The recently concluded 2026 Alibaba Cloud Summit was regarded by many attendees as one of the most disruptive launch events in the 17-year history of Alibaba Cloud.
Unlike previous technological upgrades centered around a single model, chip, or cloud product, this time, Alibaba Cloud unveiled its full-stack product lineup for the first time centered on Agents. From underlying chips and cloud infrastructure to models and inference platforms, and further to the developer-oriented service system, almost all capabilities have been redefined.
To some extent, this summit was more like a public declaration of a paradigm shift in the industry: AI is evolving from an auxiliary tool to an executing entity, and Alibaba Cloud is shifting its focus from serving humans to serving Agents.
"We have truly entered the Agentic era," said Liu Weiguang, Senior Vice President of Alibaba Cloud Intelligence Group and President of the Public Cloud Business Unit, at the summit. "For the Agent era, Alibaba Cloud is fully ready across the full stack."
Full-Stack AI Upgrade
Over the past few years, the core logic of the large model industry has continued to evolve, shifting from primarily solving conversational problems to possessing long-text comprehension and the ability to reason through complex issues. Post-2025, AI has begun to enter the action-taking phase, driving its commercialization process into a new era.
Shortly before the summit, Alibaba's latest financial report showed that revenue from AI-related products has maintained triple-digit year-over-year growth for 11 consecutive quarters. The proportion of AI-related product revenue in Alibaba Cloud's external commercial revenue exceeded 30% for the first time. The Annual Recurring Revenue (ARR) for AI models and application services has surpassed 8 billion RMB and is expected to exceed 30 billion RMB by the end of the year. Eddie Wu, Chief Executive Officer of Alibaba Group, stated that it is expected that in the next year, the proportion of revenue from AI-related products will exceed 50%, becoming the primary engine driving the growth of cloud business revenue. Alibaba's full-stack AI technology investment has officially crossed the initial incubation stage and entered a positive cycle of scaled commercial returns.
It is reported that as of March 2026, the number of customers on the Bailian platform has grown eightfold year-over-year, and the scale of token consumption over the past three months has increased significantly compared to the previous quarter. Data from QuestMobile shows that Qwen has become the fastest-growing app in terms of user growth this year, with monthly active users surging 4,241% year-over-year.
This also serves as the crucial background for Alibaba Cloud's comprehensive shift towards Agents this time. In Liu Weiguang's view, the three most core changes in the Agentic era are: first, a leapfrog improvement in model capabilities, enabling AI to autonomously complete complex, long-chain tasks, making itself the engine of capability growth; second, AI can directly deliver results rather than just completing processes, and can even directly create "outputs that were previously impossible"; third, a transformation in human-computer interaction models, giving birth to a new paradigm of AI-native organizations. Agents, which surpass human workers by several orders of magnitude and can work continuously 24/7, are emerging. Human employees and Agents form a hybrid work network, even creating new workflows.
Toby Xu, Chief Financial Officer of Alibaba, stated during the earnings call that Alibaba "sees a historical opportunity (in investing in AI)" and "this window of opportunity for us might just be a few years." Feifei Li, Chief Technology Officer of Alibaba Cloud Intelligence Group, also believes that over the next two to three years, the biggest growth opportunity in cloud computing will come from the operation, orchestration, and collaboration of massive Agents. He stated: "Including traditional enterprises and new AI-native organizations and companies, all are rapidly evolving from human-centric workflows and SaaS-based workflows written for humans, towards Agent-centric workflows that provide support for Agent work."
Cloud Built for Agents
At this summit, "for Agents," "helping Agents," and "facing Agents" became high-frequency concepts, and the pace of innovation in product restructuring, process optimization, and application implementation is also advancing in sync.
Liu Weiguang summarized the four cornerstones of the Agentic era: first, model capabilities supporting long-horizon tasks for Agents, providing the necessary leapfrog iteration speed and reliability for production-grade Agents; second, Agentic Cloud supporting Agent operational workloads; third, tools and services supporting flexible Agent invocation, including Skills, MCP, CLI, etc.; fourth, performance and supply capabilities supporting the explosion of inference.
These four demands correspond precisely to Alibaba Cloud's five strategic layouts: underlying chips, Agent Cloud, omni-modal model matrix, model service platform, and Agent applications. Liu Weiguang announced that Alibaba has taken the lead in becoming a cloud and AI vendor that has fully integrated the five-layer full stack.
