Apple Intelligence Cleared for China on Alibaba's Qwen Models
China's Ministry of Industry and Information Technology has approved Alibaba's Tongyi Qianwen models for use in Apple Intelligence in China.
Tongyi Qianwen, better known as Qwen, is a family of large language models, systems trained on large amounts of text to write drafts and answer questions. The clearance, reported on October 14, also covers Tongyi Wanxiang, Alibaba's tool for generating images and video from prompts. It removes the last model-level regulatory hurdle for Apple Intelligence in mainland China, where the service had not yet rolled out nearly two years after its launch. Engadget
The timing is tied to hardware. Chinese regulatory filings included network access permits, the licenses phones need to connect to local networks, for the iPhone Duo and iPhone 18 Pro. The iPhone Duo was set to launch in China the next week with a new and improved Siri AI as a flagship software feature. Engadget
Qwen was confirmed to be central to Apple Intelligence on compatible iPhone, iPad, Mac and Vision Pro models sold to Chinese customers. That scope matters. It points to a full deployment across Apple's product line in China, like installing the same engine across different cars, not a single-device exception for the new handset. China approved Alibaba AI models for use in the iPhone Duo and other Apple handsets, according to the same October 14 reporting. SCMP
Apple had already received clearance to roll out Apple Intelligence in mainland China in July but had not yet done so at the time of the report. The Apple Intelligence AI service was registered with China's cyberspace regulator, a step reported on July 15. That regulator statement did not give a launch date for Apple Intelligence in China. Reuters
The MIIT action follows an 18-month build-up. In February 2025, Apple teamed up with Alibaba to develop AI features for iPhone in China. Apple and Alibaba then submitted their co-developed Chinese AI features for approval by China's cyberspace regulator. Reuters
Two August data points filled in the technical design. On August 8, Apple said Mac users in China could connect to Alibaba's Qwen AI service. Reuters On August 14, Reuters reported that Apple was training its own AI model for the China market with Alibaba's support. Reuters
In the United States, Apple worked with Google and OpenAI to train Apple Intelligence models and hand off complex questions. The China configuration is different in its suppliers but similar in structure, with local foundation models, the large base models other features build on, and external handoff partners rather than a single monolithic model.
The broader context here is localization of the inference stack, the place where user requests are actually processed. Apple is maintaining separate model supply chains by jurisdiction. Qwen and Wanxiang handle the regulated generative functions inside China, while the August reporting on Apple training its own model with Alibaba support suggests Apple retains control over product-specific tuning and Siri behavior. For engineers shipping across regions, that implies divergent evaluation, safety testing and release gating. A prompt that routes one way in the US may route another way in China, with different latency, or response speed, capability and failure modes.
In my view, the delay is the point worth studying. Holding a flagship software feature for nearly two years while waiting for model and service approvals shows how launch criticality has shifted. Hardware readiness and OS readiness were not sufficient. Model registration and network access permits set the schedule. Tying the final MIIT approval to the iPhone Duo launch week aligns software availability with a hardware refresh, which gives Apple a clean install base for telemetry, or usage data from devices, and iteration.
What this enables, if rollout proceeds, is parity. Chinese customers on compatible devices would finally get the Apple Intelligence features that have been standard elsewhere, powered by domestic models and reached through Siri, writing tools and image generation. My kids adopted new phones for the camera. They kept using them for the software. Siri will face that same test. If the new assistant handles everyday Mandarin queries reliably and hands off cleanly when it cannot, adoption will follow quickly.
