Changes in the global AI computing supply landscape are reshaping Alibaba (09988)'s infrastructure investment plans. During the second fiscal quarter earnings call, Alibaba management revealed that due to changes in computing policies across different global regions, the company's long-term 20GW data center construction target may be revised downward, and the overall capital expenditure plan will be reduced accordingly. At the same time, Alibaba will accelerate the construction of its domestic chip supply chain centered on T-Head, and it is expected that after domestic computing products ramp up continuously in 2028, cloud business revenue growth is expected to recover to above 50%. Eddie Wu stated during the call that in the first quarter of 2027, T-Head's next-generation domestic chips will be fully capable of supporting ultra-large-scale model training. "In fact, we have already begun training the next-generation model on this year's generation of T-Head chips. As T-Head's domestic chip capacity continues to expand, in the long run, this will be a long-term support for Alibaba's model development. Because in the future, whether it is Alibaba, partners, or customers, the vast majority of models will rely entirely on domestic chips for training, whether it is pre-training or post-training."
More notably, Alibaba for the first time provided a cash flow breakeven timetable for this round of AI infrastructure investment cycle. Management expects that 2027 and the first half of 2028 will be the peak period of this round of computing investment, and with the large-scale expansion of the domestic computing supply chain, Alibaba is expected to achieve positive cash flow in the second half of 2028. This adjustment does not mean Alibaba is abandoning AI infrastructure expansion, but rather the company is reconfiguring its computing investment direction based on changes in the global chip supply environment: reducing dependence on overseas computing supply, increasing investment in domestic chips, and placing greater emphasis on return on investment for capital expenditure.
In terms of financial performance, Alibaba's revenue this quarter was 272 billion yuan, up 10% year-on-year; adjusted EBITDA reached 32.6 billion yuan, up 260% year-on-year; GAAP net profit was 64.4 billion yuan, up 213% year-on-year. Revenue from the AI, cloud, and computing services segment grew 53% year-on-year to 61 billion yuan, with external customer revenue growing 64%, accelerating for the tenth consecutive quarter.
20GW Computing Target May Be Lowered, Cash Flow Positive Expected in Second Half of 2028
Alibaba is reassessing its long-term computing construction scale and capital expenditure pace. During the call, regarding the impact of global chip supply restrictions on the capital expenditure roadmap, management stated that significant changes in computing policies across different global regions will have a certain impact on the long-term construction of global data centers. Based on preliminary assessments, the long-term 20GW construction target may be revised downward, and the overall capital expenditure plan will also be reduced accordingly. However, corresponding to the possible adjustment of the overall computing construction target, Alibaba will adopt a more aggressive pace to advance the construction of its domestic computing supply chain. This quarter, Alibaba's capital expenditure reached 50.5 billion yuan, approximately 7.5 billion US dollars. Although lower than the previous quarter's 67.7 billion yuan, it grew significantly year-on-year, mainly driven by accelerated computing scale ramp-up, rising prices of chips and other components, and increased CPU computing reserves to address rising AI agent customer adoption.
High-intensity investment is still consuming cash. This quarter, Alibaba's free cash flow showed a net outflow of 34.5 billion yuan, compared to a net outflow of 21.8 billion yuan in the same period last year. Management stated that the decrease in cash flow was mainly due to increased cloud infrastructure investment, partially offset by reduced losses in instant retail and increased prepayments from cloud service customers.
However, Alibaba has provided clear expectations for future cash flow improvement. Management stated that assuming the domestic computing supply chain based on T-Head continues to ramp up from 2027 and 2028 and progresses according to the established pace, the company is expected to achieve positive cash flow in the second half of 2028. The peak of this round of sustained investment is expected to be concentrated in 2027 and the first half of 2028. After 2028, as the domestic computing supply chain achieves large-scale and sustained expansion, the company will gradually move past the stage of highest AI infrastructure investment intensity.
This means that the key change in Alibaba's AI infrastructure investment cycle may not just be the adjustment of capital expenditure scale, but more importantly the transformation of the computing supply structure. The release of domestic chip capacity will become an important support for the company to control computing costs, reduce supply chain uncertainty, and improve cash flow.
