Alibaba Group announced its financial results for the fourth quarter and the full fiscal year 2026 today. The company reported Q4 revenue of 2433.8 billion yuan, a 3% year-over-year increase. Excluding the revenue from disposed businesses such as Sun Art Retail and Intime, the comparable revenue growth would have been 11%. Net profit for the quarter stood at 254.76 billion yuan. For the full fiscal year 2026, revenue reached 10236.7 billion yuan, up 3% year-over-year, with an annual net profit of 1059.04 billion yuan.
Following the earnings release, Alibaba Group's Executive Chairman Joe Tsai, CEO Eddie Wu, CFO Toby Xu, and CEO of Taobao & Tmall Group, Fan Jiang, hosted a conference call to discuss the results and answer questions from analysts.
The following are key highlights from the analyst Q&A session:
**Goldman Sachs Analyst Ronald Keung:** Thank you for sharing the scale and progress in AI MaaS and applications. Could you break down the growth in Annual Recurring Revenue (ARR)? What proportion comes from self-developed models like Qwen versus third-party models? Also, given the recent increase in token prices, what impact does this have on the profit margins of your MaaS and cloud businesses?
**Eddie Wu:** Thank you. This quarter marks the first time we have disclosed revenue from our model and application services, which primarily consists of two components: API service revenue from the Bailian MaaS platform and subscription revenue from AI-native software. Currently, the vast majority of this revenue comes from Bailian MaaS API services. Alibaba Cloud Bailian is an open platform hosting both our self-developed models and third-party open-source or proprietary models. However, in terms of current revenue scale, the majority is generated by our self-developed models, including the Qwen foundational model, voice models, video models, and others.
Regarding your second question, a significant shift has occurred in the industry over recent quarters: AI is evolving from conversational chatbots to intelligent agents (Agents). These agents must help customers complete complex reasoning tasks, leading to a sustained surge in demand for model inference. Precisely because agents can solve more complex problems, customer acceptance remains high despite our token price adjustments. Demand continues to be robust; in fact, current supply cannot fully meet demand, with a significant queue of customers waiting for service.
Fundamentally, the gross margin of the MaaS business is inherently higher than that of the IaaS business. Several other key factors are at play: First, inference technology is continuously advancing, with token output per server and per GPU card improving each quarter. Second, model capabilities are strengthening, and we expect model pricing to maintain an upward trend over the next 1-2 years. Consequently, the rapid growth of the MaaS business in the coming quarters will have a very positive impact on our overall gross margin.
**UBS Analyst Kenneth Fong:** Congratulations on the excellent performance in AI. Could you discuss the return on investment (ROI) for AI? While AI investment has driven 40% growth in the cloud business, it has also weighed on the group's free cash flow and earnings per share (EPS). How should investors evaluate the ROI of this investment? What framework does management use to balance aggressive AI investment with earnings stability?
**Toby Xu:** Let me address the first part. You may have noticed negative free cash flow this quarter, primarily due to our substantial investments in AI over the past year. We recognize this as a historic opportunity and are committed to advancing our investments, which is the main reason for the negative free cash flow.
Looking ahead over the next two years, we will continue our firm commitment to investment, as this window of opportunity lasts only a few years. From a cash flow perspective, the core operational logic remains largely unchanged. First, Taobao and Tmall, as the primary contributors to the group's operating cash flow, continue to provide stable cash flow. Over the next two years, as losses from the on-demand retail business narrow significantly and the AIGC business transitions from loss to profitability, the consumer business's operating cash flow is expected to show strong positive growth.
Second, as Eddie mentioned, investments in cloud infrastructure will drive accelerated revenue growth for the AI+Cloud business, coupled with continuous gross margin improvement. This will further enhance the net cash flow from the cloud segment, which in turn can fund ongoing cloud infrastructure investments.
Third, the group's balance sheet remains solid. As of March 31, net cash (excluding debt maturing beyond five years) was approximately $38 billion. Excluding all long-term debt, it was about $59 billion, providing strong support for cloud infrastructure investment.
Finally, we possess strong capital market financing capabilities and can utilize various financing methods to meet strategic development needs. That addresses the cash flow and capital reserve question. Eddie will now add to this.
**Eddie Wu:** Regarding the future ROI of AI investments, an analogy can be made: AI is more akin to manufacturing. To generate more revenue, we must build two core "factories" – the AI training factory and the AI inference factory. The scale of these factories determines our future revenue scale.
The core of these factories is AI data center construction, which indeed consumes a significant amount of the group's internal cash flow. However, the return path for these essential infrastructure investments is very clear. In terms of ToB commercialization, whether it's revenue from cloud IaaS services, the MaaS platform, or the model layer brought by AI-native software, our servers currently have almost no idle resources. Therefore, the return on investment for AI data center construction over the next 3-5 years is highly certain.
**Jefferies Analyst Thomas Chong:** Thank you. A question on the on-demand retail business: You mentioned significant improvement in Unit Economics (UE) in your prepared remarks. Could you elaborate on the drivers, such as Average Order Value (AOV), subsidy ratios, and fulfillment efficiency? Also, you provided a two-year outlook last quarter. Has your view on the market landscape or UE for the next three years been updated?
