Apple Intelligence is at last heading to China, and it will run in part on Alibaba's Qwen models. China's Cyberspace Administration has added Apple's on-device generative AI to its list of approved providers, the regulatory clearance Apple had been waiting on for well over a year, and Apple will draw on models from both Alibaba and Baidu for users in the country. Alibaba's US-listed shares jumped around 5 percent on the news, a measure of how much investors value getting Chinese models inside the iPhone.
The regulatory piece is the real unlock. China requires generative-AI services to use locally vetted, approved models and to clear a government review, which is exactly why Apple Intelligence had rolled out almost everywhere except Apple's single most important manufacturing base and one of its largest markets. By partnering with domestic champions and landing on the Cyberspace Administration's approved list, which also includes homegrown players like Huawei, Apple has finally found a way to thread that needle. Using two local partners, Alibaba and Baidu, rather than one, gives it some redundancy and political cover at the same time.
For iPhone, iPad, Mac, and Vision Pro owners in China, the plan is to fold Qwen's capabilities, text and image understanding and generation, into the operating system, so people can use them without hopping between separate apps. That matters commercially, because Apple has been losing ground in China to domestic rivals whose phones already ship with strong on-device AI baked in. Getting a comparable feature set into iOS is less about leapfrogging those rivals than about no longer being conspicuously behind them.
The deal is a significant win for Alibaba, whose Qwen family has quietly become one of the most widely used model lines in the world, and it is a vivid reminder of how differently the AI map is drawn inside China, where the models a global company reaches for are Chinese ones. One striking technical side note underlines how fast the ground is shifting, PrismML, a Khosla Ventures backed spinout from Caltech, has compressed an open Qwen model from roughly 54 GB down to under 4 GB, small enough for its full 27 billion parameters to run on an iPhone 15 or newer. On-device AI that once needed a data center is increasingly fitting in a pocket.
Why it matters goes beyond one market. This closes one of the longest running gaps in Apple's AI story, its conspicuous absence in China, and it does so by leaning on a Chinese lab rather than its own model or a Western one. For Alibaba, it is validation at the highest possible level, its models running inside the iPhone. And for anyone tracking the widening split between the American and Chinese AI ecosystems, it is a concrete example of that split in action, the same phone with a different brain depending on which side of the border it is sold on.