Domestic Computing to Scale Up in 2028, T-Head's Next-Generation Chip V900 to Enter Mass Production in Q1 Next Year
Domestic computing will become an important engine for Alibaba's future cloud business growth. CEO Eddie Wu stated that Alibaba's overall planning for its long-term cloud business target in 2030 has always been built on the supply chain plan based on T-Head domestic chips. As capacity ramps up, the proportion of domestic computing in the company's overall computing will continue to increase and become the main engine supporting cloud revenue growth.
T-Head has now established a self-developed chip portfolio covering GPU, CPU, storage, and network chips. Among them, the Zhenwu M890, since its launch in August this year, has seen continuous growth in shipments of related super node instances and has now achieved large-scale commercial use. The next-generation Zhenwu V900 is expected to begin mass production in the first quarter of 2027, with computing performance reaching three times that of the M890. The next-generation super node based on V900 will support large-scale network interconnection and can handle ultra-large-scale model training and inference workloads, expected to achieve large-scale commercial use in the second quarter of 2027. In terms of subsequent product planning, Alibaba plans to expand the V900 supply chain and promote large-scale mass production in 2027, and advance the J900 in 2028.
The synergy between domestic chips and models is also accelerating. Eddie Wu revealed that Qwen has already been training its next-generation model on T-Head domestic chips, and Alibaba Cloud Model Studio has also been providing model inference services based on domestic chips. Management believes that deep integration between self-developed chips and models will continue to reduce unit computing costs and improve AI application development efficiency. This layout also constitutes the source of Alibaba's confidence in future cloud business growth.
Regarding investors' questions about the new global computing supply policy, management stated that the impact of related new regulations on next year's products and services still has significant uncertainty, but the progress of domestic computing has exceeded previous expectations. For the 2028 fiscal year, Alibaba has given a cloud business operating target growth range of 30% to 45%. Management also expects that as domestic computing product scale continues to expand in the 2028 calendar year, Alibaba Cloud business is expected to recover to above 50% high-speed growth at that time. Alibaba reiterated its 2030 calendar year target: achieving 100 billion US dollars in external cloud revenue, with long-term profit margin exceeding 20%.
AI Cloud Revenue Accelerates for Tenth Consecutive Quarter, Annualized Revenue Breaks Through 10 Billion US Dollars
Despite adjustments in computing supply and capital expenditure plans, Alibaba's AI and cloud business is still maintaining strong growth. This quarter, Alibaba's AI, cloud, and computing services segment revenue reached 61 billion yuan, up 53% year-on-year; external customer revenue reached 47 billion yuan, up 64% year-on-year, accelerating for the tenth consecutive quarter. Excluding computing equipment solution revenue, external customer recurring revenue grew 67% year-on-year to 37 billion yuan. AI-related product revenue reached 17.9 billion yuan, achieving triple-digit growth for the 13th consecutive quarter, accounting for 49% of Alibaba Cloud's external recurring revenue. The corresponding annualized revenue scale reached 71.7 billion yuan, breaking through 10 billion US dollars.
From the perspective of customer demand, growth comes not only from model training and inference at AI laboratories, but also from customers across various Chinese industries, including large state-owned enterprises, as well as fields such as smart manufacturing, automotive, embodied intelligence, and biotechnology.
Profitability is also improving. Alibaba management stated that the adjusted EBITDA of the AI, cloud, and computing services segment grew 121% year-on-year, with profit margin expanding, mainly benefiting from revenue growth in computing, storage, MaaS, and AI applications, as well as pricing optimization, market supply relief, and economies of scale.
At the same time, Alibaba is accelerating the commercialization of AI applications. The newly released Qwen4 Max has 2.4 trillion backbone parameters, continuously improving in agent tasks and coding; Apple Intelligence deeply integrated with Qwen is expected to launch for Chinese users in December. On the enterprise side, Qwen Work has launched an e-commerce version to help merchants achieve cross-platform operational automation; on the consumer side, the Qwen App is expanding to more than 20 daily life service scenarios.
Financial data shows that the AI laboratory and applications segment revenue this quarter was 4.4 billion yuan, up 51% year-on-year, with losses narrowing quarter-on-quarter to 10.5 billion yuan.
