Alibaba Group (BABA) announced their Q3 FY2026 earnings yesterday (19 March). This quarter’s performance missed estimates and share price closed 7.16% lower than the previous trading day.
Earlier this week (16 March), Alibaba just announced the establishment of the Alibaba Token Hub (ATH) business group to be led by Eddie Wu. The following day (17 March), it then officially launched “Wukong”, a to-B AI agent platform.
We will share our views shortly, but first, here is the transcript we translated from the executives’ Chinese responses in the Q&A part of the earnings call. The text here might differ from the translation provided by officially appointed 3rd party interpreters during the earnings call itself. You can find the full earnings report on Alibaba Group’s investor relations web site.
Q&A Session
Q: Robin Zhu – AllianceBernstein:
Could you walk us through, now that ATH business group has been established, what new changes we should expect in terms of collaboration across different cloud and AI-related businesses from a design perspective? From a strategic standpoint, what kind of changes does this new structure bring, or what new goals does it help achieve compared with the previous setup?
In addition, could management elaborate on your priorities in AI? For example, is your top priority gaining market share and revenue growth, as mentioned earlier? Or is it achieving the strongest model capabilities? Or expanding consumer-facing use cases? Or developing AI agents? Could you rank these priorities for us? Thank you.
A: Eddie Wu, Group CEO:
Thank you for the question. The establishment of ATH is closely tied to the current era and the broader context of technological transformation. From the second half of 2025 through the first few months of 2026, we have already seen that AI has entered the era of AI agent-driven development.
The key difference between this agent-driven era and earlier stages of AI lies in the much tighter integration between models, applications, and underlying hardware. This close coordination is critical to both improving model performance and advancing applications. In the early days of AI, much of the pretraining data came from relatively static datasets. As we move into the AI agent era, however, improvements in model capability—and the ability to build better applications—are increasingly driven by the tight coupling between models and applications, as well as real-time data feedback from actual user scenarios. Looking at the AI stack in five layers—applications at the top, followed by models, then AI infrastructure such as cloud computing, and further down to the chip layer—we believe that in the agent era, the close integration between models and applications becomes especially critical.
Let me also explain the relationships among the different businesses within ATH. Based on where we see the industry heading, it is increasingly clear that in the future, the application layer for AI agents will develop into a highly vibrant ecosystem with extensive innovation and a wide range of use cases. At the application layer, our own strategy includes the Qwen app as our consumer-facing (to-C) agent, and we aim to build “Wukong” into an agent for enterprise users (to-B) in China. At the same time, we expect a large number of industry-specific and vertical AI applications, as well as various AI agent solutions tailored to different scenarios, to emerge and support the development of different industries. Therefore, beyond the application layer, we also need a strong MaaS (Model-as-a-Service) business. MaaS serves as the bridge between the model layer and the application layer. In addition to supporting our own internal applications, a robust MaaS offering will enable us to support a wide range of external AI applications across industries. We believe the market value and addressable opportunity here will be very substantial.
From this perspective, the AI application layer will become the largest distribution channel for AI tokens. The strongest model capabilities will attract more applications to use our models, while a more efficient and capable MaaS product will better connect models and applications. This is the core business logic behind the design of this business group. If we focus specifically on the model and application layers, our top priority is clearly to build the most capable models. Only the strongest models can drive the expansion of use cases across industries, and only the strongest models can attract developers and enterprises to adopt our MaaS services.
However, I want to emphasize that building the strongest models requires deep integration with various industries. This includes our own to-C and to-B applications, as well as connecting with a broad range of industry applications through MaaS, so that more users can interact with our models and help form a data flywheel and business loop. Through this approach—leveraging more scenarios, more data, and more users—we can continuously iterate and improve model capabilities, ultimately creating a self-reinforcing data flywheel. This is also a key reason why we established the ATH business group. To summarize, our top priority is to enhance model capabilities. But achieving this requires coordinated efforts across the model layer, application layer, and MaaS within ATH, in order to drive sustained, long-term improvements in model performance.
Q: Joyce Ju – Merrill Lynch:
Good evening, management, and congratulations on the solid progress you’ve made in cloud and AI. My question is about CMR (Customer Management Revenue). We’ve seen a slowdown in growth, particularly in the December quarter, given the macro pressures on China’s overall online retail sector. We observed overall online retail grew only 2% in 2025, but more recently, there’s an acceleration in January and February. Could you share your outlook for CMR trends as we move into the March quarter? Also, are you starting to see any improvement on the consumer side? Thank you.
A: Jiang Fan, Ecommerce CEO:
Let me take this question. As you mentioned, in the December quarter, factors such as weak macro consumption, a warmer winter, and the later timing of the Chinese New Year created significant headwinds for our growth. At the same time, due to the extended promotional season, we increased our investment in consumer incentives compared to previous years. As a result, both our CMR growth and e-commerce profitability slowed in the December quarter.
Since the beginning of the first quarter, however, we have observed a clear recovery in consumption. At the same time, supported by our quick commerce strategy, both GMV in our physical goods e-commerce business and CMR growth have rebounded significantly, and our e-commerce profitability has also improved meaningfully.
