Alibaba Group Holding Limited

Good day, ladies and gentlemen. Thank you for standing by. Welcome to Alibaba Group's September quarter 2025 results conference call. At this time, all participants are on listen-only mode. After management's prepared remarks, there will be a Q&A session. I would now like to turn the call over to Lydia Liu, head of investor relations of Alibaba Group. Please go ahead.

Thank you. Good day, everyone. Welcome to our September quarter 2025 earnings conference call. With me today from Alibaba are Zhou Cai, Chairman, Eddie Wu, Chief Executive Officer, Toby Xu, Chief Financial Officer, Jiang Fan, Chief Executive Officer of Alibaba eCommerce Business Group. I would like to remind you that this call is also being webcast on our corporate website. A replay of the call will be available on our website later today. Just a few forward-looking statements before we begin today. Today's discussions may contain forward-looking statements, particularly statements about our business and financial results that are subject to risk and uncertainties which could cause actual results to differ materially from those contained in the forward-looking statements. Please refer to the safe harbor statements that appear in our press release and investor presentation provided today. Please note that certain financial measures that we use on this call are expressed on a non-GAAP basis. Our GAAP results and reconciliations of GAAP to non-GAAP measures can be found in our earnings press release. With that, I'm going to turn the call over to Eddie.

Welcome to Alibaba Group's quarterly earnings call. Over the past quarter, Alibaba delivered steady and healthy growth. Our total revenue increased 15% year-over-year, excluding SunArt and InTime. Our continued investment in core businesses is yielding results, with China e-commerce, CMR, growing 10% and cloud intelligence revenue rising 34%. Let me walk you through the latest developments across our AI plus cloud and consumption businesses. Sustained strong demand for AI and rising usage of public cloud drove Alibaba's cloud's 34% revenue growth this quarter, while revenue from external customers accelerated by 29%. AI-related products continued to post triple-digit year-over-year growth for the ninth consecutive quarter. In the cloud computing market, two major trends are becoming increasingly apparent. First, as AI applications scale, more developers and enterprise customers are choosing vendors with full-stack AI technology portfolios. Second, customers are deepening and broadening their use of AI, which is significantly increasing demand for compute, storage and other traditional cloud services. Together, these forces are accelerating revenue growth driven by external customer demand. This quarter, we continued to strengthen our full-stack AI capabilities, spanning high-performance AI infrastructure, foundation models and AI development frameworks. Our flagship model, QN3 Max, ranks among the global leaders in benchmarks for real-world coding tasks, agent tool use capabilities, and other specialized valuations. Our full-stack AI capabilities are now a defining competitive advantage. Alibaba Cloud is gaining market share across multiple segments. In the hybrid cloud market, Alibaba Cloud has become a key player, growing more than 20% year over year, outpacing the industry and steadily expanding market share. Our financial cloud business is also growing faster than the market, with market share continuing to rise. In China's AI cloud market, we're also the clear leader with a market share larger than the combined total of the second to fourth largest providers. Recently, businesses such as the NBA, Marriott, China UnionPay, and Bosch have partnered with Alibaba Cloud on AI initiatives. Last week, we officially launched the Q1 app, which aims to be the most advanced personal AI assistant powered by our latest models. In the first week of its public beta, The Q1 app has already surpassed 10 million in new downloads. The launch of the Q1 app marks Alibaba's commitment to both AI for enterprise and AI for consumer. In enterprise-focused AI, our goal is to build a world-leading full-stack AI provider serving businesses across all industries. For consumers, we aim to build native AI-first applications by leveraging our best-in-class models and Alibaba's extensive ecosystem. On the one hand, QN3 Max's intelligence and world-class tool use capabilities, combined with Alibaba's rich consumer and lifestyle use cases, contributed to exceptional user retention in the QN apps beta release. We believe this is the right moment to scale our consumer AI efforts. On the other hand, the synergy between AI and the broader Alibaba ecosystem is a powerful multiplier. Alibaba is the only company in China with both a leading large model and extensive lifestyle and commerce use cases. QN will gradually integrate e-commerce, map navigation, local services and more, becoming an AI-powered entry point for everyday life. With AI innovation and ecosystem collaboration reinforcing each other, we're confident in our ability to deliver substantial user value. In consumption, we continue to deepen collaboration across businesses, and the benefits of our large integrated platform are becoming increasingly evident. This quarter, China e-commerce CMR grew 10%. Our quick commerce business saw significant improvement in unit economics with greater fulfillment efficiency, stronger user retention, higher average order value, and expanding scale. The growth of quick commerce business contributed to rapid growth in Tableau App's monthly active consumers and supported CMR expansions. Brands on Tmall are also accelerating their adoption of on-demand retail. As of October 31, approximately 3,500 brands on Tmall have onboarded their offline stores to our QuickCommerce business. Going forward, we will further enhance synergy between QuickCommerce and the broader Alibaba ecosystem, continue improving unit economics, and meet consumers' fast-growing demand for immediate access to diverse products and services. On October 1, AMAP's daily active users reached a historical high of 360 million. In September, we launched the AMAP Street Stars feature, which has significantly boosted user engagement. In October, AMAP Street Stars averaged more than 70 million daily active users, with average daily user reviews more than triple the amount of the same period last year, indicating strong future growth potential. AMAP Streetstars has built a trust-based rating system for local offline services using user-consented metrics, such as the user's credit rating. We believe that enhancing consumer trust is essential to strengthening consumer confidence, enabling merchants to focus on operations while giving consumers greater peace of mind, supporting the healthy and sustainable growth of the local offline services sector. Looking ahead, we'll continue investing decisively in our two core strategic pillars, AI plus cloud and consumption. We will advance both enterprise and consumer-focused AI, unlock deeper synergies across Alibaba's businesses, and use these engines to drive Alibaba's long-term growth and carry the company to the next level. Thank you. I will now hand over to Toby.

