On August 3rd, Alibaba released its flagship model, Qwen 3.8, and launched the public beta of "Qwen Office," marking a simultaneous rollout of both the model and a productivity application.
This dual release fills a critical gap in Alibaba's AI narrative: Qwen 3.8 provides more advanced and cost-effective model capabilities, while Qwen Office integrates the model into high-frequency workflows.
The core of this trade lies in whether AI investments can translate into revenue. Today, Alibaba's Hong Kong-listed shares surged over 6% intraday, a positive signal from the capital markets.
Alibaba's valuation has long been driven by two competing narratives. E-commerce continues to generate cash flow but is pressured by consumer spending growth, platform subsidies, and investments in instant retail. Meanwhile, AI and cloud computing offer higher growth expectations, but whether capital expenditure will convert into profits still requires more performance validation. If Qwen Office can generate enterprise-level usage, Alibaba could create its next product for token flow.
The Battle for the Office Market Intensifies
Over the past two years, most office agents have focused on solving individual productivity issues. An employee could use an agent to search for information, write reports, or create spreadsheets, shortening their delivery time. However, enterprise efficiency depends on handoffs between multiple departments: whether market data enters the sales system, whether sales commitments are synchronized with legal and delivery teams, and whether financial metrics remain consistent across different teams. If data remains siloed, permissions are incompatible, and downstream teams need to re-verify information, the organization's total delivery cycle may not shorten even if one step speeds up.
Therefore, the enterprise office market requires a different product form. The product must understand organizational structure and permissions, connect to databases, knowledge bases, and business systems, support cross-departmental workflow orchestration, and retain audit, version, and responsibility records. While a personal agent pursues the completion quality of a single task, enterprise-level products must also handle data governance, standard reuse, and multi-person collaboration. Ultimately, organizations purchase a manageable and replicable production system.
Competition in the domestic office market has subsequently expanded. WPS is advancing from document entry points to WPS 365 and the native office agent "WPS Lingxi," with strengths in formatting capabilities, document stock, and government enterprise clients. Tencent is entering through desktop agents. WorkBuddy can read and write local files, autonomously decompose tasks, and call tools, while the enterprise version can connect to internal systems. WeCom, QQ, Tencent Docs, and Tencent Cloud provide distribution and infrastructure for it.
ByteDance is integrating the capabilities of its Doubao and Feishu product lines to launch the Doubao Enterprise Edition for organizations. The Seed model handles complex tasks, data processing, multimedia generation, and computer and browser operations, while Feishu provides enterprise knowledge, organizational permissions, and a collaboration environment. AI-generated documents, spreadsheets, and other outputs can directly flow within Feishu, allowing ByteDance to extend personal AI usage to team collaboration.
Alibaba's approach is closer to a complete enterprise agent architecture. Qwen Office is integrated from QoderWork, MuleRun, and Wukong, simultaneously supporting desktop agents, cloud agents, and enterprise collaboration agents. It will later connect to DingTalk IM, enterprise databases, and workflows. Qwen determines the ceiling of intelligence, DingTalk provides organizational relationships, permissions, and process entry points, and Alibaba Cloud handles data, computing power, and token billing.
This full-stack combination is a relatively clear advantage for Alibaba, but it also tests whether the three systems can form a unified experience. A law firm could deposit an M&A due diligence process as an organizational-level Skill, and a multinational team could let agents continue projects after employees leave for the day. Such scenarios transform personal experience into enterprise assets. Once in a production environment, the token consumption generated is far higher than a single Q&A or piece of text generation.
From Stronger Models to Greater Token Consumption
Qwen 3.8 provides a new model foundation for this entry point. According to Alibaba's disclosures, Qwen 3.8-Max has a total of 2.4 trillion parameters, with 95 billion activated parameters. It supports a 100 million token context and focuses on improving capabilities in coding, cowork, and long-cycle agent tasks. Domestic API pricing is set at 12 RMB per million tokens for input and 36 RMB for output, with a cache hit price of 1.5 RMB.
Model capability determines whether an agent can complete complex tasks, while unit cost determines whether enterprises are willing to scale usage. When both improve simultaneously, a typical demand elasticity appears: as the cost per single call decreases, more tasks become economically viable. An agent might expand from generating a single PPT slide to continuously reading hundreds of documents, calling multiple systems, and iterating repeatedly, causing the total token consumption of a single task to actually increase.
Office scenarios can provide more stable and higher-intensity call volumes compared to consumer-side Q&A. Consumer-side Q&A often consists of short dialogues, where user willingness to pay and retention can fluctuate easily. Enterprise tasks often involve long documents, images, videos, and historical knowledge, and include multiple cycles of planning, tool calls, and result verification. The call intensity of a due diligence, audit, or business analysis task could be far higher than an average chat.
Once an enterprise embeds an agent into its standard processes, the migration cost also increases with permission configurations, knowledge bases, and Skill deposits, resulting in revenue quality that is typically superior to one-time traffic conversion.
Agents' contribution to cloud vendors' performance is gradually becoming evident. In the quarter ending March 2026, Alibaba Cloud revenue grew 38% year-over-year, with external commercialization revenue growing 40%. AI-related product revenue achieved triple-digit growth for the 11th consecutive quarter, accounting for 30% of external cloud revenue. The number of Bailian customers grew 8 times year-over-year.
Securities firms have a remarkably consistent view of this change. Citigroup defines Alibaba as the primary beneficiary of China's Token economy and estimates that MaaS could become Alibaba Cloud's largest revenue product. Morgan Stanley projects 45% cloud revenue growth in the first quarter of fiscal 2027, with the cloud EBITA margin rising from 9% in the previous quarter to 11%. HSBC believes that an increasing MaaS share, expanded deployment of self-developed chips, and cloud product price increases will jointly improve cloud profit margins.
These changes reflect the core impact of AI on Alibaba's valuation: the market is now using token call volume, MaaS revenue, and cloud profit margins to measure the return on Alibaba's AI investments. Qwen 3.8 provides capabilities close to frontier models at a lower price, which may suppress revenue per token in the short term. However, lower costs will also drive customers to delegate more processes to agents. As long as call volume growth outpaces price declines, and self-developed chips, caching, and sparse architectures continue to reduce inference costs, cloud revenue and profit margins can improve simultaneously.
Market Requires AI Scenario Expansion
As the capabilities of leading models continue to improve and inference prices keep falling, competition in the AI industry is extending to the scenario side. Model leaderboards determine the upper limit of technology, but real-world scenarios determine call frequency, customer retention, and payment scale. The difficulty of scenario expansion lies in enterprise workflows. Companies will ask whether agents can stably complete long-cycle tasks, inherit organizational permissions and protect data, interface with existing systems and retain audit records, and calculate the labor and time saved by each task.
Therefore, the industry is searching for high-frequency, long-context, repeatable tasks. Programming, office work, customer service, marketing, and professional research have become the main entry points. Among these, office work covers more departments and involves more complex data relationships. Vendors need to simultaneously address model effectiveness, system integration, and organizational governance. The product form will gradually evolve from a single-point assistant to an enterprise-level execution platform.
To judge whether a scenario is viable, the market also needs a set of harder metrics: the number of enterprise clients and payment rates, token consumption and retention per client, task completion rates, the reuse frequency of organizational-level Skills, and the revenue and profit generated from call volume. For Alibaba, Qwen Office shoulders the task of bringing Qwen into enterprise workflows and then converting those tasks into call volume for Alibaba Cloud. For the entire industry, the next phase of the AI narrative will come from more real-world scenarios and the ability of these scenarios to sustain the continuous flow of tokens.
Comments