**Fan Jiang:** Thank you. First, after a year of investment, the on-demand retail business has achieved rapid development, with a fundamental shift in market share. Compared to the March quarter last year (before large-scale investment), this year's order volume and market share have significantly increased. The overall order volume from January to March was 2.7 times that of the same period last year, with non-catering retail orders tripling.
Since April, while maintaining order volume, we have driven significant UE improvement by enhancing logistics fulfillment efficiency and optimizing order structure, particularly with a further rise in AOV. We are confident in achieving breakeven UE by the end of fiscal year 2027.
While continuously optimizing UE, we will continue to enhance user and merchant experience through innovation to maintain long-term competitiveness in the on-demand retail space. We are confident in achieving overall profitability for the on-demand retail business at this new scale and market share level.
This quarter, the on-demand retail business significantly boosted the physical goods e-commerce segment, particularly in new customer acquisition, increasing user engagement, fulfilling diverse consumption scenarios, driving transaction volume and commercialization, and improving logistics infrastructure, thereby promoting the overall development of the Taobao and Tmall Group. It notably drove growth in categories like food and fresh produce and also accelerated the growth of related businesses like Freshippo and Tmall Supermarket. This quarter, both Gross Merchandise Volume (GMV) and Customer Management Revenue (CMR) for physical goods e-commerce showed positive growth momentum, with on-demand retail playing a clear and positive role.
**Nomura Analyst Jialong Shi:** Thank you. A follow-up on the MaaS business: Compared to other leading AI platform companies and AI startups in China, what is Alibaba's core advantage in the MaaS field? In the U.S., AI Agents (especially AI coding) are the fastest-growing area for AI commercialization. When do you expect similar growth rates for AI coding in China? Additionally, Chinese enterprises seem less willing to pay for SaaS products compared to U.S. firms. Could this lead to a less promising commercialization outlook for AI coding products in China versus the U.S.?
**Eddie Wu:** Thank you. Alibaba Cloud Bailian is positioned as an open AI inference platform, and currently, the platform's revenue is primarily contributed by our self-developed models. Compared to AI startups, our scale and breadth of investment in models far exceed those of startups focused on a single model or domain—though these startups possess strong technical expertise and business acumen in their specific areas and are progressing rapidly. Strictly speaking, in the MaaS field, these startups are more like partners of Alibaba Cloud.
Alibaba's advantage lies in our emphasis on broad R&D across multiple model domains: including the code-centric Qwen foundational model, video models (Wanxiang, HappyHouse), future-oriented world models, voice models, etc. We believe that in many future business scenarios, users will require a combination of multiple model capabilities to meet their needs. This is a core differentiator from AI startups, while we also maintain cooperative relationships with them.
Regarding the growth rate of AI coding in China, we judge that a growth rate similar to the U.S. has already arrived. Based on data from the Bailian platform and our partner-friendly AI startups, from November-December last year to May this year, the surge in API demand from numerous enterprises has almost entirely stemmed from improvements in AI coding capabilities. AI coding is not simply replacing software engineers. As model capabilities improve and AI coding integrates with agent runtime environments (like data domains such as Harness engineering), AI coding-driven agents for complex tasks can now accomplish tasks in various digital work scenarios.
Whether in the U.S. or China, the core driver of this wave of AI demand growth is the enhancement of AI coding capabilities. Combined with computer or digital tool scenarios, AI coding can theoretically solve almost all complex tasks in digital work, which will be a significant growth trend over the next 2-3 years.
Regarding the willingness of Chinese enterprises to pay for SaaS, we believe this will change in the era of large models. As model capabilities become increasingly powerful, able to genuinely solve complex enterprise tasks and provide core value, Chinese enterprises, like their U.S. counterparts, are willing to pay for such intelligent capabilities. Theoretically, as long as the value brought by tokens exceeds their cost, enterprise demand for tokens is unlimited, making the growth of AI demand a long-term certainty.
To share a data point: The overall growth rate of the Bailian platform is extremely fast. Data from May-June this year shows growth exceeding 10 times compared to November-December last year. Moreover, very recently, the ARR (Annualized Recurring Revenue) for our AI Model and Application Services has already surpassed 8 billion yuan. Breaking the 10 billion yuan ARR mark this quarter is a very certain prospect. Therefore, we observe that both Chinese and U.S. enterprises are willing to pay for intelligent capabilities that solve real work tasks—this is a universal trend.
**Macquarie Analyst Ellie Jiang:** Thank you. A question with a global comparison: Overseas peers seem to have captured near-term opportunities in enterprise agent workflows faster, while consumer-side monetization lags. Given Alibaba's investments across infrastructure, models, cloud, the Qwen app, and other areas, how do you assess the strategic priority and resource allocation between ToB and ToC businesses? If enterprise-side development continues to improve, would you consider shifting more resources from ToC to areas like MaaS?
**Eddie Wu:** Thank you. From the essence of AI, it is a computational paradigm revolution, with the core purpose of helping users complete tasks and solve problems. From this perspective, the essence of ToB and ToC is consistent. However, currently, both globally and in China, the area with stronger customer willingness to pay is indeed ToB, because the Return on Investment (ROI) for enterprises is easier to quantify. Therefore, the majority of our inference resources are allocated to ToB commercialization.