Thank you, Eddie. We are continuing our focus and discipline on AI plus cloud and consumption, and we see strong momentum from these strategies with gains in technology, market share, consumers, and user engagement. Now let's look at the financial results. On a consolidated basis, total revenue was RMB 247.8 billion. Excluding revenue from sunnah and in time, revenue on a like-for-like basis would have grown by 15% year-over-year. Total adjusted EBITDA decreased 78%, primarily due to our strategic investments in quick commerce business to grow its user base and transaction volume, partly offset by double-digit revenue growth in China e-commerce group and cloud intelligence group, and improved operating efficiencies across various businesses, including AIDC and Hu Jing DME. Our gap net income was RMB 20.6 billion, a decrease of 53%, primarily attributable to the decrease in income from operations. Operating cash flow was RMB 10.1 billion, a decrease of RMB 21.3 billion compared to the same quarter last year. The year-over-year decrease was mainly attributed to our increased strategic investments in quick commerce business. Free cash flow was an outflow of RMB 21.8 billion, which reflected our significant investments in quick commerce business and AI plus cloud infrastructure. We are reinvesting our free cash flow to create a winning quick commerce business and to be a leader in AI. Our strong balance sheet backed by US dollars 41 billion in net cash gives us confidence for this reinvestment strategy. Revenue from Alibaba China e-commerce group was RMB 132.6 billion, an increase of 16%. Customer management revenue increased by 10%, primarily due to the improvement of take rate, which benefited from the increasing penetration of Chen Zhantui and the addition of software service fees. Revenue from our quick commerce business increased 60%, During the quarter, we executed our plan to grow the scale of our quick commerce business, improve user experience, and narrow UE loss. The adjusted EBITDA from Alibaba China e-commerce group was RMB 10.5 billion. Excluding loss from our quick commerce business, our Alibaba China e-commerce group EBITDA would have grown at mid single digit year over year for the quarter. Going forward, this adjusted EBITDA may fluctuate quarter over quarter due to intense competition and a significant investment in user experience. Revenue from AIDC grew 10%. Aliexpress, in particular, has developed its Aliexpress direct model that leverages local inventories in over 30 countries. Art Express has also enhanced the range of our product offerings by launching the Brand Plus program, providing go-to-market solutions to Chinese brands going overseas. A combination of logistics optimization and investment efficiency enhancement resulted in AIDC's adjusted beta profit of RMB $162 million this quarter. Looking ahead, While we continue to enhance operating efficiency, AIDC adjusted beta may fluctuate quarter over quarter due to tactical investments in select markets. Our cloud business delivered another quarter of accelerated growth as both growth of cloud segment revenue and revenue from external customers accelerated to 34% and 29% respectively. This momentum was primarily driven by public cloud revenue growth, including the increasing adoption of AI-related products. AI-related product revenue continued to grow at a triple-digit pace. AI-related product revenue this quarter accounted for over 20% of revenue from external customers, with its contribution continuing to increase. We are seeing accelerated adoption of our AI products across a broader range of enterprise customers with a growing focus on value-added applications, including coding assistance. The adjusted epitome margin remained relatively stable at 9%. We will continue to invest in customer growth and technology innovation to increase adoption of AI infrastructure cloud and strengthen our market leadership. All other segment revenue was a decrease by 25%, mainly due to the disposal of Sunnah and in-time businesses. All others adjusted EBITDA with a loss of RMB 3.4 billion, primarily due to the increased investment in technology businesses, partly offset by the improving operating results of other businesses. Hu Jing DME has achieved profitability for three consecutive quarters. The All Others segment comprises a set of innovative initiatives, including several strategic AI-driven technology infrastructure and businesses, including our foundation model and AI apps. We are excited to continue investing in these initiatives for future growth. Thank you. That's the end of our prepared remarks. We can open up for Q&A.

