Apple gave Alibaba a wake-up call.
On August 8, Apple's official website updated a Chinese guide.
The content is short, with only one core sentence: Eligible Mac users in mainland China can allow Apple Intelligence to call Alibaba's Qwen.
After enabling this feature, users can directly call Qwen via Siri to answer questions, analyze photos and documents; Apple's writing tools can also call Qwen to generate text and images.
At first glance, this is a standard positive piece of news.
Alibaba's model has been integrated into Apple's end devices.
But after reading this guide, I started thinking about another thing instead.
Why is Apple calling Qwen, instead of users using Qwen directly?
These two statements differ by only a few words, but they represent two completely different business models behind them.
In this cooperation, Qwen provides intelligence, Siri is responsible for interaction, Mac controls the end device, and Apple owns the users.
Qwen has become more powerful, but the entrance still does not belong to Alibaba.
I. Why does Apple need Qwen?
Apple faces a special dilemma in China. It has the most profitable end product lines in the world, including iPhone, iPad and Mac. But in China, it cannot fully use its own Apple Intelligence system.
The same Siri that can perform various tasks in the US cannot do the same in China. The same writing tools that have full capabilities overseas cannot access corresponding functions in the Chinese mainland.
In the earnings report for the third quarter of Apple's 2026 fiscal year, Cook specifically emphasized that the Mac business in China region hit its best single-quarter performance in history.
But behind the apparent prosperity, the Chinese PC market is changing.
According to Omdia data, in the first quarter of this year, Mac's shipment in mainland China was about 800,000 units, with a market share of 9%. Lenovo took up 31% of the market share, and Huawei took 16%.
More importantly, domestic manufacturers are turning AI into the underlying infrastructure of operating systems.
Lenovo's "Tianxi" personal agent has become the core of its AI PC strategy. Huawei is also integrating AI into its HarmonyOS ecosystem.
Apple has to move fast and make the right choice precisely.
It does not need a super app that tries to overshadow its own services on its devices. What it needs is a Chinese model that meets Chinese regulatory requirements, has strong model capabilities, and is willing to play a supporting role and embed into the Apple ecosystem as a called party.
Alibaba's Qwen just meets all three of these conditions.
In this scenario, every party holds different cards. Apple controls the user entrance, Siri is the interaction entrance, and Mac is the device entrance. Qwen provides intelligence, and Alibaba is the model provider.
The more I look at it, the more I feel that this role allocation may be the most realistic form of large model commercialization in the future. Not everyone needs to build a super app. Models will become as ubiquitous as electricity and chips, and the entrance will be increasingly controlled by end devices and super applications.
Today, for the first time, Alibaba has entered a world-class consumer end device as an intelligent infrastructure provider.
But is this the starting point of victory, or the beginning of complete loss of control over user access?
II. Qwen is becoming a new product of Alibaba Cloud
Looking back at Alibaba's moves in the AI field this year, we can clearly see a highly intensive evolution curve.
During this year's Spring Festival, the Qwen app became an instant hit. By June, the monthly active users of the Qwen app on mobile terminals reached 167 million, reaching its peak popularity for a while.
But if we broaden our perspective, you will find that what really excites Alibaba's core senior management has never been the C-end chat box, but the B-end cloud and infrastructure business.
On August 3, Alibaba officially released its flagship model Qwen3.8-Max. On the same day, the LMSYS Org large model arena ranking was updated, and Qwen ranked second in the world in overall performance, second only to Anthropic Claude Fable 5.
It has 2.4 trillion parameters, and its programming capability ranks top 4 in the world. In official experiments, the model ran unattended for 16 consecutive days relying only on a blank folder, independently completed a full open source project and submitted 265 pieces of code in total.
In the Hugging Face global developer community, Qwen has long ranked first among global open source base models, and is one of the open source bases with the largest number of derived models and the highest total download volume so far.
All these impressive data essentially convey one signal: Alibaba is building Qwen into a standardized industrial product that can be seamlessly called by cloud providers, enterprises and third-party end devices, just like water, electricity and computing power.
On the same day when Qwen3.8 was released, Alibaba also did another major thing: it launched the full public beta of "Qwen Office".