Instant Retail Continues to Reduce Losses, Expected to Achieve Profitability in Fiscal Year 2029
In addition to AI and cloud business, the profit improvement in Alibaba's consumer business is also an important support for this quarter's performance. This quarter, Alibaba's e-commerce group adjusted EBITDA grew 243% year-on-year to 38.9 billion yuan, mainly driven by significant loss reduction in the domestic instant retail business. Domestic instant retail business revenue reached 58.8 billion yuan, up 22% year-on-year, mainly driven by the growth of Hema's online and offline channels.
However, management acknowledged that since entering the fourth quarter, the domestic consumption environment remains under pressure, consumer willingness to spend has become more conservative, and competition in some categories is relatively intense. Platform operating indicators are generally consistent with China's physical goods online retail trends, with essential categories performing better than discretionary consumer categories.
In terms of instant retail, Alibaba stated that in the third quarter, online transaction value for the non-catering portion grew 40% year-on-year, and the front-warehouse business developed rapidly in categories such as food, fresh produce, and daily necessities. The unit economics model optimization for the food delivery business is in line with expectations, market share remains stable, and average order value continues to increase.
Regarding the future profit path, management expects that instant retail losses will continue to narrow in the December quarter, maintains the target of halving losses in fiscal year 2027, and plans to achieve instant retail business profitability in fiscal year 2029.
Earnings Call Transcript:
Alibaba CEO Eddie Wu: This quarter, Alibaba Group's revenue growth accelerated, and profitability continued to improve. Group total revenue in the AI, cloud, and computing services segment grew 10% year-on-year, with external revenue growth accelerating to 64%. At the same time, the year-on-year growth rate of external customer recurring revenue increased to 67%. Adjusted EBITDA margin rose to 14.5%. AI-related product revenue achieved triple-digit growth for the 13th consecutive quarter, with annualized revenue scale breaking through 10 billion US dollars.
The profitability of our various e-commerce businesses improved significantly. In instant retail (quick commerce), we maintained market share while significantly narrowing losses. The rapid improvement of AI model capabilities is accelerating the deployment of AI applications across industries, driving training and inference demand. The large-scale deployment of AI agents further boosts computing demand. The combined result of these factors is that global computing demand will continue to exceed supply in the coming years. This strong demand provides a solid foundation for the large-scale commercialization of AI.
Next, I will elaborate from four aspects: AI and cloud commercialization; T-Head's AI chips; models and AI applications; and our consumer business.
First, both revenue and profit growth in AI and cloud are accelerating. This quarter, external customer revenue for AI, cloud, and computing services grew 64% year-on-year, the tenth consecutive quarter of accelerating growth. Adjusted EBITDA grew 121% year-on-year, with EBITDA margin expanding to 14.5%. This performance was driven by growth in computing, storage, MaaS, and AI applications, as well as synergies. AI-related products generated 17.9 billion yuan in revenue this quarter, accounting for 49% of Alibaba Cloud's external recurring revenue. AI has become the core revenue pillar of our cloud business. The annualized revenue scale of AI-related products has broken through 10 billion US dollars.
This strong revenue growth reflects broad customer demand for our "AI + Cloud" services; in addition to training and inference demand from AI laboratories, we continue to see strong demand from customers across various Chinese industries, including large state-owned enterprises, as well as companies in smart manufacturing, automotive, embodied intelligence, and biotechnology.
Second, we continue to scale up T-Head's AI chips and strengthen supply chain capabilities. Starting with chips: T-Head has established a full-stack self-developed chip portfolio covering GPU, CPU, storage, and network chips. Since its launch in August, shipments of Super Node instances based on Zhenwu M890 have continued to climb and have now achieved large-scale commercial use. We expect to begin mass production of the next-generation Zhenwu V900 chip in the first quarter of 2027. The computing performance of Zhenwu V900 is three times that of M890; the next-generation super node based on Zhenwu V900 supports large-scale network interconnection and can handle workloads such as ultra-large-scale model training and inference. These super nodes will become the cornerstone of Alibaba Cloud's AI infrastructure. The Zhenwu V900 super node cluster will achieve large-scale commercial use in the second quarter of 2027. T-Head is accelerating its three-generation GPU roadmap. We have fully rolled out Zhenwu M890 this year; we plan to expand the supply chain and bring V900 into large-scale mass production in 2027, and advance J900 in 2028.
