On August 14, Apple updated its mainland China website to read "Apple Intelligence is now available in mainland China." Two icons sit side by side beneath the headline: Apple Intelligence, and Alibaba Group's Qwen.
After a two-year wait, devices sold in China finally have system-level AI.
But reading this as "Apple's AI has finally arrived" undersells it. Because the day before it went live, Reuters published something else.
Three things in one week
In mid-July, China's cyberspace regulator published its updated registry of generative AI services, and Apple's "Apple Intelligence" appeared on it. Reporting from the same period indicated the approved arrangement involves Alibaba's Qwen, with Baidu as an additional partner. That cleared the single most important regulatory hurdle standing between Apple and the Chinese market.
On August 8, Apple updated its official documentation, confirming that its AI features would support a Qwen extension, with Writing Tools and Siri as the two integration points.
On August 13, Reuters published an exclusive citing people familiar with the matter: Apple is training its own large language model for the Chinese market, with Alibaba's support.
Then the website changed to "now available."
How it works
None of this turns on by itself.
The requirements are specific: macOS 26.6 or later, a device sold through Apple's China channel, and an Apple ID set to the mainland China region.
Activation is manual. Users open System Settings, find "Apple Intelligence & Siri," configure the Qwen extension there, then log into or register a Qwen account before the service becomes active.
Qwen plugs into two places. The first is Writing Tools, where users can call on it inside apps like Notes and Mail to draft and rewrite text. The second is Siri: when a user issues a more complex instruction, Siri first determines whether the task needs Qwen, then asks the user for permission, and only then hands the work over.
Throughout that flow, Siri remains the one taking the instruction, making the judgment call, and asking the user's consent. Qwen is the party being called.
Apple handed over only half
Apple has not given its AI capability to Alibaba.
It has kept its own on-device processing and Private Cloud Compute framework, routing only those tasks that genuinely require a cloud model to a compliance-approved third party.
The design reduces to a sentence: Apple handles orchestration, the local model handles execution, and only what exceeds local capability gets outsourced.
The reasoning is practical. Mainland China has explicit regulatory requirements around data leaving the country and around generated content. Apple's own cloud models cannot satisfy that framework, so routing to a domestic model that has already cleared registration is the only workable path available.
The comparison with the international version makes it clearer. Abroad, this layer of Apple Intelligence connects to OpenAI. In China, that slot is filled by Alibaba. The system architecture did not change. What changed is which model gets called.
Apple does not plan to keep renting
Reuters' August 13 report changes what this arrangement is.
According to the people it cites, Apple is training its own large language model for the Chinese market with Alibaba's support. Qwen, meanwhile, is still slated for integration into the China version of Apple Intelligence.
The report also notes that exactly how Apple's in-house China model and third-party models will work together remains unclear.
If that reporting holds up, Apple's China AI strategy becomes fairly legible: it is not choosing a Chinese model. It is building its own China AI capability while using a domestic model to get there.
This fits Apple's long-standing product logic. Apple wants control over user experience. It will use a partner's capability, but it rarely hands the core experience away.
So Qwen entering Apple Intelligence does not mean Apple has handed China AI to Alibaba. The more likely reading is that Apple is using Alibaba's capability to get from "no China AI" to "a China AI stack of its own."
What Alibaba actually got
Even so, what this means for Alibaba should not be underestimated.
What is genuinely scarce in the model business has never been parameters or benchmark scores. What is scarce is distribution.
A model can post strong benchmarks, attract plenty of developers and run its own app. But if it never enters the operating system, it struggles to become something ordinary people use every day without thinking about it.
Apple Intelligence is exactly that kind of entry point.
The difference is substantial. Opening an AI app is a decision: I am going to use AI. Once AI lives inside the system, users may not even register which model they are using. Draft an email, rewrite a paragraph, interpret an image — the AI just does it, inside the system.
Model brands are moving from the foreground to the background.
One detail is easy to miss: users have to register a Qwen account. That means Alibaba is getting more than call volume. It is getting a direct channel to users, opened from inside an Apple device.
That said, calling this a win for Alibaba is premature.
First, the feature requires manual activation and an account registration. Those two frictions will meaningfully depress real adoption, and adoption is what determines whether this entry point is worth anything.
Second, the final shape of Apple Intelligence in China is not settled. Which capabilities come from Qwen, which come from Apple's own model, and what role Baidu plays have not been fully disclosed.
Third, and most consequentially: if the Reuters reporting is accurate, the pace of Apple's in-house model will determine how long Qwen stays in this arrangement. Today's entry point is not a permanent seat.
The axis of competition is shifting
For the past year, the loudest question in the model business has been whose model is strongest.
But as AI moves into phones, computers and operating systems, the competitive logic is changing. Capability still matters, but distribution is gaining weight.
A model can lead on benchmarks and still be just another standalone service if users have to open an app every time. Once a model is inside the operating system, users are not choosing a model at all. The AI simply shows up inside their workflow.
That is why this partnership is worth examining on its own. What Apple brings is not a headcount of users, but a different mode of distribution. What Alibaba supplies is not just a model, but the piece of local capability Apple needs to operate in China's AI market.
Which means the question worth asking now is no longer why Apple chose Qwen.
It is this: once Apple's own China model is ready, who gets that slot?
Information in this article comes from Apple's mainland China website and official documentation, China's generative AI service registry, and reporting by Reuters and others. Neither Apple nor Alibaba has publicly commented. The final product shape of Apple Intelligence in China, and the division of labor among partner models, remains subject to change. This article does not constitute investment or purchasing advice.