 million shares, which accounted for about 0.98% of our total shares outstanding for 2020. In addition, we declared a special dividend of 2 billion Hong Kong dollars in Q3, reflecting our confidence in Kuaishu's long-term growth prospects and a solid financial position. Looking ahead, we'll continue to prioritize the user needs and execute our strategy to empower all of our business areas while exploring more diversified growth avenues. These initiatives will reinforce our competitive edge and average changing market and enable us to create a long-term value for our users, partners, and shareholders. That concludes our prepared remarks. Now, let's move into the Q&A session.

Thank you. If you have any questions, We will now begin the question and answer session. If you'd like to ask questions, please press star 1 and wait for a name to be announced. The first question comes from Felix Liu of UBS. Please go ahead. 请提问。

Thank you for accepting my question, and congratulations on our very strong performance in the third quarter. My question is about CleanAI. The market is very concerned about the competition pattern of the video-generated AI industry. What is the competition strategy of CleanAI and the direction of the next generation? How will we look at the development direction of video-generated AI after SORA2? And what are the opportunities for the follow-up 2C? Thank you, management, for taking my question, and congratulations on the very strong third quarter results. My question is on CleanAI. The market is very focused on the competitive landscape of video gen AI. Can management share more color with competition strategy from here, and where do you plan to develop and drive evolution in Clean from here? After the launch of SOAR 2, how do we see the development of the overall video gen AI industry, and do you anticipate more opportunities on the 2C side of video gen AI? Thank you.

This is a similar question. Previously, we saw that there were a lot of industrial workers on the video generation channel, including other large telecom companies and some creative companies. Thank you for your question.

The surge of entrance from to start-ups reflects just how attractive and promising the video generation market is. That said, we believe video generation is still far from maturity in both product and technology. With a growing number of market participants, we expect accelerated innovation across the industry, meeting more user needs, penetrating a wider range of use cases, and pushing the market to expand even more.

Next, I would like to share with you about Kerlin's product positioning and competitive strategy. First of all, our goal is to focus. Kerlin's vision is that everyone can tell a good story with AI. We will use AI to create this goal, focusing on resources, deepening the ability of technology and products. The four-piece model and the unit model are essentially moving towards the four-piece model. And we believe that the four-piece model will be the core technology of the four-piece model. In the entire industry, in addition to video creation, there are also many other areas of music, such as interactive scenes, data production for GSM, and so on. We will keep our ability to penetrate into the models and products of each application scene sharp, but we will still focus on our core goals and AI video creation scenes in the future.

As for claiming positioning and competitive strategy, we have zeroed in on a key goal to empower everyone to craft competitive stories with AI. Our first industry focus is film and television, where we are dedicating our resources to deepening our tech and product capabilities. Video models, like large language models, are essentially evolving toward world models. We see video models as the key technology for world models. Applications can extend far beyond film and TV production. They can reach interactive experiences, and data generation for embedded intelligence. While we will continue sharpening our model and product capabilities across diverse application scenarios, our strategic focus right now is squarely set on AI-powered film and TV production.

Our main goal is to continue to build on the imagination of technology and products. Compared to the original model, the video production model has two characteristics. First, it is highly complex. The original model is relatively simple in terms of macro structure, while the video model is made up of many different modules. The complex system also means that there is a lot of technical space and innovation space. The second characteristic is that video production is an open-source problem. For example, the input end can include text, images, sounds, and sports tracks. There are many modes such as image, video, and sound. The two characteristics of complexity and openness mean that the space between the technology and product selection is large, which means that both technology and product innovation have a larger space. Klink has been working hard to realize the dynamic and technical boundaries of product imagination and user demand, and the three-way combination of breakthrough capabilities. With this goal in mind, we have been advancing our technology leadership and product creativity, and we'll continue on this path.

video models differ from language models in two ways first they are highly complex while language models are relatively simple at the macro level video models consist of a wide range of different modules this complexity also gives us significant room for technological breakthroughs and innovation second video generation is an open-ended domain input be text pictures on ocean trajectories and outputs can be diverse modalities including images video and sound these two characteristics complexity and openness allow greater flexibility in technology and product choices which in turn provides significant room for technology and product innovation calling ai aims to bring together product creativity inside the user's inability to push technological boundaries For example, in April, we read our concept of interaction called MVL. Building on this, we are continuously upgrading our foundation model and product capabilities, exploring more high-model products.