Thank you, Toby. Hi, everyone. You're welcome to ask questions in Chinese or English. A third-party translator will provide consecutive interpretation for the Q&A session. The translation is for convenience purpose only. In the case of any discrepancy, our management statement in the original language will prevail. If you are unable to hear the Chinese translation, bilingual transcripts of this call will be available on our website within one week after the end of the meeting. 大家好,今天的电话会欢迎您用中文或英文提问。 我们会有三方工作人员提供实时交替传译,翻译目的方便大家理解。 如有任何疑义,请以我们管理层原始语言所做的陈述为准。 如无法听到中文翻译,本次电话会议的双语记录将在会议结束后

Operator, please go ahead with Q&A session. Thank you. To give more people the opportunity to ask questions today, please keep yourself to no more than one question at a time. Thank you. Your first question today comes from Gary Yu and Morgan Stanley. Please go ahead.

Hi, thank you for the opportunity and congratulations on a strong set of results. My question is related to cloud business. How should we look at the growth outlook going forward? Should we continue to expect growth to accelerate? And on the demand side, given we don't have a big AI company like in the U.S., How should we look at the key driver driving the external revenue growth going forward? Thank you.

关于云方面,尤其想了解云业务未来的增长险景, 是不是还有加速的可能性? In addition, in terms of demand, since China does not have such a large AI company as the United States, what are the main driving factors for our external customer income?

Thank you very much for your question. I will answer your first question first. I think from We are now seeing that the customer demand is still very good. Then we are now Aliyun's AI server, our up and down rhythm is actually still seriously unable to keep up with the growth rate of customer orders. We are still expanding the number of orders in hand. So from the current data demand, we can see that the future growth potential is still in the process of continuous acceleration. Then in demand, in fact, we see that In fact, the needs of the application aspects of the enterprise are still growing very rapidly. Now, among the many companies that we support, whether it is product development, including the process of product manufacturing, and the daily use of the products after the customer buys them, in the entire application aspects, the needs of AI are still penetrating. So we see that whether it is the enterprise as a model training or reasoning, or even their end customers start to use their developed AI products, all of these will use cloud computing. So we see that the demand of customers and enterprises is actually a relatively large one. We see that it is a big potential in the process of growth. So we think that for the future needs of AI,

Thank you for those questions.

Let me start with the first one. Certainly, we see that customer demand for AI is and remains very strong. In fact, we're not even able to keep pace with the growth in customer demand as in orders in terms of the pace at which we can deploy new servers. So we certainly do see the demand for AI is accelerating. In terms of where that demand is coming from, it's really coming from all aspects of enterprise operations as AI adoption continues to not only accelerate but deepen with applications across product development, throughout manufacturing processes, and also in terms of supporting the enterprises and customers use their products. So in all of those places, AI adoption continues to deepen. And of course, all of this activity around model training and inference requires the use of compute as well. So essentially, we're talking about huge potential and continually growing demands among real customers engaged in real-world use cases. Therefore, our conviction in future AI demand growth is strong.

Next question, please.

Thank you. Your next question comes from Kenneth Fong at UBS. Please go ahead.

Hi. Good evening, management, and thanks for taking my questions. Congrats on the strong performance in our QuickCommerce initiative. Can management share some key progress for QuickCommerce and Synergy to our core e-commerce so far? Given the Synergy, what's the outlook for December quarter CML and EBITDA for our core e-commerce? Thank you.

关于此晚上好,感谢接受我的提问。 同样演出和你们本季度在其实零售这一方面取得的加持的业绩, Then I would like to know about the latest significant progress in the field of real-time sales and some of the co-effects it has achieved. Based on these progress, what is the outlook for the CMI of the quarter in December and the core transformer EBITDA?

I will answer this question first. First of all, I would like to thank you for your question. In the past few months, we have focused on maintaining the market share and optimizing our benefits. We believe that we have achieved very significant progress in this regard. On the one hand, the optimization of order structure and the scale effect have also led to a significant decline in logistics costs. Since November, the advantage of flash purchases is that the loss of each order has been reduced by half in July and August. On such a basis, the order share of Taobao flash purchases is stable and the GMA share is stable. There is also a significant pull on the relevant 15 e-commerce categories. 我可以展开讲一下。 首先,订单结构的优化, 那过去两个月, 平台的高比单价订单占比提升, 按照最新的数据来看, 我们非茶饮的订单 已经上涨到75%以上。 那么闪购最新的比单价 环比8月份也涨了超过两位数。 比单价的提升也带动了 淘宝闪购整体GMV份额的拉动。 其次,随着订单份额的扩大, 淘宝闪购的物流规模效应正在凸显, 那体现在... Thank you for your question.