Q: Gary Yu – Morgan Stanley:
Thank you for taking my question. My question is about quick commerce. Over the past few months, Alibaba has achieved some phased progress in terms of market share gains and improvements in UE. Looking ahead, what will be your priority? Will it be to further expand market share, or to take this opportunity to further optimize UE and narrow losses?
Also, how should we think about the synergy between quick commerce and traditional e-commerce, and how will these synergies translate into faster CMR growth over time? Thank you.
A: Jiang Fan:
Let me take this question. Yes, we continue to see that as we gain market share, improvements in logistics efficiency, enhanced monetization capabilities, and optimization of our order mix are all driving continued improvements in unit economics. We believe our UE will continue to improve going forward.
Over the past year, we have also seen that quick commerce, Taobao Flash Shopping, has significantly contributed to overall platform growth. The broader e-commerce platform, including Flash Shopping, added 150 million annual active buyers, while our physical goods e-commerce business added 100 million annual active buyers. Annual active buyers on Taobao physical goods e-commerce increased by more than the total growth over the past three years combined. New users, compared to more mature users, tend to have lower average order value and purchase frequency in the short term. However, we aim to continuously increase their ARPU and purchase frequency over time, which will serve as a new growth engine for the platform in the coming years. We are also seeing strong momentum from Flash Shopping in driving growth in related categories, particularly in food, fresh produce, and health-related segments. It has also accelerated the development of quick commerce businesses such as Freshippo and Tmall Supermarket.
In terms of future outlook, we maintain our target of exceeding RMB 1 trillion in total GMV for quick commerce by fiscal year 2028. At that scale, we believe we can achieve positive cash flow at scale. We also expect the quick commerce segment to achieve overall profitability by fiscal year 2029. Flash Shopping and quick commerce have evolved into foundational businesses for Taobao and Tmall. They play a key role in acquiring new users, increasing user engagement, addressing diverse consumption scenarios, driving transactions and monetization, and enhancing logistics and supply chain capabilities. Strategically, they are critical to the long-term development of Taobao and Tmall in the AI era. Over the next two years, we will remain committed to investing toward the goal of surpassing RMB 1 trillion in GMV, while also strengthening our leading market position.
Q: Alicia Yap – Citigroup:
Thank you for taking my question. I have a few questions regarding your chip business, specifically T-Head. There have been recent reports suggesting that Alibaba may spin off T-Head for an IPO—could management share any updates on this? Also, could you provide some performance metrics for T-Head? For example, beyond the 470,000 chips you mentioned have been delivered, what is the corresponding revenue? And what kind of growth rate do you expect over the next year? You also mentioned that around 60% of demand comes from external customers. Could you elaborate on these use cases? Are chips provided to external customers mainly used for inference, while internal usage is more focused on model training, or other purposes? Finally, how do T-Head’s chips compare with other domestic chip offerings?
A: Eddie Wu:
Thank you for the question. T-Head is a critical component of Alibaba’s overall AI strategy, and I’m happy to take this opportunity to elaborate. Within China’s current AI chip ecosystem, we believe T-Head ranks among the top tier in terms of both technological and product capabilities. Our product portfolio covers the full AI workflow—from training and fine-tuning to inference. T-Head chips are already being deployed at scale within Alibaba Cloud, both in training scenarios and in inference scenarios such as our Bailian platform.
Across Alibaba Cloud’s public and hybrid cloud offerings, more than 60% of T-Head chips are now being used by external commercial customers, spanning industries such as internet services, autonomous driving, and intelligent manufacturing. These external customers use T-Head chips for both training and inference workloads. In addition, our software stack is well adapted to the CUDA ecosystem, which significantly reduces the time and effort required for customers to migrate their systems. Another key point is T-Head’s strategic importance to Alibaba. While domestic chips still lag behind international counterparts in manufacturing process and performance, we focus on deep co-design between our chips, Alibaba Cloud infrastructure, and Tongyi Qianwen models. This enables us to deliver superior cost-performance compared to standalone chip providers. Our goal is to maximize the cost-efficiency of AI capabilities and to make T-Head a key enabler in reducing inference costs on the Bailian platform.
Beyond cost efficiency, T-Head also plays a crucial role in ensuring computing supply. Over the next three to five years, global AI compute capacity is expected to remain in short supply, with even tighter constraints in China. As the only cloud provider in China with in-house chip design capabilities, T-Head is strategically vital for Alibaba in securing compute resources. This will support stronger growth across our cloud and AI businesses, including MaaS.
In terms of performance, over the past two years, T-Head has successfully commercialized its products, with cumulative shipments exceeding 470,000 chips and annualized revenue reaching the RMB 10 billion level. Looking ahead to 2026 and 2027, we expect continued expansion in the production of high-quality AI chips. This will provide sufficient compute capacity for our AI initiatives and serve as a strong growth driver for the group. It will also contribute positively to profitability over time. In summary, the value of T-Head lies not only in cost optimization, but more importantly in supply assurance—something that is critical in an era of constrained compute resources. As for a potential IPO, we do not rule out that possibility, but there is currently no specific timetable.