In the long term, however, AI will ultimately become a human assistant, encompassing both work assistants and personal/learning assistants. The essence is using AI to help people solve tasks, with scenarios spanning ToB, ToC, work, and life. We believe the customer acceptance and willingness to pay in the ToC business require a certain investment cycle. But as technology advances and AI assistants can help users accomplish more tasks, a mature business model will gradually form on the ToC side. This model is already emerging overseas, and we expect significant progress in the commercialization of ToC AI assistants in China within the next 1-2 years.
**Bank of America Merrill Lynch Analyst Joyce Ju:** Thank you. A follow-up on the future development of the cloud business: Beyond accelerating growth, could you discuss the trend in EBITDA gross margin? In the coming quarters, as business accelerates, do you anticipate a gross margin expansion trend similar to that seen with international peers?
**Eddie Wu:** Thank you. Currently, the penetration of AI technology across industries is still in its early stages. However, the core objective for Alibaba Cloud and the AI business is to achieve growth in revenue, user token consumption, and market share at a pace exceeding the industry average, solidifying an absolute market leadership position. Profit margin is a secondary goal at this stage.
Several industry characteristics, however, will drive gross margin improvement. First, over the next 3-5 years, the growth of AI demand will face physical bottlenecks. The construction cycle for AI data centers, the production cycle for chips and memory, and the pace of capacity expansion are all challenging to fully align with the growth of AI demand.
Leveraging its massive customer scale effect and the scale advantage formed from historical IT capital expenditure (CAPEX), Alibaba Cloud, in the current tight supply-demand environment, faces a cost for deploying a new server this year that is over 100% higher than two years ago. This rise in server replacement costs exerts an upward pull on pricing for both new and existing customers, which is expected to positively impact the asset pricing of cloud services in the long term.
Second, the rapid growth of Alibaba Cloud's MaaS business contributes significantly. The gross margin of the MaaS business is notably higher than that of traditional IaaS or other IT asset-based businesses. Simultaneously, as inference technology is optimized, output per GPU card will continue to increase. The same server generates higher revenue and gross profit when utilized by the Bailian platform compared to providing simple computing power services in traditional cloud computing. Therefore, the increasing proportion of MaaS business revenue will continuously drive overall gross margin improvement.
Third, the synergistic effect of full-stack technological advantages: The large-scale mass production of T-Head's self-developed AI chips will help us build China's most cost-effective inference platform, forming a strong synergy with our models and further enhancing gross margin.
Based on these objective factors, we judge that Alibaba Cloud's gross margin will achieve significant improvement over the next 1-2 years, and this change will gradually become apparent over the most recent 1-2 quarters.
**Morgan Stanley Analyst Gary Yu:** Thank you. Two questions: First, to achieve the long-term revenue targets for MaaS and the cloud business, what level of capital expenditure (CAPEX) needs to be maintained? Second, what is the current penetration rate of T-Head chips within the cloud business? As penetration increases, what magnitude of gross margin improvement is expected?
**Eddie Wu:** Thank you. Regarding the first question, in last quarter's earnings call, we outlined a high revenue target for the next five years—representing a 10x growth compared to Alibaba Cloud's external revenue in 2022 (before the large model boom). Roughly calculated, supporting this long-term business target requires data center assets at least 10 times those of 2022. Therefore, the scale of data centers to be built in the future will be more than 10 times that of 2022. This target will be achieved not only through CAPEX but also through operational expenditure (OPEX) methods like leasing computing power.
Compared to the previously announced three-year CAPEX plan of 380 billion yuan, the related investment required to achieve the five-year target will far exceed 380 billion yuan. However, the current situation is more complex. Not all computing centers need to be self-built: we can acquire computing power through leasing; as T-Head chip production capacity expands, we can sell AI servers equipped with T-Head chips to computing center or data center service providers, or jointly build data centers with them. We will expand data center capacity through multiple methods, but the core requirement is to target a scale 10 times that of 2022.
Regarding the second question, the deployment proportion of T-Head's self-developed chips within Alibaba Cloud is currently still relatively low. Besides TPUs, our CPUs and storage network chips are all fully self-developed across the stack, with the potential for future large-scale deployment of a fully self-developed chip stack. The current low penetration is mainly constrained by domestic semiconductor production capacity in China, which has been continuously expanding in recent years. As the penetration rate of T-Head's fully self-developed chips increases, it will have a significantly positive impact on gross margin.
It should be noted that several interrelated factors exist: Domestic semiconductor process technology still lags behind advanced international levels, resulting in chips that are slightly inferior in power consumption and efficiency compared to top international chips. However, the gross margin of mainstream international AI chips is as high as 60%-80%. Therefore, even with room for improvement in the performance and power consumption of domestic chips, there exists a huge space for cost-performance optimization. Ultimately, the extent of gross margin improvement from this cost-performance advantage will depend on the future pace of production capacity expansion and the replacement ratio of existing chip inventory.