At the chip layer, during the event, Alibaba Cloud released Zhenwu M890, its latest parallel computing chip for the Agent era, along with the supporting ICNSwitch interconnect chip, aiming to provide continuous upgrade and optimization services for workloads in the Agentic era.
At the Agent Cloud layer, the role of the cloud is to provide critical support for Agent workloads, supporting the development and application of Agents; at the same time, enabling Agents to use a rich array of cloud products just like humans. To this end, Alibaba Cloud has restructured its cloud system from training to inference, specifically addressing the issues of Agent workloads characterized by "irregular elasticity, short lifecycles, and instantaneous scaling up and down."
At the model layer, Alibaba Cloud officially released Qwen 3.7-Max, which places greater emphasis on Agent capabilities compared to the previous generation. It is introduced that 3.7-Max can fully autonomously complete ultra-long-horizon complex tasks for agents lasting up to 35 hours.
At the model service and application layer, Alibaba Cloud launched the official website for "Qwen Cloud," focusing on being "built for Agents." It encapsulates full-chain capabilities such as model selection, model invocation, authentication configuration, and usage queries into Skills, facilitating direct invocation by Agents. In addition, Alibaba Cloud simultaneously launched a Skills portal, uniformly encapsulating common functions such as shopping, travel, and payment into standardized modules that can be directly invoked by Agents.
Front-end industry scenarios each have their unique characteristics, while the back-end model ecosystem is flourishing. Every industry has its own business processes, and each model has different capability boundaries, invocation methods, and usage costs. Liu Weiguang stated that after Agents break through the critical point, they can work non-stop 24/7, creating an endless demand for AI and cloud. Through the five-layer full-stack service upgrade, Alibaba Cloud is building "China's largest AI factory."
Making the Token Business Work
A more critical signal emerges from the capital market: Agents are changing not only the technical architecture but also the business model of the cloud. In the Agent market system, the ability to generate commercial value and the ability to complete tasks are almost equally important.
In the past, cloud vendors primarily sold CPUs, storage, bandwidth, and virtual machines; today, more and more enterprises are directly purchasing model invocation capabilities, inference capabilities, and the ability of Agents to execute tasks. At the summit, Alibaba Cloud explicitly stated that Agent-driven MaaS (Model as a Service) revenue will replace ECS (Elastic Compute Service) as the largest product line—meaning that Alibaba Cloud's growth engine is comprehensively switching to AI revenue measured in tokens.
Yu Wenyuan, Head of Technology at Alibaba Cloud Bailian, pointed out that traditional model invocation is usually just a single question and answer, but Agents are different. They need to understand goals, break down tasks, plan paths, invoke tools, read data, execute tasks, and continuously reason based on task feedback. This means that what Agents bring is not just a simple increase in model invocation, but an exponential increase in the scale of token consumption.
What vendors need to think about is no longer just providing API invocation services, but rather, centering around the actual invocation methods of Agents, combining context reuse, resource scheduling, batch inference, reserved resources, and subscription services, enabling Agents to gradually move from primitive token consumption towards controllable costs, deterministic capacity, and reliable peak performance.
Feifei Li emphasized: "Efficiently generating tokens and turning tokens into the intelligence of real production applications and executable Actions for production applications is the core problem that Agent Cloud needs to solve." Yu Wenyuan also stated: "The positioning of Bailian is no longer a model API service platform, but an Agent inference service platform."
In addition, Alibaba Cloud is also further strengthening its open ecosystem. The role it attempts to play is an irreplaceable infrastructure platform in the AI era. At this summit, Alibaba Cloud announced that multiple leading model vendors, including MiniMax, Zhipu, Kimi, StepFun, and AISphere, will soon be integrated into the Bailian platform.
From the repeated emphasis on the importance of AI in financial reports to the bold stride of full-stack adaptation for Agents at the cloud summit, Alibaba Cloud's actions are sending a signal to the market: the era is changing, and enterprises also need to "rebuild" their DNA. Standing at this critical juncture, whoever takes the lead in securing an advantageous position will be able to grasp the future.
Author | Chen Cun Editor | Qiu Jiangyong Art Editor | Malia Executive Producer | Zhao Chen