Q: CITIC:
Good evening, management, and thank you for taking my question. My question is about the commercial targets of Alibaba Group’s AI strategy. You mentioned a goal of exceeding USD 100 billion in revenue over the next five years. Could management provide more details behind this target—for example, the implied CAGR through 2031 and the key growth drivers?
Also, how should we think about the timeline for margin improvement at Alibaba Cloud as revenue scales?
A: Eddie Wu:
Thank you for your question. We believe that achieving over USD 100 billion in AI- and cloud-related revenue over the next five years is highly visible, given the size of the market opportunity, as well as our current foundation and product capabilities.
The most important growth driver will come from breakthroughs in large model capabilities. In the first few months of 2026, we have already seen clear signs that large models are becoming capable of handling complex enterprise (to-B) workflows. As more companies begin deploying AI agents powered by large models to perform real work tasks, we are seeing a fundamental shift in how enterprises view spending. Instead of treating token consumption as part of IT budgets, companies are increasingly treating it as part of production or R&D costs—as a core factor of production. This is a structural, long-term driver of AI demand.
We see three primary growth drivers going forward:
First, MaaS, driven by large models, will be the core engine of growth. Demand will come from multiple sources, including our own applications, our customers’ applications, and a wide range of industry-specific AI applications and software. We believe MaaS will be a key contributor to future AI and cloud revenue growth.
Second, in addition to public MaaS services, there will be a significant enterprise market for private deployment of inference and training. Many mid- to large-sized enterprises will continue to require private or hybrid setups, depending on their business models, security requirements, and specific use cases. This represents another major incremental opportunity for Alibaba Cloud’s AI infrastructure.
Third, a growth driver that is often overlooked is the expansion of traditional cloud computing in the AI agent era. Traditional cloud services, largely CPU-based, were originally designed for IT engineers—perhaps a few million to at most ten million potential users in China. However, in the age of AI agents, the number of “users” could scale to billions of agents. These agents will require extensive traditional cloud resources—CPUs, databases, storage, and memory—to support their long-term operation. Transforming traditional cloud computing from a product designed for human engineers into one optimized for AI agents represents a massive opportunity. This transformation is also a key focus of Alibaba Cloud’s ongoing upgrade efforts.
As revenue continues to scale and our AI business model evolves from primarily selling compute resources to delivering higher-value intelligent capabilities, we expect a meaningful upgrade in our business model. Combined with cost efficiencies from our self-developed T-Head chips, we believe cloud profitability will improve over time. However, margin expansion is unlikely to be linear. It may come in step changes driven by scale effects or breakthroughs such as increased adoption of our proprietary chips. The pace will depend on how quickly we achieve economies of scale across our products and infrastructure.
As for the implied CAGR from 2026 to 2031, it is straightforward to calculate mathematically. That said, both investment cycles and revenue growth in this space will not be linear. Investments made today may take one to two years to translate into meaningful revenue growth. Overall, we remain confident in achieving our five-year target.
Q: Alex Yao – JP Morgan:
Thank you for the opportunity. Let me shift the topic to e-commerce. Previously, we mentioned that e-commerce would enter a three-year investment cycle. Was this driven by the opportunity we see in quick commerce, along with related strategic adjustments? If no adjustments had been made, would we already be roughly in the middle of this three-year cycle? And going forward, should we expect a transition toward financial improvement and eventually a relatively stable harvest phase by the end of the three years? More broadly, could you share how you think about the current stage and positioning of this business cycle? Thank you.
A: Jiang Fan:
As I mentioned earlier, we are making significant investments this year in quick commerce, which we see as a major opportunity. We will continue to invest over the next two years to achieve our goal of surpassing RMB 1 trillion in quick commerce GMV. We also believe that, after this investment phase, quick commerce will begin to generate positive economic returns for our overall e-commerce segment.
On AI, as we’ve discussed, its impact on e-commerce will be substantial. That said, in the context of AI, a three-year timeframe is actually quite long—AI is evolving on a weekly and monthly basis. Therefore, we are also actively increasing our investments in AI. Whether on the user side or the merchant side, we will continue to roll out new experiences and upgrade merchant operating models powered by AI this year. AI is expected to drive significant upgrades across many areas of e-commerce. In addition, leveraging our B2B capabilities, we see substantial new opportunities emerging in e-commerce, and we will make every effort to capture these opportunities.
[THE END]
Disclaimer:
This translated transcript was compiled by Momentum Works from the publicly available earnings call of Alibaba Group held on 19 March 2026. It is intended solely for informational and analytical purposes. The minutes may contain unintentional errors or omissions of some details due to audio quality, accents, and real-time interpretation during the call.
All spoken content remains the copyright and intellectual property of Alibaba Group. Please refer to the company’s official recording or transcript for complete accuracy and authoritative reference.
Any analysis, commentary, or opinions provided by Momentum Works are independent, based on our own research and perspectives, and do not represent the views of Alibaba Group or any other organisations mentioned.