The infrastructure driven by our self-developed domestic chips is becoming the foundation of Alibaba's full-stack AI capabilities. It will increasingly support model training, inference, and agent workloads for Alibaba Group and Alibaba Cloud customers, supporting the group's AI revenue growth. Qwen has already trained its next-generation model on T-Head's domestic chips, and Model Studio has also been providing model inference services based on domestic chips. Our full-stack AI infrastructure will give us a clear advantage in this domestic computing supply chain; deep integration between self-developed chips and models will continue to reduce unit computing costs and help us achieve higher efficiency in AI application development.
Combining all the above factors, our 2030 calendar year target remains unchanged: achieving 100 billion US dollars in external cloud revenue, with profit margin exceeding 20%.
Third, we continue to advance models and AI agent applications. Our newly released foundation model Qwen4 Max has 2.4 trillion backbone parameters. We continue to improve its key capabilities, including agent tasks and coding, in which it is already among the world's leading models. Leading partners have also recognized Qwen's capabilities. Apple Intelligence deeply integrated with Qwen will launch for Chinese users in December. On the enterprise side, Qwen Work was released in August and quickly became a leading enterprise-level AI agent. We recently launched the e-commerce version of Qwen Work, with the proportion of paying e-commerce users continuing to rise. On the consumer side, the Qwen App is becoming a personalized assistant that helps users get things done, currently providing services covering more than 20 daily life scenarios.
Fourth, e-commerce business profitability improved significantly, with Alibaba's China e-commerce EBITA and profit margin both improving. This quarter, our e-commerce business achieved steady revenue growth and significantly improved profits. China e-commerce's adjusted EBITA and profit margin also improved significantly, reflecting its stable profitability. Our instant retail business maintained market share, losses continued to narrow, and unit economics improved quarter-on-quarter.
Two quarters ago, we began to see AI investment crossing the commercialization inflection point. Last quarter, growth accelerated and profit margins expanded. This quarter, these investments are translating into commercial results more quickly, with clearer returns. Our years of investment in computing and cloud infrastructure have brought us full-stack product advantages and economies of scale, which are becoming Alibaba's key competitive advantages in the AI industry. They allow us to flexibly respond to industry changes and deliver AI products at scale to meet customers' rising AI demands.
Looking ahead, we will continue to invest in AI infrastructure to expand our leading edge and strengthen long-term competitiveness.
Alibaba CFO Toby Xu: Welcome to Alibaba Group's second quarter earnings call. This quarter, thanks to our strategic focus on "AI + Cloud growth" and the profitability sustainability of our consumer business, the company achieved strong growth in both revenue and profit. AI, cloud, and computing services revenue grew 53% year-on-year, with EBITDA margin rising to 14.5% quarter-on-quarter. Recurring revenue from external customers accelerated to 67% year-on-year growth, and this strong growth reflects a broader demand base for "AI + Cloud" services. On the consumer side, e-commerce group EBITDA grew 243% year-on-year this quarter, mainly driven by significant loss reduction in instant retail. In addition, domestic e-commerce business contributed stable operating profit, global wholesale business profit was strong, and we narrowed losses in international e-commerce business.
Now let's look at this quarter's financial results. Total revenue grew 10% year-on-year to 272 billion yuan, driven by strong momentum in cloud business and instant retail. Adjusted EBITDA increased 260% to 32.6 billion yuan, mainly attributable to improved operating results in Taobao instant retail and cloud business, as well as improved operational efficiency across businesses. GAAP net profit was 64.4 billion yuan, up 213% year-on-year, mainly driven by increased net investment gains and improved operating profit. Operating cash flow performance was strong this quarter, growing 55% year-on-year to 15.7 billion yuan.
Capital expenditure this quarter was 50.5 billion yuan (approximately 7.5 billion US dollars). This is due to continued investment in AI infrastructure to meet strong and accelerating customer demand. The significant year-on-year increase was due to accelerated computing scale ramp-up, rising prices of chips and other components, and increased CPU computing reserves in anticipation of rising AI agent customer adoption. Compared to the previous quarter's 67.7 billion yuan (approximately 10 billion US dollars), it decreased, reflecting fluctuations in the procurement cycle. We will maintain a prudent investment pace, continuously tracking and evaluating the expected return on invested capital (ROIC) for future capital expenditure.
