Wu Yongming projected that in the June quarter, the annualized recurring revenue (ARR) from AI models and application services will surpass 10 billion yuan, and by the end of the year, it will exceed 30 billion yuan. It is anticipated that within the next year, the revenue share from AI-related products will surpass 50%, becoming the primary driver of Alibaba Cloud’s revenue growth. Compared to 2022, before the surge in large-scale model development, the scale of data centers to be built in the future will experience growth of more than tenfold.
Facing the historic opportunities brought by AI, Alibaba is at a critical juncture where technological dividends are being transformed into commercial dividends.
At the Q4 2026 earnings call for analysts, Alibaba CEO Daniel Zhang stated: 'AI is driving a comprehensive upgrade across Alibaba Cloud's businesses, with growth momentum shifting entirely from traditional computing and storage to models, computing power, and Agent services.'
Daniel Zhang forecasted that by the June quarter, the annualized recurring revenue (ARR) from AI models and application services, including the Model as a Service (MaaS) platform Bailian, will exceed 10 billion yuan and surpass 30 billion yuan by the end of the year. 'The high-profit margin advantage of this revenue stream is becoming increasingly prominent and will serve as a pillar for our future healthy and high-quality revenue growth.'
In the first quarter, revenue from AI-related products accounted for more than 30% of Alibaba Cloud's external commercialization income for the first time, with quarterly revenue reaching 8.971 billion yuan. It is projected that within the next year, the proportion of revenue from AI-related products will exceed 50%, becoming the primary driver of Alibaba Cloud's revenue growth.
To meet the enormous demand for AI, Alibaba announced that its future capital expenditures will exceed the originally planned 380 billion yuan. Daniel Zhang revealed that compared to 2022, before the large model boom, the scale of data centers to be built in the future 'will essentially represent more than a tenfold increase over 2022.'
It was reported that Pingtouge, a subsidiary of Alibaba, has achieved scaled mass production of its self-developed GPU chips, with more than 60% of the computing power already serving external commercial clients.
The inflection point for AI commercialization has arrived, and the demand for Tokens is 'almost limitless.'
The market is highly focused on the actual monetization capabilities after the implementation of large models. During the conference, Daniel Zhang disclosed that the annualized recurring revenue (ARR) from AI models and application services, including the Bailian MaaS platform, is growing rapidly and has recently surpassed 8 billion yuan. He projected that by the June quarter, this figure will exceed 10 billion yuan 'and reach 30 billion yuan by the end of the year.'
This exponential surge is primarily driven by enterprises' demand for complex workloads, particularly the leap in AI Coding (Code Intelligence) capabilities. Addressing concerns about the low willingness of domestic enterprises to pay for SaaS, Daniel Zhang offered a striking assessment: 'As long as the value created within an enterprise by the tasks completed exceeds the token cost, the demand for API tokens will be virtually limitless.'
Daniel Zhang emphasized that compared to traditional SaaS businesses, MaaS businesses currently enjoy higher gross margins and remain in short supply. 'The high-profit margin advantage of this revenue stream is gradually becoming evident and will serve as a pillar for our future healthy and high-quality revenue growth.'
There is no overcapacity in computing power: 'Not a single card is idle.'
In response to concerns that the substantial capital expenditure on AI infrastructure could weigh on free cash flow, Alibaba's management conveyed unwavering resolve. CFO Xu Hong stated that, with a net cash position exceeding USD 59 billion (excluding debt maturing in over five years), the group will firmly invest in the critical opportunity window over the next two years.
Regarding return on investment, Wu Yongming compared AI to the manufacturing industry, emphasizing the core role of building two major data center factories: 'AI training' and 'AI inference.' 'Whether it’s commercialized services for cloud IaaS or generating additional revenue atop models via the Maas platform and AI-native software, we have almost no idle cards during the service period.' He added, 'Given the anticipated demand over the next three to five years, our significant investments in AI data center construction will yield highly predictable returns.'
To support the enormous future demand, Alibaba has provided an ambitious capacity expansion plan. Wu Yongming revealed that, compared to 2022 before the surge in large models, the scale of data centers to be built will experience growth of more than tenfold.
In light of the rapid investment in expanding computing power, Alibaba's capital expenditure will exceed its previously planned RMB 380 billion.