In addition to the continued breakthrough in technology and product, we have also built up an ecosystem of all-round creators and a prosperous ecosystem through a variety of operations. For example, CleanAI's future partner plan, through the integration of the core resources of CleanAI, to match the high-value business orders of many scenarios, not including the well-known brands of NBA and Michigan City. Recently, we have also used NextGen's new global image to challenge the competition to make CleanAI's creators in Fushan, Ghana, and Tokyo International Film Festival get more exposure and continue to expand the brand influence of CleanAI products.

Alongside the content that went through in our product to the police, we have also a wide range of operational initiatives to foster an end-creator mechanism and a thriving content creation ecosystem. For example, our Cling AI Future Partner Program integrates key resources from both Kuaishou and Cling AI to precisely match creators with high-value commercialization opportunities across diverse scenarios. The program has supported well-known brands such as the NBA and Mishra Ice Cream and Tea. We also recently leveraged the Cling AI Next Gen Creative Contest, helping Cling AI creators gain exposure at international film festivals in Busan, Cannes, and Tokyo, further expanding Cling AI's global brand visibility and influence.

And about the recent Zorro 2 that everyone is enthusiastic about, not only achieved technical breakthroughs at multiple levels, but also deepened the integration of social interaction functions, and accelerated the landing of Western consumer AI applications. This makes us more confident in the future commercialization of video production. Although our main experience is still facing professional creators to improve their experience and welfare. But we will also continue to explore the Western application scene,

As for the latest buzz around Sora 2, it has made technology breakthroughs on multiple fronts and integrated closely with social interaction features. This has really accelerated the rollout of consumer-level AI applications and strengthened our confidence in the future commercial scalability of video generation. For us, our main focus is still on professional creators, improving their experience and willingness to pay. At the same time, we are actively exploring consumer-facing use cases. When the time is right, we will advance the productization of clinging ice technology, embedding social features to speed up consumer-level applications and commercialization.

Thank you, operator. Next question, please.

The next question is from Lincoln Kong from Kaohsiung. The next question comes from Lincoln Kong from Goldman Sachs. Please go ahead.

Thank you for accepting my question. Congratulations to the company for achieving a good performance. I would like to ask about the whole A.I. In addition to this Kling A.I. and the one that Mr. Yi mentioned just now, the one rack in the online sales business, can you tell us more about it? So thank you, management, for taking my question and congrats on a very solid result. So my question is about the AI-empowered business. So on top of a clean AI and the one rack system we've been talking about for online marketing services, could management elaborate more on AI large language model to empower our Kuaishou content ecosystem and how to improve our operational efficiency front. Thank you.

Thank you for your question. In 2015, it was widely regarded as the key year for deep application of AI. In this year, AI technology representing multi-modal production agents is continually explored, more efficient, and more suitable for the application form of user needs. AI technology is moving towards the systemic line of industrial value. Under such a big backdrop, we are also gradually building a set of core needs for users, such as existing business scenarios, accelerating the complete technology and application system of AI landing, energy, content, ecology, commercial ecology, and internal organizational infrastructure.

Thank you for your question. 2025 is widely regarded as AI's first year advancing into deep applications. Throughout the year, AI technologies represented by a multi-model generation and AI agents have consistently moved toward richer and more efficient applications that are more aligned with the user needs. This marks a systematic step toward unlocking AI's industrial scale value. Against this backdrop, we have progressively developed a comprehensive AI technology and application system centered on user needs and rooted in our existing business areas. It is designed to accelerate AI adoption to empower our content and business ecosystems, as well as our organizational infrastructure.