Free cash flow showed a net outflow of 34.5 billion yuan, compared to a net outflow of 21.8 billion yuan in the same period last year. The decrease in cash flow was mainly due to cloud infrastructure investment, partially offset by reduced losses in Taobao instant retail and increased prepayments from cloud service customers. Compared to the previous quarter's net outflow of 44.7 billion yuan, it improved, reflecting our prudent cash management. As of September 30, 2026, we continued to maintain a strong net cash position of 32.9 billion US dollars, up from 30.7 billion US dollars in the previous quarter. Excluding debt with maturities exceeding 5 years, our net cash position was approximately 49 billion US dollars. Strong operating cash flow and net cash position, combined with realized investment gains, give us confidence and financial flexibility to invest in long-term growth. We will maintain prudent investment in AI capabilities, focus on return on invested capital, and aim to achieve positive free cash flow.
Now let's look at results by segment. Alibaba e-commerce group revenue remained stable, with CMR (customer management revenue) declining 5% year-on-year; excluding the impact of new business development plans on domestic revenue, it grew 1% on a comparable basis. Domestic instant retail business revenue was 58.8 billion yuan, up 22% year-on-year, mainly driven by strong growth in Hema's online and offline channels. Alibaba e-commerce group adjusted EBITDA grew 243% to 38.9 billion yuan, mainly driven by significant loss reduction in the domestic instant retail business. Taobao instant retail continued to improve unit store economics, reflected in increased average order value. Platform e-commerce profitability was stable, relying on stable operating profit from domestic e-commerce, reduced losses in international e-commerce business, and strong growth in global wholesale business.
Now let's look at business updates and results for AI, cloud, and computing services. Segment total revenue was 61 billion yuan, up 53% year-on-year; external customer revenue was 47 billion yuan, accelerating to 64%. External customer recurring revenue (excluding computing equipment solution revenue) grew 67% to 37 billion yuan. Driven by strong demand for AI computing and core cloud services (including storage, computing, security, and networking), customers continue to expand their use of our AI cloud services. AI-related product revenue this quarter was 17.9 billion yuan, corresponding to an annualized revenue scale of 71.7 billion yuan. This is the 13th consecutive quarter of triple-digit growth, accounting for 49% of external recurring revenue. Adjusted EBITDA margin expanded to 14.9%, benefiting from pricing optimization, market supply relief, and operating leverage from scale. We reiterate our 2030 calendar year guidance: external cloud revenue reaching 100 billion US dollars, with long-term profit margin exceeding 20%, driven by expanding enterprise adoption and AI products being more deeply embedded in core business processes.
Next, let's look at the AI laboratory and applications segment. Segment total revenue was 4.4 billion yuan, up 51% year-on-year, mainly driven by rising demand for native AI products. Segment losses narrowed quarter-on-quarter to 10.5 billion yuan. We observed that the Tongyi App improved efficiency in both model training and marketing investment. We are also continuously upgrading our AI agent products for enterprises and consumers through innovations in Tongyi Work and the Tongyi App. In October, we launched the e-commerce version of Tongyi Work, which is integrated into Tongyi Work and can achieve end-to-end cross-platform operational automation for e-commerce merchants. In addition, we launched Qingo — a cloud-based personal agent embedded in the Tongyi App.
Revenue from other segments declined 5% year-on-year to 27.7 billion yuan, mainly due to decreased Cainiao revenue. Adjusted EBITDA for other segments was a loss of 3.3 billion yuan, mainly due to increased investment in technology businesses.
In summary, these strong financial results demonstrate that our co-designed, deeply integrated AI technology stack continues to advance, combined with the profitability sustainability of our consumer business, enabling us to well capture increasingly diverse demand across industries and create greater value for shareholders. Building on this momentum, we will continue to balance prudent growth investment with focus on return on invested capital, proactively manage cash, and accelerate progress toward positive free cash flow.
Q&A Session:
Q: Good evening management, thank you for taking my question. My question is about cloud revenue outlook. First, thank you very much for the guidance management provided, which is achieving 100 billion US dollars in cloud revenue by 2030. But considering the latest BIS announcement, how should we view the cloud revenue growth path for the remaining two quarters of this fiscal year? Also, are we confident in achieving more than 50% revenue growth in the next fiscal year, which is fiscal year 2028? If we have such confidence, considering the current situation, where does this confidence come from and what is the basis? Thank you.