The competitive moat of full-stack technology is evident, with significant improvement in cloud business gross margins.
Under the physical bottleneck where long-term AI demand remains challenging to fully satisfy, Alibaba’s full-stack technological advantages and economies of scale are becoming increasingly apparent.
Wu Yongming pointed out that due to soaring reset costs of newly built servers (increasing by over 100% compared to two years ago), this will positively influence asset pricing for cloud services. Combined with the inherently high gross margin of MaaS operations and continuous optimization of inference technologies, 'the revenue and gross profit levels generated by the same server on the Bailian platform surpass those of our traditional cloud computing's straightforward computing power services.'
At the foundational computing level, Pingtouge's self-developed GPU chips have achieved scaled mass production, with 'over 60% of the computing power serving external commercial clients.' Wu Yongming stated that as the penetration rate of Pingtouge’s full-stack self-developed chips increases, leveraging the significant gross margin differential between domestic and international chips, Alibaba aims to provide China’s most cost-effective inference platform. 'We anticipate a notable improvement in Alibaba Cloud’s gross margin within the next one to two years.'
Consumer business rebounds, with instant retail expected to achieve overall profitability.
Beyond the "AI + Cloud" domain, Alibaba's consumer business has also seen substantial recovery. This quarter, on a comparable basis, core customer management revenue (CMR) grew by 8% year-over-year.
Regarding the near-field e-commerce (instant retail) segment, management disclosed that the overall order volume from January to March this year was 2.7 times that of the same period last year. Through improvements in logistics efficiency and an increase in average order value (AOV), unit economics (UE) have significantly improved. Management anticipates that, while maintaining scale and market share, they are confident in achieving positive UE before the end of the new fiscal year, ultimately moving toward overall profitability. This will not only substantially narrow losses but also provide stronger operating cash flow support for the group's future AI investments.
The following is a transcript of the earnings call (assisted by AI translation).
CEO Daniel Zhang:
Dear investors, welcome to Alibaba Group’s fourth-quarter earnings call for the fiscal year 2026. Over the past quarter, Alibaba’s high-intensity investments in its two strategic priorities—"AI + Cloud" and "consumer"—have rapidly translated into tangible business results, with group revenue growing by 11% year-over-year.
This quarter, the external revenue growth of the Cloud Intelligence Group accelerated to 40%, with AI-related product revenue achieving triple-digit growth for the eleventh consecutive quarter. On a comparable basis, core customer management revenue (CMR) in China's e-commerce business increased by 8% year-over-year, while the near-field e-commerce business achieved significant improvements in unit economics while maintaining market share. We are at an inflection point in the evolution from conversational chatbots to autonomous AI agents, which is directly driving explosive growth across our three core workload categories: training, inference, and agent orchestration.
Against this backdrop, Alibaba's AI has moved past the initial investment phase and begun large-scale commercialization. Next, I will elaborate on four key aspects: AI commercialization, cloud infrastructure, the AI application ecosystem, and our consumer business. First, the tipping point for AI and cloud commercialization has arrived. This quarter, the annualized revenue from AI-related products in the Cloud Intelligence Group exceeded RMB 35.8 billion and continues to maintain triple-digit growth. Revenue from AI-related products now accounts for 30% of the Cloud Intelligence Group’s external revenue. We expect that within approximately one year, revenue from AI-related products will surpass the 50% threshold, becoming the primary driver of cloud business revenue growth.
Therefore, given the certainty of long-term AI demand and our full-stack technological advantage, the external revenue growth of the Cloud Intelligence Group is expected to continue accelerating beyond the current 40% level over the next few quarters. We anticipate this trajectory will remain robust in the medium to long term. This reflects the role of Alibaba Cloud in driving comprehensive value upgrades across its entire business, with its growth focus having fully shifted from traditional computing and storage to models, AI computing, and agent services.
We are also witnessing exponential growth in revenue from AI models and application services, driven by foundational model services and AI-native software as new revenue streams. Over the past three months, token consumption on our model service platform has grown significantly quarter-over-quarter, as enterprise customers accelerate their transition from simple tasks to production-grade scale and complex workloads, fueling continued demand for model and application services. On the Model Foundry platform, we expect the annualized recurring revenue (ARR) from model and application services, including the model platform, to exceed RMB 10 billion in the June quarter and reach RMB 30 billion by year-end.