Over the past few months, we've focused on optimizing our unit economics in QuickCommerce while maintaining our market share. And we believe we've made significant progress on this front. The order mix has improved and the economies of scale from growing order volume have driven clear reductions in logistics costs. Since November, the per order UE loss for QuickCommerce has been cut by 50% compared to July, August. So on this basis, QuickCommerce has maintained stable order share with GMV share holding steady and trending upward. And we're also seeing meaningful uplift in related physical e-commerce categories. Let me expand a bit further on those points. First, in terms of order mix optimization, over the past two months, the share of higher e-commerce average order value, higher AOV orders has increased. According to the latest data, non-beverage orders now account for over 75% of total orders. Most recently, AOV for QuickCommerce has grown by double digits compared to August, which has contributed to an increase in QuickCommerce's overall GMV share. On the second point about logistics, As the order volume scales, QuickCommerce is realizing very clear economies of scale in fulfillment logistics. Delivery speed is now faster than the same period last year, while average logistics cost per order has declined significantly. In fact, the average cost per order is now lower than it was before we started making large scale investments in QuickCommerce. So these two factors together have enabled us to achieve our near-term target, namely cutting by half the per-order loss versus July-August. And importantly, during this phase of narrowing UE losses, both user retention and purchase frequency have outperformed expectations.

In addition to food and drink, we also see a rapid development of retail goods. So, retail shopping has been a major factor to drive the movement of related products and businesses, especially in the food, health, supermarkets, and other food and e-commerce categories. For example, the retail shopping order of He Ma and Mao Chao increased by 30% in August. In the past few months, we have also been actively promoting brand merchants to join Taobao retail shopping, and we will also accelerate the coordination and integration of related business products and retail shopping. In general, we firmly believe that There is a huge potential for retail shopping and Ali Ecology. In the first stage, we have completed the rapid expansion of the scale. In the second stage, we have optimized and optimized to meet the expectations. We have established a foundation for the long-term sustainable development of the take-out business, which has also increased our determination to invest in in-person sales for the long term. In the next stage, we will continue to engage in customer experience, focus on the management of high-value users, and focus on the development of sales. Taotian is one of the core strategies of the Taotian platform upgrade. Our goal is to bring 10 billion transactions to the platform in three years, and to drive the growth of the overall market share of related products.

Beyond food delivery, we're also seeing rapid growth in retail categories via Quick Commerce, clearly driving growth across related categories and businesses, especially groceries, healthcare products and supermarket segments within physical e-commerce. For example, Freshipo and Tmall supermarket quick commerce orders are up 30% from August. Over recent months, we've also actively onboarded merchants and brands onto Taobao Instant Commerce, and we will further accelerate integration and synergy between key retail categories and the quick commerce model going forward. So in summary, we firmly believe that the QuickCommerce model holds immense potential for synergy with the broader Alibaba ecosystem. In phase one, we successfully achieved rapid scale expansion. In phase two, UE optimization is progressing in line with our expectations, laying a solid foundation for the long-term sustainability of the QuickCommerce business and reinforcing our confidence in sustained long-term investment in QuickCommerce. In the next phase, we will continue to refine the user experience through operational upgrading with a focus on serving high-value users and a focus on expanding retail categories. QuickCommerce is a core strategic pillar in the Taobao Tmall Group's platform upgrade. Our goal is to generate 1 trillion RMB in GMV for the platform within three years, thereby driving market share gains across the related categories.

Let me answer your second question about CMR and EBITDA fluctuations. So from back to the whole Chinese e-commerce group, This EBITDA, first of all, let's talk about this buyback. Buyback of this EBITDA, because of this overall investment, so from this perspective, this quarter, September quarter, will be a high point. With the improvement of the entire efficiency, including a single significant improvement and stable scale, the entire investment, we expect to be significant in the next quarter. Of course, for this point, Thanks, this is Toby.

Let me take the second part of your question about CMR and Avatar. So as Jiang Fan just shared with you, QuickCommerce is having a very significant effect in terms of enhancing user engagement as well as driving transactions in relevant categories. So that of course has a positive impact on CMR. So the main thing that we need to do in this next phase is to better integrate and achieve synergies across conventional e-commerce and QuickCommerce. so as to more fully realize that impact. However, we are in an investment phase right now, so this is relevant to EBITDA. I think likely the September quarter will be the quarter during which the scale of those investments are the highest. And as efficiency improves, and UE improves, and as the scale of this business stabilizes, we can expect to see, I think, by next quarter, a significant sizing down in the scale of those investments. Of course, having said that, we will dynamically adjust the pace and size of our investments in line with market competitions.