A: Thank you for your question. In our overall planning for the 2030 long-term target, our AI infrastructure has always been based on the T-Head domestic chip supply chain plan. So as capacity ramps up, the proportion of domestic computing in our overall computing will continue to increase, becoming the main engine supporting cloud revenue growth, and correspondingly reducing dependence on overseas computing. Currently, AI computing is still in short supply globally, especially in China, and we think it will be particularly tight in the next two years. As the AI full-stack products and supply chain centered on T-Head continue to advance and scale up, I think the previously mentioned target of achieving 100 billion US dollars in cloud external revenue and over 20% profit margin in the 2030 calendar year still has strong visibility.
As for next year's situation, we believe the new global computing export control regulations affect next year's products and services, and the entire global supply chain has significant uncertainty; but on the other hand, our progress in domestic computing has exceeded original expectations. So considering various situations, there may be some impact on next year's operating plan targets, but we are still evaluating. Considering all factors, we expect that for the next fiscal year, the cloud business operating target range will fall in the 30% to 45% growth range. At the same time, as domestic computing product scale continues to expand in the 2028 calendar year, we expect to recover to above 50% high-speed growth in the 2028 calendar year.
Q: Given the current global chip supply restrictions, how should we view our capital expenditure roadmap? How will we achieve the 20 gigawatt computing target in the future? Of course, the current situation is still evolving, and the pace of capital expenditure may also change accordingly, but is it possible that we could achieve positive free cash flow earlier than originally expected? Thank you.
A: Computing policies across different global regions have indeed undergone significant changes, which will have a certain impact on the long-term construction of global data centers. Based on our preliminary assessment, the long-term 20 gigawatt construction target may be revised downward, and the overall capital expenditure plan will also be reduced accordingly. But at the same time, we will adopt a more aggressive investment and advancement pace in the domestic computing supply chain. In the current preliminary assessment, assuming the domestic computing supply chain based on T-Head continues to ramp up from the 2027 and 2028 calendar years and progresses according to the established pace, we expect to achieve positive cash flow in the second half of 2028. That is, the peak of this round of sustained investment is concentrated in 2027 and the first half of 2028; after 2028, as the domestic computing supply chain achieves large-scale and sustained expansion, we will move past the stage of highest AI infrastructure investment intensity. Therefore, we remain very confident about long-term cash flow planning.
Q: Regarding the core e-commerce business. Given the current weak macroeconomic environment and your cautious guidance for the fourth quarter, what is the outlook for Taotian Group for the December quarter, especially regarding CMR growth and the adjusted EBITA development path? Regarding instant retail, could you provide an update on the December quarter and your timetable for further loss reduction next year? Thank you.
A: Let me answer this question. First, in the third quarter, our overall e-commerce segment performance was in line with expectations, and domestic e-commerce remained stable. The instant retail segment maintained good year-on-year growth while significantly reducing losses; profitability levels of other business segments also improved noticeably. As you just mentioned, we also see that since entering the fourth quarter, the overall domestic consumption environment remains under pressure, consumer willingness to spend has become more conservative; at the same time, we have been facing relatively intense market competition in some categories. So since the fourth quarter, the platform's main operating indicators have been consistent with China's physical goods online retail trends. In the current consumption environment, essential categories are showing better trends than discretionary consumer categories, and the platform will further strengthen the competitiveness of essential categories in daily sales.
Instant retail remains the core growth driver for the platform in the coming years. In the third quarter, online transaction value for the non-catering portion of instant retail grew 40% year-on-year, leading the industry, especially with rapid development in the front-warehouse business. In categories such as food, fresh produce, and daily necessities, instant retail continues to change users' consumption habits and drive growth in related categories. In food delivery, our unit economics optimization is in line with expectations, market share remains stable, and average order value continues to rise. In the December quarter, we expect instant retail losses to continue to narrow, and we continue to maintain the target of halving losses next year, and achieve instant retail profitability in fiscal year 2029.
Beyond B2C e-commerce and instant retail, other businesses in our e-commerce segment, such as B2B and Xianyu, have also continued to make progress over the past two years. This quarter, we also integrated the international and domestic business segments of B2B, connecting domestic supply with overseas demand. With the continued support of AI technology, we believe this integration will further unlock greater potential in the B2B segment. Xianyu achieved 29% order growth year-on-year last quarter and has been profitable for multiple consecutive quarters. We also believe these businesses will bring sustained revenue growth and profit contribution to the overall e-commerce segment in the future.