The healthy margin profile of this revenue stream is becoming increasingly evident, making it a source of healthy, high-quality growth. Second, our AI infrastructure supports our full technology stack and constitutes a lasting competitive moat. Our self-developed GPU chips have achieved scaled mass production, with more than 60% of computing power now serving external customers across verticals such as internet, financial services, and autonomous driving.
As the only AI cloud service provider in China capable of delivering self-developed AI chips at scale, we have ensured the autonomy and controllability of our computing supply chain in an environment of scarce computing resources, while providing customers with highly competitive AI inference and training services. This structural advantage is conducive to our revenue growth and gross margin improvement. Meanwhile, our cloud products are accelerating their AI-driven upgrades.
The surge in agent workloads has significantly increased demand for traditional cloud products built around CPUs, storage, and containers. We are upgrading these products into infrastructure solutions optimized for the agent era. Thirdly, at the application layer, we have constructed a complete closed loop encompassing AI-native software to a full-fledged agent ecosystem. Alibaba Token Hub (ATH) continues to launch new products that bridge consumer and enterprise environments, achieving breakthrough progress in AI-native software and code agents.
The Tongyi Qianwen model continues to iterate in reasoning, coding, and agent capabilities. On the enterprise side, we have launched a series of products covering intelligent workplace tools, AI coding, and business operation management, helping businesses unlock and enhance productivity. On the consumer side, the Tongyi app fully integrates the commercial service capabilities of Taobao and Tmall.
On May 7, the Tongyi app has now been deeply embedded in the ecosystem, covering Taobao, Alipay, AutoNavi Maps, and Fliggy, making it China’s first one-stop personal assistant that seamlessly connects daily life, productivity, and learning. Fourthly, in our consumer business and at the group level, we prioritize learning from value creation.
Beyond AI, our consumer strategy continues to advance steadily, with CMR growth rebounding significantly this quarter. On a comparable basis, CMR grew by 8% year-over-year. We continue to improve user experience and merchant operational efficiency, with our proximity e-commerce business achieving significant unit economic benefits while maintaining stable market share and scale.
In summary, the return on our investments in AI, cloud, and consumer businesses is becoming increasingly clear. Revenue growth in AI + Cloud is accelerating, accompanied by margin improvement. The ARR for models and application services continues to grow, and the operational efficiency of our consumer business keeps improving.
Facing the historic opportunity represented by AI, Alibaba is at a critical juncture where technological investments are beginning to yield commercial returns. We will maintain strategic focus and leverage our full-stack AI capabilities to support long-term growth.
CFO Xu Hong:
Our strategic priorities remain firmly focused on AI + Cloud and consumer businesses. Multiple growth catalysts, including technological advancements and business innovations, are creating strong tailwinds in the AI + Cloud domain. Our full-stack models, computing frameworks, and applications hold established leadership positions at every layer.
The robust growth of our AI + Cloud business and the clear monetization path of our model platform give us confidence to make significant investments to expand our leadership in the consumer space. This quarter, we achieved strong CMR growth on a comparable basis.
In our near-field e-commerce operations, we continue to achieve quarter-over-quarter improvements in unit economics. Turning now to this quarter's financial results. On a consolidated basis, total revenue was 243.4 billion yuan. Excluding the revenue contributions from Intime and Sun Art, comparable revenue growth would have been 11%. Adjusted EBITDA declined by 84%, primarily due to our strategic investments in technology businesses, near-field e-commerce, and user experience, partially offset by improved operating performance driven by the continued growth of consumer management services within the cloud business and enhanced operational efficiencies across various segments.
Our net profit was 23.5 billion yuan, representing a year-over-year increase of 96%, mainly attributable to higher gains from equity investments measured at fair value compared to the same period last year, as well as losses from the disposal of Intime and Sun Art in the prior-year period, partly offset by the decline in adjusted EBITDA. Net cash inflow from operating activities was 9.4 billion yuan. Free cash flow was an outflow of 17.3 billion yuan.
We are reinvesting our operating cash flow to strengthen our competitive advantage in artificial intelligence (AI). As of March 31, 2026, our cash holdings amounted to approximately $38 billion, including debt with maturities exceeding five years. Our net cash position stands at around $59 billion.