The CMR of e-commerce is mainly affected by the payment procedures and the technology of full-time push. Because last year, we started receiving payment procedures in September. So in the next quarter, we expect the increase to slow down. We have always emphasized that our most important goal is the long-term market share. In this process, we will firmly invest in consumers, invest in business growth,

When it comes to CMR in the e-commerce business, there will be an impact. From the base effect in respect of the payment processing fee as well as the rollout of QZT, we started charging the payment processing fee in September of last year. And so starting from next quarter, we could expect to see a slowdown in growth due to that base effect. But as we've consistently emphasized, our primary and foremost objective is to secure market share for the medium and long term. And during this process, we will continue to decisively invest in consumers and in merchants, and we will resolutely move ahead with business model upgrading of our e-commerce platform. And during that process, you can therefore expect that there will be short-term fluctuations in CMR and in hepatitis.

Peter, let's take the next question.

Thank you. Your next question comes from Alex Yao at JP Morgan.

Please go ahead. 那就像刚才蒋凡总聊的我们第一阶段的投资已经结束 现在进入到第二阶段就是效率优化的一个阶段 那我想问的问题是随着我们效率优化然后取得这个成本的节省 我们在这个节省上面是怎么思考 在这个价值链上key stakeholder 上面的利益分配 就比如说对消费者端的补贴 是维持原来的那个强度 那么就是说这个财务的改善是相对渐进的 包括跟商户这边的补贴的比例 就是我们的这个成本节省 会在这三个这个 我们的优异还有多少空间继续提升?

Thank you very much for the opportunity. So as Jiang Fan just said, we've now completed the first phase of these investments. We're now in the second phase where we are enhancing efficiency. So my question is, as the efficiency is optimized and we obtain cost savings, what are we going to do with those cost savings? How will the benefits of those cost savings be allocated or distributed across the value chain among the different key stakeholders? Assume, for example, that we're going to continue to maintain the same level of intensity with respect to subsidies to consumers, and there's this ongoing incremental improvement in the financial performance of the business then what will that mean in terms of subsidies for merchants? You know, the cost savings will need to be allocated or distributed somehow across the three key stakeholders, consumer, merchant, and platform. And then so if we don't decrease those subsidies to consumers and we continue to follow the same path that we're on, now rely on optimization of user mix as well as driving increase in order share and driving higher basket sizes. What does that mean for UE and how much scope is there going forward for UE growth?

Let me answer this question. I have already expressed a part of it in my speech. 那就是我们在这段时间的优异提升 首先一方面是这个比单价的上涨 那意味着我们每单的收入会涨 因为这个我们的收入是跟比单价相关的 然后刚才我讲到我们的物流的这个效率 随着这个规模提升 是有一个显著的优化的 对 然后我觉得未来我们还是有很多的空间 那一方面就是我们的这个消费者 因为我们在过去的几个月 还是以新客大量新客为主 然后随着我们的这个 and the consumer's return to a higher annuality with our platform. In the process, we will also increase the bid price and then change our way of supplementing. At the same time, the traffic in the past few months in Taobao, including the rapid rise of the retail channel, It has become a channel for over 100,000 users every day. There is also a lot of commercial space in it. I think this is also a great opportunity for improvement in the future. Of course, the whole market is still a competitive market. We will also look at our opportunities according to the competitive situation in the market, and we will also adjust our strategy in a dynamic way.

Let me take this question and it's actually related to in part some of the things that I was sharing with you a bit earlier. So what we've been doing in this period of time is enhancing the user experience and at the same time increasing the average order value. So that means that the revenues attributable to each order will increase because our revenues are proportionate to average order value. I also spoke earlier about how we've optimized logistics, fulfillment logistics efficiency, and we'll continue to drive improvement there with scale. So I think going forward there is still considerable scope there on the one hand in respect of consumers because over the past few months it's really been primarily new consumers in this business. And what we're doing is converting those users into users with a higher level of stickiness across the platform as a whole. And through that process, we'll continue to increase average order size, average order value, and to modify the ways in which we provide subsidies. Also, if you look at traffic on the Taobao app over the past few months, including on the QuickCommerce channel, which is rapidly to the point where it now has over 100 million daily users on the channel. I think it speaks to the fact that there's considerable potential for monetization, further monetization, and I think that that's an opportunity also to improve UE in the future. Now, having said that, again, the market is a highly competitive market, So we will be looking at those opportunities, but adjusting our approach dynamically in line with market dynamics.