If we put the three events of releasing the strongest model, launching Qwen Office and integrating into Apple Mac in the same time period, Alibaba's AI layout has become extremely clear:
The bottom layer is the extreme iteration of the Qwen series models;
The upper layer is the huge computing power base composed of Alibaba Cloud Zhenwu M890 super nodes;
The next upper layer is to make models API-based and platform-based through ModelStudio platform;
The top layer is to embed Qwen into office, development and enterprise workflows.
As you can see, Alibaba is still relying on infrastructure services to enable more enterprises and users to use AI.
This is highly consistent with what Alibaba has been best at in the past. In the early days, the core of Taobao was to make it easy for goods to be sold anywhere, Alipay was to make transactions happen seamlessly, and Alibaba Cloud was to allow others to develop applications easily.
Following this logic, Qwen's future position may be to allow everyone to call the capabilities of Qwen, rather than just letting everyone use the Qwen app directly.
This is undoubtedly a huge business.
But problems also arise here.
III. The stronger the model, the more important the entrance becomes
Imagine a scenario. What will happen after a Mac user enables Qwen?
He says a request to Siri. Qwen processes the request and sends the answer to Siri, then Siri outputs the answer to the user.
The impression left in the user's mind is that Siri is getting smarter and smarter.
As for whether the model called behind the scene is Qwen, Claude or any other model, 99% of ordinary users do not care at all, nor can they perceive the difference.
Qwen puts in all the efforts, but Apple takes away all user trust and brand premium.
Looking at the entire technology industry, a complete industrial chain is taking shape.
Hardware manufacturers control the end device entrance, operating systems control the interaction entrance, WeChat/TikTok control the social and traffic entrance, and office giants control the work scenario entrance. Pure model companies are forced to the very bottom of the industrial chain, at the computing power and inference layer.
The stronger the model is, the more likely it is to be hidden behind the scene.
In the Internet era, the core assets are user relationship and traffic entrance; but in the AI era, the industrial chain is restructured as: User ➔ End device/Entrance ➔ Agent scheduling layer ➔ Model ➔ Computing power.
As you can see, the model is pushed two layers backward in the chain.
The more perfect Qwen performs, the smoother the system packaging will be, and the less users need to contact Qwen directly.
For Alibaba, which has strong ambitions to build a platform, this is not a technological breakthrough, but a business role choice related to its survival in the future.
IV. What Alibaba really lacks is not the top model in the world
Alibaba has proved with its achievements that it can build world-class models, but it has not yet proved another thing: how to build an AI native application that users actively open every day and cannot forget.
The explosive popularity of the Qwen app after the Spring Festival once brought Alibaba a great C-end opportunity. But when the subsidy fades, all large model companies will face a problem:
A pure Chatbot (AI chat box) is not a high-frequency rigid demand at all.
Why would users chat with an AI for no reason every day? If AI cannot help you finish the work directly, the chat box is just a toy that will be abandoned after the novelty wears off.
Alibaba's senior management understands this better than anyone else. Qwen cannot only be a Chatbot. It has to enter the office, e-commerce and search sectors. It has to enter mobile phones, PCs. It has to embed into browsers, enterprise workflows and cloud services.
Among all scenarios, the office sector is the battlefield that cannot be lost at all. Because the office scenario is a natural super Agent entrance in the AI era.
The value of chat is to tell me the answer, while the value of office AI is to complete the work for me. These are two completely different product logics.
Office software naturally masters documents, emails, calendars, meetings, address books, approvals, projects and enterprise data.
After the Agent is fully implemented, office software will become the work entrance. Users cannot even work without opening it.
Therefore, Qwen Office is not just a new product, it is a game that Alibaba cannot afford to lose in the entrance war.
V. Qwen Office is Alibaba's high-stakes bet on entrance
When it comes to office sector, we have to mention Tencent and ByteDance.
The real competitors of Qwen Office are Tencent's WorkBuddy and ByteDance's combination of Doubao and Feishu, not any single AI chat application.
Let's talk about Tencent first.
WorkBuddy was released in March this year. By June, its monthly visits exceeded 20.97 million times, surpassing the sum of the second and third place in the industry.
Frankly speaking, this number made me re-evaluate Tencent's strength in this field.
In the past two years, Tencent has not been the most high-profile player in the large AI model competition. But WorkBuddy has blazed its own path. What it is fighting for is one thing: the work entrance, not just an Agent product.