This strong balance sheet gives us confidence to invest for growth. Now turning to our consumer business. Revenue from Alibaba's China e-commerce segment was 122 billion yuan, growing by 6%. Customer management revenue (CMR) increased by 1%.
To support merchants in growing their businesses on our platform and encourage higher spending, this quarter we upgraded our business development program for select merchants. Under this initiative, subsidies provided by the platform to these merchants are directly linked to their marketing expenditures on our platform. For accounting purposes, these subsidies, previously recorded as sales and marketing expenses, are now classified as deductions from CMR. As a result, CMR grew by 1% year-over-year this quarter. Excluding the accounting impact of this program on revenue, CMR would have grown by 8% on a comparable basis.
Our near-field e-commerce business achieved revenue growth of 57%, reaching 20 billion yuan. The near-field business further improved unit economics and increased the average order value quarter-over-quarter, primarily driven by portfolio optimization. Adjusted EBITA for Alibaba's China e-commerce group was 24 billion yuan, declining by 40%, mainly due to investments in near-field e-commerce, user experience, and technology.
Meanwhile, customer management services made a positive contribution. Excluding losses from the near-field e-commerce business, the EBITA of Alibaba's China e-commerce group would have remained stable year-over-year and shown sequential improvement due to significant investments, user retention, and enhanced user experience. AIDC revenue grew by 6% this quarter. The adjusted loss of AIDC narrowed significantly year-over-year, approaching breakeven, benefiting from logistics optimization and improved operational efficiency.
The unit economics of Global AliExpress Choice business continued to show substantial sequential improvement. Next, let us review the business updates and financial performance of our Cloud Intelligence Group. Our cloud business delivered another quarter of accelerated growth.
Revenue from external customers accelerated to 40% growth. AI-related products continued to drive this momentum. We achieved triple-digit AI revenue growth for the eleventh consecutive quarter.
The proportion of AI revenue in external cloud revenue continued to increase, now reaching 30%. AI revenue for this quarter was 9 billion yuan. The annualized run rate is 36 billion yuan, or approximately $5.3 billion. This clearly reflects the scale and acceleration of our AI business. The adjusted profit margin remained relatively stable at 9.1%.
Revenue from all other segments decreased by 21% to RMB 65.5 billion, primarily due to the disposal of Intime and Sun Art businesses, as well as a decline in Cainiao's revenue, which was partially offset by increased revenue from Freshippo and Ele.me. The EBITA loss from all other segments amounted to RMB 21.2 billion, mainly due to increased investments in technology businesses, including foundational models and the consumer-oriented Tongyi APP. As we conclude this fiscal year, we remain committed to delivering sustained shareholder returns.
Our board has approved an annual dividend of $1.05 per share. We will continue to decisively invest in AI and consumer businesses, where we see significant long-term growth potential and compounded competitive advantages. We believe these investments will yield growth and returns over time, ultimately creating greater value for our shareholders.
(Q&A Session)
Question:
Thank you for sharing, for the first time, a very impressive scale and target for AI model and application ARR. I would like to ask: how much of this ARR is driven by our self-developed models, such as Qwen, and how much comes from third-party models? Additionally, considering the recent rise in token prices, what impact will this have on the profit margins of Model-as-a-Service (MaaS) and cloud services?
Answer:
In this quarter, we just announced the latest figures for our model and application services revenue. This revenue is primarily composed of two components: the API services of our Model-as-a-Service platform 'Bailian,' and subscription-based revenue from our AI-native software. Currently, the majority of this revenue comes from Bailian’s MaaS API services. Meanwhile, Alibaba Cloud's Bailian platform is relatively open, hosting not only our proprietary models but also third-party open-source and closed-source models. In terms of current revenue scale, the majority still comes from our self-developed models, including our Qwen foundational model, speech models, and video models.
Your second question is also quite critical. Over the past quarter or recent months, the entire industry has undergone a major shift: AI is transitioning from dialogue-based chatbots to agent operations. Agents are required to help customers complete highly complex reasoning tasks. During this reasoning process, customer demand for model inference continues to grow exponentially. At the same time, because agents can assist customers in performing more complex tasks, the price increase of API tokens has been widely accepted, with strong and sustained demand. Moreover, at present, our supply cannot fully meet customer needs, and there are many queued customers. Therefore, under the current circumstances, we believe that the gross margin of MaaS business will be higher compared to SaaS business.