Thank you. Operator, let's move on to the next question.

Thank you. Your next question comes from Ronald Kim at Goldman Sachs. Please go ahead.

Thank you, Ellie, Toby, and Lydia. I would like to ask how we should think about the 3.8 billion yuan that we said we would spend in the next three years, especially the 1.2 billion yuan that has been spent in the past four seasons. How do we think about the ratio of CAPEX to incremental revenue?

Thank you. Thank you.

So I'd like to ask about CapEx over the next three years, and I'm wondering what your thinking is, as you sit here today, regarding the $380 billion figure that you previously mentioned, in particular because over the past four quarters, I believe $120 billion has already been spent. So how should we be thinking about CapEx going forward and the incremental revenue being driven by that CapEx? And how to evaluate the correlation between CapEx and the expected incremental revenue?

Thank you. Thank you for your question.

In fact, the 38 billion KPAC investment we mentioned before is actually a three-year plan of numbers. But from what we see now, as I mentioned just now, the speed of our service delivery is still far below the growth speed of customer orders. So from what we see now, if the supply chain problem, including the rhythm of the machine room, including our overall In terms of pricing, we meet the customer's needs as quickly as possible. In this case, if we can't meet the customer's needs as quickly as possible, we may not exclude further penetration. But because the supply chain rhythm of the entire market is also affected by various aspects. But overall, we will invest in the overall AI with a relatively positive attitude. Thank you for your question.

So the 380 billion CapEx figure that we had previously mentioned was a planned figure for a three-year period. But based on what we're seeing now, and as I just mentioned, the pace at which we can add new servers is insufficient to keep up with the growth in customer orders. So looking at the CapEx situation from where we're at today, and of course there are also supply chains issues to consider as well. The pace at which we can build up IDCs and launch new servers is also part of that consideration. But essentially, we're working as fast as we can to be able to satisfy all of that customer demand. In that context, if we're not able to satisfy all of that customer demand especially well with the current pace of investment, then the wouldn't rule out further scaling up that cat-back capacity. But again, that is somewhat dependent on supply chains and availability. But in overall terms, certainly we will be investing in AI infrastructure aggressively in order to meet demand customers. So in big picture terms, I would say that the $380 billion figure we had mentioned previously might be on the small side, certainly in terms of the customer demand that we're currently seeing.

Regarding how to estimate the increase in revenue brought by capital investment, I think it's not easy to estimate the current situation because the overall AI industry is still in its early stages. We can see that in terms of the use of our entire AI infrastructure, there are actually several different areas that are changing. For example, we have directly applied it to customers for training, and we have directly applied the server to customers for reasoning. We have also used our own server to do the reasoning of the layout chain. In fact, there are also There are also our internal applications, such as Gaode, Taobao, and Qwark, which are transformed into their member services or member products. So in general, our AI infrastructure is due to the use of these applications. Due to the different applications, the revenue and net profit levels So I don't think it's going to be as stable as it is now, but in the long term, we're actually more concerned about the quality of our infrastructure and the cost ratio of the token.

Thank you.

The second part of the question had to do with the incremental revenue being driven by these capital expenditures and if there's some kind of ratio that we can calculate between X amount of CapEx investment and X amount of incremental revenue. And I don't think it's really possible to make that kind of estimate, at least for the time being. because overall the AI sector is still in the early phases of its development. And if you look at the different ways that our AI infrastructure is currently being used, that's in flux and spans several different areas. For example, we have servers that are directly rented to customers for training. We have servers that are directly rented to customers for inference. And we're also, of course, using servers ourselves for Bylian for inference, as well as for internal applications within the Alibaba group, like AMAP, like Taobao, like QN and Quark, and transforming these applications into member services or membership-based applications. products for our users. So overall, AI products and AI infrastructure are being used in all these different ways, different kinds of applications, resulting in different revenues and different gross margin levels. So I think in terms of that kind of ratio you were asking about, whatever it is, it certainly wouldn't be stable. At this point, I think in the long term, though, what we care about more is that our infrastructure is serving high-quality tokens and providing good cost-effectiveness for those tokens.

Next question, please.

Thank you. Your next question comes from Billy James at Macquarie. Please go ahead.

感谢管理层接受我的提问 我有一个问题也是一个跟进的问题 就是作为一个我们现在全站的AI服务商 很明显目前我们还是处在一个比较重要的投资周期 并且这个投资也看到已经涵盖了整个价值链的不同环节 Considering the current wave of supply chain, how do we think about the allocation of resources? Especially in the model MAPS layer, maybe for the continuous construction of the underlying capabilities, or in terms of applications, we see that there is a more aggressive push-and-pull. In the current big environment, how do we evaluate the RYC of AI investment around training and reasoning in the entire industry? Thank you.