Q: Thank you management for taking my question. My question has two parts. First, could you discuss how the latest US Commerce Department restrictions will affect model training? Will it affect the iteration speed of our flagship Qwen model? Second, related to this, as we understand, part of your funding for expanding data center capacity will come from your partners, that is, your customers. Could you share some updates on this? Have you already started reaching such deals with some customers to validate this concept? Thank you.
A: Thank you for your question. I think the new computing access regulations are being implemented, and there are many detailed rules. Our current situation is that we use computing globally and have always complied with applicable laws and regulations, so the specific impact on our products and finances may require some specific information to make a more complete assessment. But if it affects the progress of some subsequent products, I think it will also be a relatively short-term impact, because in the first quarter of 2027, our own domestic chips based on T-Head will be fully capable of supporting ultra-large-scale model training. We have actually already begun training the next-generation model on this year's generation of T-Head chips. As T-Head's domestic chip capacity continues to expand, in the long run, this is more of a long-term support for our model development. Because in the future, including ourselves, our partners, and our customers, the vast majority of models will rely entirely on domestic chips to complete the entire training, whether it is pre-training or post-training. So as our capacity continues to expand in 2027 and 2028, the pace of our long-term continuous model iteration will actually be more controllable.
Q: I mainly want to learn more about T-Head. We have heard that T-Head's domestic chip capacity continues to expand, which is encouraging, and there is also a very good future product pipeline. I want to know, in the next few years, if we talk about incremental training and incremental computing supply, how much of that proportion will go to Qwen? And a follow-up question: could management introduce the progress of T-Head's spin-off and listing? When relevant details are announced, how should we correctly assess T-Head's internal and external resource allocation, and what impact will this have on cloud business profit margins?
A: Given this new situation, we believe that in 2027 and 2028, that is, after 2027, most of our newly added AI chip supply should be mainly our self-developed domestic T-Head chips, and we will also purchase some other domestic chips to support some of our internal inference needs. So from this perspective, our own flagship chips will prioritize meeting the training needs of large models, including ourselves and our important deep partners. Therefore, this will be a relatively balanced allocation logic: first, prioritize ensuring our own model iteration; on the basis of model iteration, we will also consider the needs of important partners and customers; at the same time, at the inference level, we will try to purchase more chips of different compositions, including many other brands of domestic chips. This is roughly our situation.
Regarding the impact on cloud business profit margins, let me explain from several levels. First, in 2027 and 2028, in the Chinese domestic market, computing may be the most scarce in these two years. During these two years, due to computing scarcity, overall, the pricing and gross margin of overall computing can be relatively guaranteed. On the other hand, since they are all our self-developed chips overall, combined with our continuous scale expansion, we also have a relatively good advantage in procurement and supply chain. So compared to using externally purchased chips for cloud business and AI business, our profit margin will be a relatively positive improvement.
Regarding the question of T-Head's spin-off and listing, let me answer. We recently made an announcement that T-Head has submitted a listing application to relevant regulatory authorities on a confidential basis. Due to listing compliance requirements, we cannot disclose too much T-Head-related financial information at this stage, including other undisclosed information. But what can be confirmed is that the proportion of domestic computing in Alibaba Cloud's overall computing will continue to increase, and this trend is structurally positive for the company's entire cloud business.
Q: We are pleased to see that the losses of our AI laboratory and applications segment narrowed quarter-on-quarter. At the same time, we see many other large Chinese internet companies actually expanding their scale of investment, especially at the model and application layers. I want to know how management views where the ultimate value of AI will primarily reside? Will it mainly be at the application and agent layer? If so, how should we allocate investment in Qwen? How should we understand the differentiation between Qwen Enterprise Edition and other Tongyi models? How will this differentiation affect the scale and distribution of investment, and the loss trajectory for the next few quarters or the next one to two years? Thank you.