There are several other key points. First, the development of inference technology continues to progress. Each quarter, we see optimization results from advancements in inference technology, leading to continuous improvements in token production per server or per GPU. Additionally, model capabilities are continuously strengthening, and model pricing is expected to rise steadily over the next one to two years. From this perspective, I believe the rapid growth of the MaaS business will have a very positive impact on our gross margin in the coming quarters.
Question:
I have a question regarding the return on invested capital for our AI investments. While our AI investments have driven an impressive 40% cloud growth, they have also significantly weighed on the group's free cash flow and EBITDA. How should investors evaluate the returns on these investments? What is your management framework for balancing aggressive AI spending with profitability?
Answer (Xu Hong):
Let me answer the first part. You may have noticed, especially this quarter, that our free cash flow turned negative, so there is concern about how we manage our free cash flow. First, the primary reason for the negative free cash flow is our investment in AI over the past year. Because we recognize this historic opportunity, we have firmly committed to investing in it, which is the main factor leading to the overall negative free cash flow.
Looking two more years into the future, our investment will remain resolute. This window of opportunity is only a few years for us, so we will continue to invest firmly. Meanwhile, from the perspective of operating cash flow, there has been no significant change. Here, I would like to address two aspects:
First, from the consumer business perspective, Taobao and Tmall are our most important contributors to operating cash flow, which remains highly stable. In the next two years, as the flash purchase business significantly narrows its losses and AIDC transitions from loss to profitability, the overall consumer business will see a very positive development in operating cash flow. That is the first point.
Therefore, over the past year, we have been very determined in making these investments. Looking ahead to the next two years, we plan to continue with the same determination. Again, this is because we see it as a critical window of opportunity that will remain open for the next few years. Additionally, our view on cash flow has not changed significantly.
First, the main contributors to the group's operating cash flow are Taobao and Tmall, and that part of the cash flow is very stable. Looking ahead two years, for near-field e-commerce, losses will be significantly reduced. Meanwhile, AIDC will transition from losses to profitability. Therefore, we believe these developments will be very positive for net cash flow over the next two years.
The second important point is our investment in cloud infrastructure. As Eddie mentioned earlier, this will accelerate our cloud revenue, including AI cloud revenue, while improving gross margins. All of these will further enhance our operating cash flow returns in the cloud sector, which can also support our continued investment in cloud infrastructure.
Third, I previously mentioned that our balance sheet remains very strong. Our current net cash is approximately USD 38 billion. If we exclude debts maturing in more than five years, our net cash amounts to about USD 59 billion. Such a robust balance sheet can well support our investment in cloud infrastructure.
Finally, we maintain strong capital market financing capabilities, allowing us to raise funds through various market instruments to meet strategic development needs. Thus, I have addressed your earlier question regarding cash flow and how we plan to ensure sufficient cash reserves. Now, let’s see if Eddie has any additional comments.
Answer (Wu Yongming):
Regarding your question about AI-related investments and future ROI, I would like to take this opportunity to share our perspective. Based on the current development trends in the AI business, AI resembles a manufacturing industry. In other words, to generate higher revenue, we need to establish two core factories. The scale of these two factories will determine the size of our future revenue. One factory can be called the AI training factory, and the other, the AI inference factory. Behind these factories lies the construction of AI data centers, which will inevitably consume a significant portion of the group’s free cash flow. However, the return path for these rigid data center infrastructure investments is very clear.When we commercialize these data centers for B2B purposes, the path forward is very clear. Whether through cloud IaaS commercial services or by generating additional revenue via the Maas platform and AI-native software models, nearly all resources are utilized during the service period without idle capacity. Hence, given the demand outlook for the next three to five years, we believe the return on our substantial investment in AI data center construction is highly certain.
Question:
My question pertains to near-field e-commerce. In your prepared remarks, you mentioned improvements in UE. I would like to understand the key drivers behind metrics such as AOV, subsidy ratios, and fulfillment rates. Additionally, I recall discussing the outlook for near-field e-commerce over the next few years during our last conversation. Could you provide any updates or changes regarding market dynamics or UE? Thank you.