Thanks. TechNet. More accurate and comprehensive content understanding has been implemented. Content multi-layer construction, new research and exploration, and other core use scenarios have led to a constant improvement in human GDP. In terms of content generation, we also see a slight increase in the content of AIGC. In terms of the most core content recommendation, this quarter, by continuing to upgrade to the end-to-end generation, we recommend the big model of One Rack. In terms of empowering our content ecosystem,

AI has now been fully integrated across Kuaishou's business operations from content and user understanding to content generation and recommendations. First, in understanding content and users, our proprietary multi-model, large language model, Kuaishou, has demonstrated strong video comprehension capabilities. Based on this model, we upgraded our short video and live streaming content understanding system and launched Technex, our next-generation tagging system, which enables more accurate and comprehensive content understanding. TechNext is now being applied across key scenarios, including early-stage content management, content diversity expansion, and new interest discovery, driving higher average app usage time per user. Second, in content generation, Clang AI continues to empower mass creators, we have witnessed a significant increase in the video views volume of AIGC short video content on the platform. Third, in content recommendation, the most important area, we further expanded the boundaries of generative recommendation systems by upgrading our end-to-end generative recommendation large model, One Rec. We launched the next generation One Rec Stink large model, integrating LLM inference capabilities and combining conversational inference, personalized recommendations, and real-time feedback mechanisms into one single model system. This further enhances recommendation accuracy and strengthens user trust.

AI technology plays an important role in internal organization infrastructure and promotion. Our self-developed AI programming product called Flickr has become a smart development tool that fast engineers use on a daily basis. Normal smart single test production, smart code evaluation, smart test utilization production, etc. will also have a long-term support. Currently, there are almost 30% of the production of fast new code called Flickr. In terms of content review, we have used AI, large model applications and user images, content recognition, comment recognition, and many other scenarios to use thinking chain and powerful learning techniques to improve review and large model capabilities. Currently, the content of the platform has exceeded 99% of the initial review rate of AI, while greatly saving relevant costs and increasing review efficiency and quality. Beyond business empowerment, AI technology has played a major role in improving the efficiency of our organizational infrastructure.

Our proprietary AI coding tool, CodeFlicker, has become a core intelligent development tool used daily by our engineers at a high frequency. It supports scenarios such as automated unit testing generation, intelligent code review, and smart testing cases generation. Currently, nearly 30% of new code at Kuaishou is generated using CodeFlicker. In terms of content review, we have applied large AI models across diverse scenarios, including user profiling, content identification, and comment analysis. By leveraging COT reasoning and reinforcement learning technologies, we have enhanced our review model's capabilities. Currently, over 99% of the content on our platform is reviewed by AI. greatly reducing related costs while improving the efficiency and quality of content review. In addition, our customer service team is leveraging AI technology to prescreen and route user inquiries, provide intelligent assistance, and accumulate knowledge. As a result, over 70% of user inquiries are now directly handled and resolved by our AI-powered customer service system, significantly improving efficiencies.

As a result, we see that from AI technology innovation to application landing, the positive cycle that leads to revenue growth is taking shape within Quasio. From a long-term perspective, we believe that this all-round AI application mentality can make Quasio have a stronger market adaptability and growth potential.

Overall, a resilient self-reinforcing cycle of AI innovation, AI application, money transition, and revenue growth is taking shape at Kuaishou. In the long run, we believe this full-spectrum AI application ecosystem will further strengthen Kuaishou's market resilience and unlock new growth momentum.

Thank you all for your time. Next question, please.

The next question comes from Thomas Chong of Jefferies. Please ask your question.

晚上好,谢谢管理层介绍我的提问。 我的问题是关于线上营销方面的。 每季度我们看到我们线上营销收入增长进一步提速, 我们可否从流量、行业、还有产品等角度 详细拆解一下我们具体做了什么? Hi, good evening. Thanks, management, for taking my question. My question is about online marketing services. We have seen our online marketing revenue accelerating this quarter. Can management provide more details on what we have done from the perspective of traffic, industry sectors, as well as product offering? Thank you.