What cards does Tencent hold? WeChat, Tencent Meeting, Tencent Docs, WeCom, email, Tencent Cloud.
WorkBuddy can connect all these products together, so users do not need to open different applications. They only need to tell WorkBuddy what to do.
This idea is much smarter than building a stronger Chatbot.
Then let's talk about ByteDance.
ByteDance's advantages are more direct: it holds all five cards of users, content, recommendation, model and Agent. Doubao has a huge C-end user base, Feishu has enterprise assets, and Volcano Engine has infrastructure. ByteDance is trying to use AI to open up the full-link entrance from personal entertainment to work productivity.
Then what about Alibaba?
Alibaba has a card that neither Tencent nor ByteDance has.
DingTalk.
DingTalk serves more than 26 million enterprise organizations, with about 200 million monthly active users, of which nearly 8 million enterprises have accessed DingTalk's AI capabilities. In China's enterprise service market, this scale is overwhelming. Therefore, the change of CEO of DingTalk is very critical.
On June 11, Chen Yusen succeeded Chen Hang as the CEO of DingTalk. At the age of 34, he is the youngest division CEO in Alibaba's history.
He is the founder of MuleRun, the person who created a general Agent product from scratch within the Alibaba system.
Within one week after taking office, he promoted the integration of Wukong and MuleRun. In early July, he also added QoderWork into the unified product line. The three Agent products, which cover enterprise collaboration, cloud execution and desktop operation respectively, were integrated into "Qwen Office". (Extended reading:Chen Yusen did not follow Alibaba's old path)
Alibaba has figured out one thing: it cannot fight the entrance war in the office sector separately with multiple scattered products.
In the past, Alibaba ran four or five Agent products with similar functions at the same time internally: Wukong for DingTalk ecosystem, QoderWork for desktop, MuleRun for cloud, and JVS Claw.
Qwen Office ended this internal horse race. Chen Yusen once said a very interesting sentence in an interview with Yicai. He said that this AI office war started very early, back in 2024 when a company called Cognition launched Devin.
If Qwen is fighting for what users ask, Qwen Office is fighting for what users do next.
Only when the latter succeeds can Agent really have the opportunity to generate continuous value.
VI. Qwen has reached a fork in the road
Today's Alibaba and Qwen are standing at a fateful fork in the road.
The first path: Qwen acts as a pure AI infrastructure provider.
Alibaba will iterate models to the extreme, Alibaba Cloud sells computing power, Qwen sells APIs at high prices, and all walks of life including Apple and mobile phone manufacturers will call its models. This path is extremely logical, it fully conforms to Alibaba Cloud's business genes, and the certainty is extremely high.
Every time the model makes a breakthrough, Alibaba Cloud's moat will be deepened. But the end of this path is that Qwen will become the unsung hero behind the AI development in Silicon Valley and China, but may lose the opportunity to become a platform-level app in the next era.
The second path: Qwen builds its own AI entrance.
Qwen will fully cover DingTalk, Taobao, PCs, mobile phones, browsers and personal Agents. Qwen directly understands user demands, and the Agent completes all business behaviors for users. Eventually, Alibaba will hold the core user relationship in its own hands.
This path is extremely attractive, but extremely difficult to take. Because Alibaba has to solve the most original proposition again: without strong social bonds, what makes users willing to open your app every day?
Taobao has transactions, Alipay has payments, and DingTalk has check-in and work scenarios. What about Qwen in the future?
Alibaba can easily win the first path of being an infrastructure provider; but the second path of building a super entrance is what Alibaba has not yet proved its capability in the new AI era.
VII. What is the real reminder Apple gave to Alibaba?
Looking back at the notice on Apple's official website on August 8, everything seems to be full of profound meaning.
Apple's integration of Qwen gave Alibaba a public business lesson on ecosystem and entrance.
Apple is not telling Alibaba how powerful Qwen is. On the contrary, it is telling Alibaba: I can help you prove that your model is very powerful, but I will not help you build your own entrance.
It hides Qwen behind Siri, and inside macOS.
Users do not need to know Qwen at all.
Qwen is called, solves the problem and provides value. But users will always remember Apple.
This is the power of entrance hegemony.
A possible division of labor in the future AI industrial chain may be like this: end devices