Thank you, management team. My question is about instant retail. In the prepared remarks, you highlighted significant improvements in UE. I would like to better understand the underlying drivers, including contributions from ticket size, subsidy ratios, and fulfillment, among other factors. Furthermore, during last quarter's earnings call, you outlined a two-year outlook for instant retail. Since then, have there been any updates to this outlook? Have your perspectives on market dynamics, UE, and subsidies evolved?
Answer:
First, after a year of investment, we have witnessed rapid growth in our instant retail business, with a fundamental shift in our market share. Comparing the March quarter of this year with the same period before our large-scale investment last year, both our order volume and market share have significantly increased. From January to March, the total order volume was 2.7 times that of the same period last year, with the non-food retail segment growing threefold. Since April, while maintaining order volume, we have continued to enhance UE through improved logistics efficiency, optimized order structure, and particularly, further increases in AOV. We are confident in achieving positive UE by the end of the new fiscal year.
While optimizing UE, we will continue to enhance the experience for users and merchants through innovation to maintain our long-term competitiveness in the instant retail sector. Therefore, we are confident that we can achieve overall profitability in instant retail in the future under the new scale and market share levels. This quarter, we have also observed the positive impact of flash purchases on e-commerce, particularly in acquiring new customers, boosting user activity, meeting diversified consumer scenarios, driving transactions and commercialization, as well as enhancing logistics infrastructure, thereby promoting the overall development of Taotian’s business segments. We continue to see significant growth momentum driven by flash purchase-related categories, especially in food and fresh produce, which further supports the development of instant retail-related businesses such as Freshippo and Cat's Eye Supermarket. Our physical e-commerce has achieved robust growth this quarter in both transaction volume and CMR, with flash purchases and instant retail playing a definitive and positive role.
Question:
I would like to ask about the Maas business mentioned by management in the opening statement. How does Alibaba compare with other leading AI platform companies and AI ventures in China in terms of competitive advantages in the Maas space? In the U.S., we have seen AI agents, especially AI coding, emerge as the fastest-growing area of AI commercialization. So, when do you expect to see similar growth trends in AI coding in China? Additionally, Chinese enterprises have traditionally been reluctant to pay for SaaS products. Do you think this might limit the commercial potential of AI coding products in China compared to their U.S. counterparts?
Answer:On Alibaba Cloud’s Bailian platform, we position ourselves as an open AI inference platform. Currently, the majority of our revenue comes from proprietary models. Compared to these AI ventures, our investments in model breadth far exceed theirs. Of course, these ventures exhibit strong technical expertise and quick commercial acumen when focusing on specific models or domains, and they are making rapid progress. Therefore, within the Maas space, these ventures can, to some extent, be viewed as partners of Alibaba Cloud.
However, Alibaba takes a broader approach, emphasizing extensive research across various model domains. For example, we develop foundational models prioritizing coding capabilities, such as the Qwen base model. Our video models (whether Wanxiang or Happy House), forward-looking world models, and voice models are designed to address future business needs. We believe that in many business scenarios, users will require the integration of multiple model capabilities to meet their demands. This is one key distinction between us and leading AI ventures. At the same time, these ventures can also be considered partners of Alibaba Cloud’s Bailian platform.
Your question about AI coding is very pertinent. When will we see similar growth trends? Based on our observations, China is already experiencing such growth. Trends we observe on the Bailian platform, as well as feedback from AI ventures with whom we maintain cooperative relationships, indicate that from November and December last year to May this year, API demand from numerous companies has surged, primarily driven by advancements in AI coding capabilities.
AI coding does not simply replace the work of software engineers. Due to enhancements in AI model capabilities and the integration of AI coding with the broader agent runtime environment (including harness engineering and data domains), we are seeing complex task agents powered by AI coding that can handle nearly all digital work tasks. Whether in the U.S. or China, this wave of AI-driven demand is largely fueled by advancements in AI coding. The combination of enhanced AI coding capabilities with computing or digital tool scenarios theoretically addresses almost all future complex tasks in digital work. This represents a critical growth trend over the next two to three years. Let me add...