接您的問題,本季度線上營銷服務收入同比增加14%, The second quarter's increase in sales has been further improved, especially as domestic online marketing service revenue has exceeded 16% growth. From traffic driving factors, look at marketing material exposure and GPM dual-wheel drive revenue. The increase in marketing material exposure is related to the overall traffic growth of the big plate. On the other hand, more high-quality original marketing content helps to improve the exposure. Thank you for your question.

In Q3, online marketing services revenue grew by 14% year-over-year, accelerating from the previous quarter. with domestic online marketing services revenue increasing by over 16%. From the traffic perspective, advertising revenue was driven by both increased marketing material impressions and higher CPM. The growth in impressions was supported by overall traffic growth and by more high-quality native marketing content, which helped increase ad load. The rise in CPM was driven by our use of AI technology, such as the generative reinforcement learning bidding and end-to-end generative recommendation models, which improved the matching between user interest and advertiser needs, enhancing the personalization and matching efficiency of online marketing material recommendations.

From the outside circle of the industry, We have upgraded our private products in the service industry. and in the subsequent transition chain after the optimization of the atmosphere, help customers to improve efficiency and improve user transformation. At the same time, in the life service industry, customers focus on small and medium-sized customers, which is faster than AI customer UX, AIGC, marketing, material, production tools, and other products. In the content and sales industry, the short-term demand continues to maintain a high increase. At the same time, with the deep negative energy of AI technology, the slow growth has led to a high development.

Looking ahead at external marketing services industry-wise, lifestyle services where clients mainly rely on lead-based operations and content consumption represented by short plays and mini games were the standout sectors this quarter. In lifestyle services, we upgraded our private messaging product and optimized the subsequent conversion passes across industry verticals, helping clients reach users more efficiently and improve sales conversions. Since most of our lifestyle services clients are small and medium-sized businesses, they benefit more from products like our AI customer service, UAX placement solutions, and AIGC marketing material generation tools. In content consumption industries, deep AI empowerment, growth, rapid growth, and comic style short plays. We captured this opportunity and used to playing AI to play an active role in upstream content creation.

In the internal circulation of e-commerce marketing services, we continue to deliver full-time products to help e-commerce merchants gain more volume and transformation. At the same time, the smart price adjustment agent, a multi-layer data model, is implemented throughout the day to stabilize the price. The user interest is fully explained, and the sales budget of the merchant is put forward. Whether in the content or the price range, it strengthens the understanding of the user's desire to recover. The optimization can effectively improve the size and scalability of the conversion users. It is better to satisfy the user's fast e-commerce consumption.

In terms of our closed-loop marketing services, we continue to iterate our Omni platform marketing solution, helping e-commerce merchants achieve more incremental exposure and conversion. By leveraging intelligent bidding agents and generative large models, we enable 24-7 stable bidding and more fully uncovered user interest, which helped expand merchants' placement budgets We also strengthen our ability to capture and interpret a user's full-range interests across both content-based and shelf-based scenarios, effectively increasing the number of converted users and their purchase frequency, while better meeting users' e-commerce consumption needs on Kuaishou.

At the product level, we have upgraded the UX, AIGC, and sales material production tools. products such as live broadcasts, digital employees, and other products to help customers reduce sales thresholds, optimize customer investment, and transform. The first step is to drive the growth of online sales consumption. Specifically, this quarter's UX all-automatic advertising investment products added the cycle stability function. Cycle stability is the need for customers to express their material and price in the cycle that has already been invested. The system carries out automation, smart infrastructure, smart control, and smart creation production, thereby improving the stability of investment cycle. to help online sales customers achieve a more stable and continuous investment status. In the third quarter, UX all-automatic advertising investment products have reached more than 70% in the external cycle of consumption penetration. AIGC's sales material development tools support customer low cost, fast batch production of short-term material. At the same time, material conversion efficiency is also 10% to 20% higher than the average level of the industry. From a product perspective, we upgraded to multiple products, including how