A: Thank you for this question. Actually, I think in the industry now, different people have different views, and different companies also have different views. We now feel it is still the early stage of AI, and whether applications or models are more important in the future, I think these two need to be viewed together, because in the future, models and applications will definitely need to develop jointly, or rather, develop synergistically together. Models help applications, and applications can also synergize with model growth. But if an application does not have a stronger model, or does not have a uniquely stronger model to develop jointly with it, I think the long-term user retention of this application is still questionable. Just like, if an application uses the same models as everyone else behind it, then the replaceability of this application will also be very strong. So from our perspective, in the future, models and applications still need to develop jointly to create more unique product value, or more unique product competitiveness, so that they can be retained long-term within enterprises or on customers' desktops. So at present, our strategy is to emphasize joint development of applications and models, and how to enhance stronger product capabilities. So in the current relatively aggressive user promotion, or relatively aggressive user subsidies, we are still relatively restrained. In the long run, we are more optimistic about a unique model and a unique application creating a unique value for customers, which is a value that is relatively difficult to replace. Such a combination of application and model can be a relatively useful long-term value.
Q: Regarding our annualized revenue trend, since we mentioned a latest achievement earlier, which is that annualized revenue broke through 20 billion yuan in August. I would like to understand more specifically, could you introduce the full-year target? By the end of the fiscal year, do you think annualized revenue can reach 30 billion yuan? In this regard, I believe management is closely monitoring the annualized revenue trend and analyzing the competitive landscape. Could you introduce the current competitive landscape? The second question is about overseas strategy. Could you introduce the outlook for the proportion of domestic and overseas revenue in the cloud business? Thank you.
A: Thank you for your question. Overall, model-as-a-service business revenue is still relatively dependent on the launch of new models. In the past two months, domestically, including ourselves and our partners, the overall pace of new model launches has been somewhat slower than we originally expected. Our own Qwen 4 has just been released, and we also have high expectations for our next-generation video model, which will also begin to be released soon. So as our new models and our partners' new models are gradually released, our annualized revenue growth will begin quickly, with relatively fast growth, which we expect in December. So the full fiscal year target of 30 billion yuan in annualized revenue remains unchanged.
Regarding the impact on overseas cloud business, we are also currently assessing the impact of computing export rules on overseas cloud business, and the specific situation still needs some time to analyze and evaluate. But currently Alibaba Cloud operates 109 availability zones in 32 regions globally, including our data center in the Netherlands, as well as data centers in France and Brazil, which are gradually opening. In the future, Alibaba Cloud's overseas business will continue to be our ongoing focus. But in future overseas business, the AI portion will have various different cooperation methods, perhaps requiring more overseas partners, and even in the long term, our own chips for export is also one of our feasible options in the future. Of course, these may all need to be based on meeting domestic demand. In the long run, for full-stack infrastructure like Alibaba Cloud, first serving our Chinese companies going overseas and these overseas customers, we maintain relatively strong confidence in the long-term prospects of this business.
Q: I would like to continue asking about cloud business profit margins. Could management elaborate on the core drivers of the significant improvement in cloud profit margins this quarter? How should we understand the trend of cloud profit margins in the coming quarters? Also, given the constraints of the current computing supply environment, does Alibaba Cloud have the opportunity to raise cloud product pricing to match strong demand? Also, under the current changes in the computing environment, is there an opportunity for AI laboratory losses to narrow more quickly?
A: There are several factors driving the continuous improvement of AI cloud business profit margins. First, the proportion of high-margin AI-related products in recurring revenue continues to increase, directly boosting the overall cloud business profit margin. Second, against the backdrop of global computing scarcity and tight domestic computing supply, our bargaining power with customers has strengthened, and the gross margin space for most products has improved compared to last year. Third, as data centers continue to expand, economies of scale continue to be released, reflected in supply chain bargaining power; at the same time, after customers purchase multiple products together, migration costs rise and customer stickiness increases, which continues to benefit profit margins. In the long run, AI technology development, especially AI cloud equipped with self-developed chips, brings structural changes to Alibaba Cloud. Under the full-stack service of self-developed chips + self-developed large models, Alibaba Cloud's products are no longer standardized commodities. It is difficult for customers to directly compare prices and bid like standardized cloud resources. After communicating with many customers, we found that in the AI era, Alibaba Cloud is no longer just a supply chain procurement party for customers, but a partner that jointly develops products with customers. This product differentiation is difficult to directly compare prices, and is a structural factor supporting long-term improvement in cloud profit margins.
Comments