Regarding the long-standing issue of Chinese enterprises being less willing to pay for SaaS products, I believe there is a significant difference in the era of large models: AI models are becoming increasingly powerful. When these models can truly help users complete complex tasks at work, we observe that the willingness to pay for such intelligent capabilities is equally strong in both the U.S. and China. Theoretically, as long as the value created by these solutions within an enterprise exceeds the token cost, the demand for API tokens will be virtually unlimited. Therefore, we believe the underlying growth in AI demand remains highly long-term and certain.
Based on the data we have observed, we can share that the overall growth rate on our Bailian platform has been extremely rapid. Compared with the data from November and December of last year, the figures for May and June this year are expected to increase more than tenfold. Moreover, our recent ARR has surpassed 8 billion. This means that it is highly certain that we will exceed 10 billion in ARR this quarter. Therefore, we see that whether in Chinese or American enterprises, the willingness to pay for intelligent capabilities and to utilize these capabilities to assist in accomplishing real work tasks has become a universally accepted choice.
Regarding your other comment, we have also noticed the relatively low willingness to pay for SaaS in China. However, I believe that as models become increasingly powerful and capable of solving highly complex tasks and providing truly valuable intelligence, this situation will change. We can expect that demand for such services in China will mirror that of the United States. In a sense, when the value provided by tokens exceeds their cost, the demand for tokens will, to some extent, become limitless. Thus, we consider the growth in AI demand to be a long-term certainty.
I would also like to share some data. From November and December of last year to May this year, the ARR growth we have seen on the Bailian platform has exceeded tenfold. The ARR has now surpassed 8 billion. In my view, it is highly certain that we will achieve an ARR exceeding 10 billion this quarter.
Question:
I would like to continue discussing the topic of global comparisons. Looking globally, overseas counterparts seem to have captured the most direct growth in enterprise agent workflows, while consumer-facing applications and monetization appear to lag slightly. Looking ahead, considering that Alibaba is investing in cutting-edge innovations such as infrastructure, models, cloud services, and the Tongyi app, how do we evaluate the strategic priorities and resource allocation between our B2B and B2C initiatives? If enterprises continue to gain more traction in the future, will we consider gradually shifting more resources from the Tongyi app to cloud services and MaaS?
Answer:
Thank you very much for your question. You have raised an excellent point. However, from the fundamental principles of AI itself, it represents more of a revolutionary shift in computational paradigms. Ultimately, this paradigm shift must help users complete tasks or solve problems more effectively. From this perspective, I believe that B2B and B2C are fundamentally the same. For now, however, we observe that globally and within China, the strongest customer willingness to pay resides in the B2B domain, as ROI calculations are more straightforward and enterprises exhibit stronger payment intent. Therefore, the majority of our inference resources are allocated to commercializing B2B applications.
On another level, AI ultimately serves as an assistant to humans. As a human assistant, it can function as a work assistant, personal assistant, or learning assistant. Fundamentally, the problem to be solved remains the same: using AI to help people accomplish tasks, whether they are B2B, B2C, or related to learning, work, or daily life.
We believe that customer acceptance and willingness to pay for B2C services may still require a certain investment cycle. However, we are confident that B2C AI assistants will gradually form viable business models as technology advances and customers become more accepting, or as they assist users in completing more tasks. Such business models have already emerged overseas, and domestically, we anticipate significant commercial progress in B2C AI assistants within the next one to two years.
Question:
I would like to follow up on the previously mentioned question regarding the future development of our cloud business. Beyond acceleration, could management elaborate further on the EBITA margin, particularly in the coming quarters? Will we observe a similar trend of margin expansion as seen internationally as we accelerate? Thank you.
Answer:
In terms of current AI technology and its penetration across industries, we believe it is still in its early stages. However, regarding Alibaba Cloud and the objectives of our AI business, the primary focus remains on growth, along with achieving absolute leadership in user numbers, token consumption, and market share. We aim to outpace the industry's average growth rate, rapidly capture market share, and solidify our position as the undisputed market leader. Profitability remains our secondary objective.
However, due to several distinct characteristics of the industry at present, we assess that it will remain challenging to meet the demand for AI over the next three to five years due to various physical bottlenecks. Whether considering the construction cycles of AI data centers or the production timelines for chips, memory, and other components, as well as physical capacity expansion and overall output growth, we believe it will be difficult to support the growth in AI demand within the next three to five years.