IGC marketing material generation tools, live streaming digital human solutions, and our virtual employees. These enhancements lower the marketing threshold and improve the conversion rates, driving more online marketing services spending. Specifically in Q3, our UAX placement solutions added fixed period steady placement feature. The new feature allows clients to set their requirements for marketing materials and pricing for a specific ad placement period. while the system automatically handles intelligent infrastructure, smart dynamic fine-tuning, and smart creative content production. This enhanced the stability of the ad placement period and helped our online marketing clients achieve more consistent placement performances at more predictable costs. In Q3, our UAX placement solutions accounted for over 70% of the external marketing spending. Our AIGC marketing material generation tool enabled clients to generate short video materials rapidly at a low cost and in batches with a 10 to 20% higher material conversion efficiency than the industry average. Live streaming digital human solutions allowed our clients to run 24 seven live streams, even without streamers or venues. Our virtual employee reached a human level customer service performance in conversational accuracy, efficiency, and safety, engaging naturally across scenarios like private messaging and comments,

improving conversion efficiency for our clients.

Looking ahead, we'll continue to expand our industry client base and further deepen AI applications, empowering clients to achieve more efficient, high-quality marketing performances and better ad placements.

Thank you, operator.

Next question, please. The next question comes from Daniel Chen from J.B. Morgan.

Please go ahead. So my question is related to e-commerce. So what's the latest progress and performance of our double 11 promotion in December quarter? And if we look at the next

one to two years what's the incremental what's the key growth driver for our e-commerce business especially the live streaming e-commerce how should we look at the future growth potential thank you thank you Among them, Zhu Baoyu's tea and wine health, large clothing, including men's clothing and women's clothing, as well as the performance of the 3,000 food industry, etc. This year, we invested 1.8 billion yuan in platform flow incentives, combined with 2.1 billion yuan in red envelopes and 1.1 billion yuan in product subsidies, to help the merchant increase, convert, sell, and make the buyer more active. The J&V has sold tens of millions of products, and has reached a double growth. Based on the business type and scale of Shanda, we have developed a policy of division of cities, drive the ecological prosperity of the field of business, and stimulate Shanda to better gain power and growth. At the price of goods, our support focuses on core products, and we have launched a series of projects such as big cards, big supplies, super links, etc. Thanks for questioning.

Regarding e-commerce, while consumption has shown some resilient recovery this year, overall user spending has remained cautious and rational. During the 2011 sales promotion, we delivered the results in line with our expectations with standout performances in categories such as jewelry and gemstones, tea, wine, and wellness, apparel, including men's and women's apparel, sportswear, and family matching outfits, and fresh food. For this year's 2011 sales promotion, we invested over 18 billion RMB in platform traffic incentives, combined with 2 billion RMB in user subsidies and 1 billion RMB in merchandise subsidies. Together, these effectively enhanced the merchant sales convergence and buyer engagement, increasing the number of merchants achieving GMV of over 10 million RMB by double digits year over year. We implemented tiered support programs tailored to business type and merchant and account size, fostering a thriving e-commerce ecosystem and motivating them to achieve better growth across many domain scenarios. For shelf-based e-commerce scenarios, we focus on supporting core products where we launched a range of initiatives, including the Big Brand, Big Sipsy, and Superlinks. During this year's W11 sales promotion, the number of single products achieving over 1 million RMB GMV via the Big Brand Big Subsidy initiative surged by over 77% a year over year. Our users' mindshare for shopping on Qyzer improved during the sales promotion, with search-generated e-commerce GMV growing by over 33% a year over year.

In the near future, e-commerce business will grow rapidly, and we will focus on how to improve our e-commerce business.

How many times do we need to improve our e-commerce business? How many times do we need to improve our e-commerce business? How many times do we need to improve our e-commerce business? How many times do we need to improve our e-commerce business? . . . . .

to enhance users' trust in the platform and lead to stable growth of RPUs. Of course, the growth of e-commerce paid buyers has greater room for growth, but we will see it as a long-term outcome indicator, not a process indicator. In the short term, we will pay more attention to the health of e-commerce paid buyers.

