Apple built a China AI model with Alibaba

Apple has trained its own large language model specifically for the Chinese market, with technical support from Alibaba, according to Reuters sources. The model exists because China’s AI regulators require locally registered, government-approved systems for any generative AI product sold there, a bar Apple’s usual approach, pairing its own on-device models with a partner like OpenAI for cloud queries, cannot clear inside China. Alibaba’s Qwen model handles some of the underlying capability, while Baidu’s technology reportedly supports search-related functions, and China’s Cyberspace Administration registered Apple’s generative AI service in July after a review that delayed the China rollout of Apple Intelligence by nearly two years.

This is a dual-track approach: Apple’s global Apple Intelligence stays built on its own models plus outside partners, while the China version runs on a separate, locally compliant stack built specifically to satisfy Beijing’s content and data rules.

What it means for you

This is what AI localization actually costs once a market has hard content-approval requirements, not a config flag or a different API endpoint, but a distinct model built and registered specifically for that jurisdiction, with different partners underneath it. If you’re planning to ship an AI feature into China, or any market moving toward similarly strict AI registration rules, Apple’s two-year delay is a more realistic planning number than the weeks a routine international rollout usually takes. It’s also a data point on how far even the most vertically integrated hardware company will go to keep a single global AI stack unified: not very far, once national regulation demands otherwise.