From this perspective, historically, Alibaba Cloud has benefited from a strong customer base and economies of scale, as well as significant IT capex scale effects. Under these scale effects, due to tight market supply and demand, the cost of deploying a new server—identical to those we deployed two years ago—has more than doubled, meaning costs have increased by over 100%. This new server replacement cost has a pricing pull effect on both existing and new customer demands. We believe this will positively enhance the asset pricing of cloud services over a longer future period.
Secondly, we are also witnessing rapid growth in Alibaba Cloud's Maas business. The Maas business itself generates higher gross margins compared to our traditional IT asset-based businesses. Additionally, with ongoing optimization of inference technologies, the output per GPU card will continue to rise. Therefore, we currently observe that the revenue and gross margin generated by deploying the same server on the Bailian platform exceed those of our traditional cloud computing simple computing power services. Thus, we believe that an increase in the revenue share of this business will also enhance future gross margin levels.
Another factor is our full-stack technology advantage, including Pingtouge’s self-developed AI chips and their potential large-scale deployment, which will help us provide what may be one of the best cost-performance inference platforms in China. We believe this inference platform will also generate better gross margins and exhibit strong synergy with our models. Due to these objective factors, we anticipate a significant improvement in Alibaba Cloud's gross margin over the next 1-2 years, with visible changes expected within the next few quarters.
Question:
Regarding capital expenditure, what level of capex do we need to maintain to meet the requirements of Maas and long-term cloud business revenue? Additionally, management mentioned opportunities. What is the current deployment penetration rate of Pingtouge chips within Alibaba Cloud? As the penetration rate increases, how much of an improvement in profit margins should we expect from the use of self-developed chips? Thank you.
Answer:
Your first question is also quite important. In fact, we mentioned in last quarter's presentation that we have set a very ambitious revenue target for the next five years. Given this target, comparing 2022 and 2023—before the large AI model boom—the external revenue target for Alibaba Cloud represents roughly a tenfold increase. A rough calculation suggests that we would require at least ten times the data center assets held by Alibaba Cloud during that period to support our long-term business objectives. To some extent, the scale of data centers we aim to build in the future will reflect a more than tenfold increase compared to 2022.
This is the total goal for all our capex investments, including the substantial amount of computing power we acquire through OPEX methods. Compared to the previously mentioned three-year plan totaling 380 billion yuan, our five-year target will require significantly higher investment to obtain these computing centers, far exceeding the original 380 billion yuan. However, the current situation is more complex, and not all computing centers will necessarily be acquired through self-built capex. For instance, we may adopt OPEX leasing models. Moreover, as Pingtouge chip production expands, we might sell Pingtouge AI servers to various computing centers or data center service providers while collaborating with them on construction. Thus, multiple approaches will support the expansion of our data center capacity. Fundamentally, this reflects data demand. Regarding data center needs, we should benchmark against a tenfold increase in scale compared to pre-AI boom levels in 2022.
Concerning the deployment ratio of self-developed chips within Alibaba Cloud, it currently remains relatively low. However, I would like to share that Pingtouge’s chips include not only our own GPUs but also fully self-developed CPUs, storage, and network chips. Thus, in the future, we have the opportunity to create a fully self-developed stack encompassing GPUs, CPUs, storage, and network chips.
The current proportion is still low due to several objective reasons. For instance, overall semiconductor production capacity in China remains limited. Nevertheless, domestic semiconductor production capacity in China has been expanding steadily in recent years. Therefore, we believe that as the penetration rate of Pingtouge’s fully self-developed chipsets increases, it will have a significant impact on improving our gross margins. However, there are still some interrelated factors. For example, China’s domestic semiconductor manufacturing processes are generally less advanced than those abroad, resulting in certain gaps in performance, energy consumption, or efficiency when compared to leading foreign chips.
However, we observe that mainstream foreign AI chips command extremely high gross margins, typically ranging from 60% to 80%. Even if domestic chips achieve improvements in overall performance or power consumption, there remains a substantial room for cost-performance enhancement compared to the 60% to 80% gross margins of advanced foreign chips. The ultimate impact of this cost-performance improvement on our overall gross margin will depend on the pace of our capacity expansion and the extent to which new production replaces existing inventory after overall capacity expansion.
Editor/Jayden