Qwen 中国的iPhones 少说云端，少说退票

中文版正在随流水线自动整理，稍后刷新即可；可先阅读下方英文原文。

本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

NextFin News — When Apple confirmed that Alibaba’s Qwen would plug into Siri, writing tools, and visual features on iPhones in mainland China, the first reaction on the tape was familiar: more devices, more calls, more cloud revenue. That reading skips the product design Apple actually published. Qwen is not the invisible default for every handset. Users must switch it on. The first features largely match what Apple has shipped elsewhere since 2024. The next full Siri model stack is not included.

So the useful question is not how many iPhones exist in China. It is what Alibaba is paying—in cash, in cost, or in concession—for a seat on a high-value lock screen, and what it can do from that seat that a standalone App cannot.

A distribution seat, not a guaranteed meter

Apple has a pattern in these deals. When it needs a model it treats as core intelligence, it has been willing to pay, as in reported arrangements around Google technology for new Siri, running in Apple-controlled environments. When it controls a default doorway—as in search—it has collected very large sums from partners. When it offered ChatGPT as a Siri option, reporting indicated Apple did not fund the placement and kept room to share in any monetization on its platform.

Qwen’s first posting looks like the ChatGPT template: optional, account-gated, secondary to Apple’s own on-device and private-cloud baseline. In that posture, token fees are more likely a cost of presence than a windfall invoice to Cupertino. Apple is also guiding to tighter gross margins while memory chips stay expensive; shifting cloud inference onto a partner is convenient. Broker notes that lift Alibaba Cloud forecasts on the back of the tie-up imply annual increments that are hard to reconcile with eligible-device counts, realistic opt-in, and published Qwen list prices—unless every optimistic assumption is stacked at once. Relative to Alibaba’s existing AI cloud run-rate and capital spend, pure inference from this channel sits in the noise.

Why Alibaba still wants the switch

The constraint on phone agents in China is not model quality. It is whether WeChat, payments, shopping, and local services allow an outside assistant to act. ByteDance’s system-level helper has run into long public refusal lists across those categories, including major Alibaba and Tencent apps, and has had to lean on narrower partnerships and its own shopping surfaces.

Alibaba is both a model provider and the operator of Taobao, Alipay, Amap, and Ele.me. Public reporting has placed it on Apple’s mainland AI path. A system assistant that can only paraphrase text is a demo. One that can move a user from intent to order, pay, and navigate touches the businesses that still define Alibaba’s consumer franchise. No party has published a roadmap for deep Siri hooks into those apps; the strategic logic does not require one to be visible yet. Alibaba has already pushed agent-style shopping inside its own products at large scale, while on-site customer-management revenue—tied heavily to search ads—has been soft. Losing the first utterance of a purchase intent to a generic assistant is a longer-term threat than missing a small token line. An iPhone extension is a defensive claim on that utterance among users who spend.

What still has to go right

Apple’s recent iPhone share gains in Greater China arrived without mainland Apple Intelligence. The feature is insurance for the platform, not a growth lever Apple must rush. That limits how hard Alibaba can push for a primary-brain role and helps explain a modest, opt-in slot. Overseas, OpenAI described its own Siri integration as slow while defaults stayed off. Enablement, weekly return, and task completion will decide whether Qwen’s presence is real. Even a solid slice of a roughly 100-million eligible-device pool is incremental next to the monthly reach of leading consumer AI apps in China.

On-device models will keep absorbing simple rewrites and lookups, trimming cloud calls over time. Who eventually powers a full mainland Siri stack may not be settled until 2027 or later. Until then, December’s software update is less a cloud earnings event than a distribution test: how many people flip the switch, how often they come back, and—if commerce links appear—whether intent turns into a paid order inside Alibaba’s services rather than into a dead-end answer.

Auto-fetched and translated by AI Atlas. All rights belong to the original author.