For our future e-commerce growth drivers, in the short to medium term, we will prioritize boosting user purchase frequency, followed by increasing AR PPU. Our key initiatives to raise purchase frequency are first, we will continue to empower streamers to strengthen their private domains and operational efficiency, broadening the variety of streamers and product categories that users pay for. Second, We will maximize cross-scenario synergy. Lower purchase barriers in short-view scenarios will allow us to expand our buyer purchasing purchase rate. Moreover, as we progressively reinforce users' shopping mindset, our pan-shell-based e-commerce will better capture users' repeat purchases needs with greater certainty. We'll further enhance the operations of our key product categories and more precisely identify our core user base, keeping users' trust in the platform, having steady ARPB growth. So there is still significant room to grow our e-commerce monthly average paying users, but we view this as a long-term outcome metric rather than a short-term performance metric. In the near to medium term, we will mainly focus on the healthy structure of our e-commerce monthly average paying users.

In terms of the basic space of live e-commerce as a content platform, live e-commerce and new e-commerce have always been the basic and core areas of our e-commerce. We believe that the form of live e-commerce is in the advantage of natural transformation and will still drive it to improve its penetration in the online retail market. In the future, we still have the opportunity to have an objective space and structural growth. The future growth space of Zipo e-commerce is whether we can build a healthy ecosystem that allows businesses to run long-term. And Ziyu's powder-layer store is a long-term partner of Ziyu. Because Ziyu users have the advantage of high annuality and high recovery, we use multiple methods to help businesses do good public-private contact, obtain traffic in the apartment, and make Ziyu a good powder-layer store. But the most important and most important job is still to make good content and provide good products. Therefore, we will continue to encourage business owners and adults to expand their quality, supply, and access to water, and at the same time continue to expand the scope of business goods. We will work hard to strengthen the long-term co-development of business owners and adults. through sales channels and other channels to provide them with abundant supply and demand, and provide traffic support for high-quality content. At the same time, the platform will provide smart operation, product governance, and use AI products to supply and supply, and enhance new efficiency. A good business environment will also encourage the continued creation of good content.

Regarding the growth potential of live streaming e-commerce, as a common platform, live streaming e-commerce and trust-based e-commerce have always been the backbone of our e-commerce business and most critical operational scenarios. We believe that live streaming e-commerce with its built-in conversion advantages will continue to gain ground in the online retail market and it still holds substantial room for structural growth in the future. The long-term growth potential lies in creating a healthy ecosystem where merchants can operate sustainably with private domain follower retention, acting as a key mode given their high user stickness and repeat purchase behavior. Accordingly, we help merchants better integrate to their public and private domain strategies through a range of initiatives, acquiring traffic in the public domain while retaining followers and converting them into customers and driving repeat purchases in private domains. That said, exceptional content and superior products remain the essential foundation of our ecosystem. Therefore, we'll continue to onboard merchants and creators, expanding the pipeline for high-quality supply while continuously broadening the range of merchandise. In parallel, we will strengthen long-term collaboration with both merchants and KOLs by offering them extensive products through our distribution pool and providing traffic support for standout content. We will also equip the merchant and KOLs with our intelligent operational tools, empowering them with AI to improve efficiency and performance. A robust business ecosystem, in turn, will incentivize the continuous creation of exceptional content.

最后呢,直播电商虽然是快手的电商基本盘, 但我们也将鼓励商家多层经营, 加强权益协同效率,更好地承接用户需求的同时, 加强快手电商生态的韧性和确定性。

Finally, while live streaming e-commerce is the backbone of Kuaishou's e-commerce, we will also encourage merchants to operate across diverse scenarios and strengthen the efficiency of all many domain synergies. This will facilitate a closer alignment with the user needs and enhance the resilience and stability of Kuaishou's e-commerce ecosystem.

Thank you. Thank you. The last question, please, operator.

The next question is from Xueqing Zhang from CICC. Next question comes from Xueqing Zhang of CICC. Please go ahead.

