Apple Trained Its Own AI Model for China, Making It the First Foreign Company to Get Beijing’s Approval

Summary

Apple worked with Alibaba to train a large language model built specifically for the China market, according to Reuters. The proprietary model will run alongside Alibaba’s Qwen system as part of a dual-track strategy that lets Apple navigate China’s strict AI content regulations without handing the keys entirely to a domestic partner. Apple is the first foreign company to receive Chinese government approval to operate a proprietary AI model in the country. The rollout of Apple Intelligence features in China is expected in the coming months, with Siri AI launching on devices running iOS 27 or later. Apple briefly published a support guide in China showing Mac users how to connect Qwen to Siri and Writing Tools, then pulled it within a day.

Why it matters

Apple Intelligence has been live in the US, UK, and other markets for over a year. Getting it into China, with its 1.4 billion potential users, is the last major market gap. For those who travel to or have family connections in China, this explains why iPhones sold there have lagged behind in AI features. The fact that Apple trained its own model rather than fully outsourcing signals it wants more control over the AI experience it delivers everywhere.

HTD Says

Apple getting Beijing’s sign-off to run its own AI model in China is a bigger deal than it sounds, because it means Apple Intelligence is finally on track to land in one of the few places it has been completely absent. This is interesting on so many levels. For starters, developing this model specific to this massive market must have been a herculean effort.

On top of this, getting US government sign-off to ship overseas, getting validated by the Chinese government, and aligning with iOS 27 rollout timelines is a project manager’s nightmare. Apple trained the model for this market, betting on huge adoption. If it plays out, this could be a massive bump for Apple.

And I wonder about this dual-stack implementation. How will this work? Which model will have priority over the other? Will there be on-device “discussions” between the models? My head is spinning a bit.

Source:

MacRumors