Thank you, management, for taking my question.

This is more of a follow-up question. The company is a full-stack AI service provider and obviously is currently in an important investment cycle. And we can see that the investments you're making cover a number of different segments in the value chain. You know, considering the instability in the supply chains that are ongoing at present, I'm wondering how you consider the allocation of our resources. Because you have the model as a service, the mass layer, we also continue to build out underlying capabilities, fundamental capabilities, and on the user-facing side, we have apps including Tony or QN and AMAP, products like that that we're iterating rapidly and scaling up to users. So I'm just wondering, in the present macro environment, how should we think about, how should we evaluate the return on invested capital, ROIC, in respect of AI and

and including both training and inference.

Okay, let me answer. As a full-time AI service provider, we have our products or our infrastructure in various areas during the big AI investment cycle. But in the priority level, There are several internal priority considerations. I can share some of them. I think our most critical priority is to ensure the overall training of our machine-made models. Because in general, our AI technology needs to have more customers or more high-value scenarios. Only when our model capabilities are continuously improved will we have the opportunity to unlock or get more users and to unlock more scenarios. After unlocking more high-value scenarios, in fact, the consumption of the entire token and the quality of the token, including the client's willingness to pay for these tokens, will actually increase gradually. So this is one of our higher priority areas. Another one is how to use推理. We think that for the推理 service on BaiLian, we also have a higher priority. Because we ourselves This hundred chain platform is actually a service that can serve global customers. How to make our AI resources can be used at a high efficiency for 24 hours to wind down stocks, to make a whole AI server better, to produce more tokens in 24 hours. This is what we think is a resource pool that magnifies hundreds of chains. For us, it is also a relatively high priority area. On the other hand, Whether it is our internal AI-based reasoning services, we also include these reasoning services for external customers. The demand is also behind this. But in the choice of external customers, there is actually a priority measure. If our external customer is fully aware of our Aliyun services, for example, he wants to use storage again, he wants to use big data again, Thank you. Let me take that question.

Indeed, as a full-stack AI service provider, We are currently in a very important investment cycle for AI and investing in our products as well as in our infrastructure. So there are several different places where we're investing, and we do have some internal thinking about how we prioritize them. And I can share, I think, some of those considerations with you. First of all, I would say the most critical of those priorities, the first thing that we need to ensure is that we are able to continually train our own foundation models. Because in the AI space overall, the ability of our AI infrastructure to be able to acquire more customers or to be able to acquire more high-value use cases relies on our ability to continually iterate and upgrade our foundation models. We need to be doing that in order to be able to unlock new demand and to acquire new customers by unlocking new use cases. After that, after unlocking new higher-value use cases, Then the next thing is to look at the amount of token consumption as well as the token quality as well as the willingness of customers to pay for those tokens. And that willingness is going to continue to strengthen gradually. So I would say that that is one of the highest priorities. when it comes to allocating those investments. Another priority is around inference. I'm thinking primarily of inference as a service on Bylian. That is also a relatively high priority area for us because we've created the Bylian platform in order to be able to serve customers all around the world. ensure that those AI resources are available 24 hours a day and are being utilized 24-7 with high efficiency. So the key there is to ensure that one AI server can run at full capacity 24 hours around the clock and thereby to generate more tokens. So by the end, it's a very critical resource pool for us, and it's a relatively high priority. Separately, of course, we have internal use cases for AI inferencing. And indeed, we also have external customers who are leveraging our inferencing services to meet their demand. So that's also part of the picture. But when it comes to these external customers, we also have some criteria for prioritizing different external customers. If an external customer is utilizing all of our services across cloud, all the values on the cloud services spanning storage, spanning big data, and all of these other things, then of course that customer would be accorded a higher level of priority. If you have a customer that's merely renting a GPU to meet some very simple inferencing needs, then the demands of those customers would accordingly be given a slightly lower level of priority.