谢谢管理层接受我的提问。 我想请教一下关于AI CapEx和利润率的问题。 随着可能和其他AI赋能进展, 公司对于2025年CapEx和AI费用这边有没有更新? 2025年全年的利润率目标是否有变化? 以及在整个行业都在大力提升CapEx的情况下, Thanks for taking my question. My question is regarding CapEx and profit margins. With the progress of CLEAN and other AI-driven initiatives, does the company have any updated guidance on the CapEx and AI-related spending plans? Has the full year 2025 profit margin target being adjusted And given that the entire industry is significantly increasing capacity, how is the question of planning the capacity over the next one to two years? And what impact will AI investments have on profit margins? Thank you.

Thank you for your question. As President Yi said, we have achieved very good progress in the combination of AI technology and internal and external application scenarios this quarter. AI is also improving its internal organization infrastructure while providing business benefits. AI technology is gradually releasing the value of fast content and commercial ecology. At the same time, Kerlin is also making breakthroughs in commercialization and sales. We expect that Kerlin's annual revenue in 2025 will reach US$140 million, which is more than 100% higher than its goal of US$60 million at the beginning of this year. Due to the increasing demand for video production models, we have repeatedly increased our investment in computing. In addition to increasing our investment in reasoning computing, we have also started to increase our training in computing to maintain the technical advancedness. In addition to the investment of CapEx brought by other AI businesses, we expect that the overall CapEx output of the group in 2025 will achieve the same growth as that of the high-middle and high-small numbers last year. In terms of cost, we have recently added some investment on the attraction and preservation of AI technology, but this part of the cost is relatively controllable. Although we have increased the investment related to the above AI strategies, we are still fully committed to achieving the same increase in profit-making efficiency as this year. The overall increase in profit-making capacity also further explains that AI technology is gradually releasing the negative value of the fast content ecosystem and commercial ecosystem.

Thanks for your question. As Yixiao said, this quarter we achieved strong results by integrating AI technology across a wide range of internal and external application scenarios. AI empowered our business operations and improved the quality and efficiency of our organizational infrastructure. AI technology continues to unlock increasing value across our content and business ecosystems. At the same time, Cling AI made more solid breakthroughs in commercialization. We now expect Cling AI's full-year 2025 revenue to reach $140 million, more than double the target we set at the beginning of the year of $60 million. Given Cling AI's users' growing demand for video generation models, we have continued to ramp up our investment in computing power for Cling AI. Beyond the incremental investment in inference capacity, alongside continuous model iterations, We have recently started scaling up Culling AI's training computer power to keep Culling AI at the forefront of technology advancement. Including this and CapEx from other AI initiatives, we expect the group's total 2025 CapEx to increase in the mid to high double digits year over year. Regarding expenses, we have recently stepped up our investments in hiring and retaining AI talent. This portion of expenses remains relatively manageable, and despite the higher AI-related investments, we're confident that our full-year adjusted operating margin will continue to improve year-over-year. Our overall improvement in profitability further underscores that AI continues to unlock increasing value across Qyto's content and business ecosystems.

In short, we are pleased to see the growth of QoLing and AI technology. We are relatively clear about the future capital spending growth plan. The focus is on the upgrade of computing power and technological evolution. This is not a simple cost-benefit increase, but a logic based on the increasing value of AI technology. As the AI application scene continues to expand, the value of AI will also be further released. We are confident that in the next two years, we will continue to achieve further increases in profit scale and steady growth of profit capacity.

Thanks to the better-than-expected progress in the production of AI technology of our businesses, we have launched our X-Groys plan with a focus on upgrading computing power and technology. This goes beyond a simple rise in costs and expenses. It is in our strategy of leveraging leads, in AI to drive greater value. As AI applicants continue to expand across areas, their potential value will be a lot. We're confident that we can continue to steadily grow our profits, improving profitability over the next two years, and we look forward to sharing our progress along the way. Thank you.

Thank you all very much. That's the end of the Q&A session.

Thank you once again for joining us today.

If you have any further questions, please contact our Capital Market and IRR team at any time. Thank you.