Your second question is also very good. I think actually we see this question mainly from There are two ways of looking at it. The first is from the demand side. From the demand side, we see two very obvious advances. One is that we see the entire machine model, whether it is the basic model we see, or whether it is the video production model, or the future more like the full-modal model, the ability of this aspect is constantly improving. The whole scaling law trend is still not We say that in fact, the industry has not hit the wall yet. That is, we have seen a lot of breakthroughs from the progress of Gemini 3. In this field, we think that with the improvement of model capabilities, the whole industry can say that there are more and more things that AI models can do. There will be more and more scenes that can be adapted. In the scene of this adaptation, or in the process of improving the ability of these tasks that can be driven, there is also an improvement in the penetration rate of these tasks in various industries. So with these two dynamic movements, within three years, we can see that the need for AI is still a very decisive thing. In the case of the promotion of such a high real-time demand, we look at the supply section. The supply section, maybe you analysts have also seen it. Now, from the second half of this year, maybe the world, whether it's from FAB manufacturers to DRAM manufacturers, and then to the storage factory and then to the CPU, right? Each part of the entire AI server, each part is out of stock. And this wave, this wave of out of stock is driven by AI needs. In the future, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, the entire supply chain, we see that the growth of demand and supply is actually difficult to have a rapid improvement. So we think that in these three years, the overall AI resources are still in the state of supply and demand. And we actually also see that in our industry, including ourselves, including the American hype scalers, they are not only the new GPUs that are basically running out, but even the previous generation, Moving on to the second question, which I thought was a really good one.

I think there are two pieces to this issue. First is the supply side. Second is the demand side. Second is the supply side. So if we look at foundation models, and these could be video generation models, they could be omni-modal models going forward, but the capabilities continue to increase and be enhanced. And we're not yet seeing any issues in terms of scaling law. Nobody's hit the wall yet, so to speak, in the industry. We continue to make a lot of progress and make very important breakthroughs in terms of model capabilities. As the models become more powerful, then the AI models will be able to do more things in the real world. They'll be able to serve a larger variety of different use cases. And that will result in these models serving a lot of tasks. As their capabilities increase, they become stronger. And as these tasks become more deeply embedded across all industries, all aspects of business operations. So with those two drivers, we see in the next three-year period a highly definitive trend of demand for AI. And with all of this rapid growth in demand, We also need to be thinking about the supply side. I'm sure that you as analysts have also been looking at the supply side. Starting from the second half of this year, I think we've seen worldwide, if you look at fabs, if you look at DRAM vendors, storage companies, CPU manufacturers, across all of those different links in the value chain, that go to making AI servers, there is a situation of undersupply. Supply is unable to keep up with demand for all of these components globally. And I think that you can expect that to continue throughout this scaling up an investment cycle driven by real demand for AI. We know that the supply side is going to be a relatively large bottleneck. So I think that it could be at least a period of two or three years for those different suppliers, those different venues, to be able to ramp up their production capacity. So in this period of two to three years, we can expect to continue to see rapid increase in demand demand and not to be driving the supply side. So I think in the next three years to come, AI resources will continue to be undersupplied with demand outstripping supply. And what we can see internally in the industry, and if we look at the hyperscalers in the US, all of the latest GPUs that are running are running at full capacity, and not just them, the last generation GPUs, even GPUs from three to five years ago, so several generations back, those GPUs are to this day still running at full capacity. So, you know, looking ahead to the next, say, three years, we don't really see much of an issue in terms of a so-called AI bubble.

Well, let's take the last question. Please, operator.

Thank you. Your last question comes from Zhilong Shi from Nomura.

Thank you, Mr. Manager. Thank you, Mr. Manager, for accepting my question. I would like to ask, in the last phone call, Mr. Manager mentioned that 88 is currently in China's large consumer market to get a larger market share. Thank you.

So in the last earnings call, management shared that Alibaba intends to grow its market share in the consumption market in China. And we've seen that over the past few months, your investments in QuickCommerce have indeed resulted in an increase in market share. So I'd like to know, apart from QuickCommerce, apart from InstantCommerce, What are the other sub-sectors in the consumption market that you see as good opportunities for investment where you will consider scaling up your investments?

Let me answer this question. In fact, in the past few years, Ali has entered a lot of categories. In addition to what we are doing now, For example, we invested a lot in real-time sales this year. In fact, we are similar to Hema, including this offline commercial O2O model, including such as Feizhu, including such as high-quality local life, we all have such a layout. Now we still need to integrate more of our existing business, and then be able to open various businesses, and then make us have better co-efficient between businesses, and then to achieve such a Thank you.

This is Jiang Fan.

Let me take this question. You know, Alibaba has been investing strategically in the consumption market over many years, and we've entered a huge number of different categories and some verticals. So apart from QuickCommerce, which we've been investing in heavily, we've talked a lot about it, we also, of course, have Freshable, we have the offline O2O model, as well as Fliggy, as well as AMAP, and, of course, local services. So that's our landscape or matrix of businesses that we've been investing in. And I think what we need to be doing now really is working to integrate, connect those businesses, and to drive more synergies across those existing businesses. And in that way, we can achieve further increase in our market share in that larger consumption market. Thank you.

Thank you. Thank you everyone for joining us today. We look forward to speaking with you again on our December quarter earnings call.