Apple began testing its long-delayed Apple Intelligence China features with registered software developers on Wednesday, deploying Alibaba‘s Qwen model to power generative tasks. The trial arrives almost two years after the system launched in Western markets, clearing the path for an expected public release across mainland China in December.
Key points
- Apple has opened developer testing for Apple Intelligence in China using Alibaba’s Qwen model for complex cloud queries.
- The rollout covers 18 initial tools, including call translation and message proofreading, while holding back the updated Siri voice assistant.
- Foreign AI models from OpenAI and Google remain inaccessible in China due to strict domestic data localisation and censorship rules.
- More than 110 million iPhones in China can run the software, though local rivals already offer deeply integrated mobile AI agents.
How Apple adapted its AI for mainland China
The developer beta introduces Apple Intelligence features to users in mainland China for the first time since the software debuted in North America in autumn 2024. In overseas markets, Apple routes complex user queries that exceed on-device processing limits to its Private Cloud Compute infrastructure or third-party models such as OpenAI’s ChatGPT. In China, regulatory barriers make that setup impossible.
Chinese internet laws require artificial intelligence providers to register algorithms with state regulators, operate within local data centres, and process citizen data strictly within national borders. Because Western firms including OpenAI, Google, and Anthropic block access from mainland IP addresses and lack state approval, Apple had to find a domestic partner. The company received formal regulatory clearance in July after assessing several local technology giants, including Baidu, ByteDance, and Tencent, before settling on Alibaba.
Under the arrangement, Alibaba’s open-weight Qwen model handles server-side generative text tasks that the iPhone’s local neural engine cannot run alone. The initial test build contains 18 discrete features. Users gain access to real-time phone call translation, system-wide proofreading, message summarisation, and automated transcription clean-ups. Notably absent from the test build is Apple’s rebuilt Siri assistant, which relies on deeper system hooks and more complex reasoning layers that remain in development for the Chinese language.
Why Alibaba’s Qwen secured the partnership
Alibaba develops Qwen as a suite of proprietary and open-weight models capable of drafting text, translating languages, parsing documents, and generating programming code. Unlike closed systems developed in the United States, Alibaba distributes weights for many Qwen variants openly, allowing enterprise customers and independent engineers to run and adapt the software on their own infrastructure.
That distribution model has turned Qwen into the most downloaded open-weight model family in the world. Industry research tracking open foundation models found that by February 2026, Qwen served as the architectural base for 69 percent of newly adapted models created by software developers, up from roughly half a year earlier. During the same period, European open models saw their developer share drop to four percent.
Apple is not the first multinational business to select Qwen to navigate Chinese compliance requirements. German car manufacturer BMW is integrating an in-car virtual assistant powered by Qwen for its Neue Klasse vehicles assembled in China from 2026. The model family has also made inroads into international sport, with UEFA signing a multi-year deal with Alibaba to deploy Qwen for automated content moderation and digital supporter engagement across European competitions.
Regulatory hurdles and structural concessions
The gap between the American debut of Apple Intelligence and its Chinese developer preview reflects the friction foreign device makers face under Beijing’s strict software regime. Regulators require all commercial generative systems to pass rigorous safety reviews, submit training corpora details, and adhere to strict political filtering standards.
Chinese authorities are tightening those requirements further as deployment scales. As officials draft China AI safety standards to govern foundation models from raw data collection through post-release updates, device makers must ensure every external API connection complies with domestic law. Apple could not simply export its existing server architecture to Chinese server farms without extensive code audits and algorithmic registration filings.
Industry analysts view the arrangement as a pragmatic compromise that alters Apple’s standard operating model. Zhang Yi, founder of the research firm iiMedia, told AFP that turning to domestic partners “signifies a structural concession by Apple regarding its pricing and defining power over the AI experience in the Chinese market”. Zhang pointed out that legal compliance costs, rather than fundamental engineering roadblocks, dictated the two-year delay.
By yielding cloud inference to Alibaba, Apple protects its handset sales while sacrificing the unified global software stack it prefers to maintain across iOS devices.
The competitive field in the Chinese smartphone market
Apple faces an uphill climb to convince Chinese consumers that its AI offering justifies upgrading older hardware. Domestic competitors such as Huawei, Xiaomi, Oppo, and Vivo spent the past two years building proprietary AI assistants directly into their mobile operating systems, tailoring them to Chinese social platforms and payment networks.
Ivan Lam, a senior analyst at Counterpoint Research, said the update “fills a long-standing gap in Apple’s AI offering” but added that the release does not necessarily provide the hardware maker with “a significant advantage” in an already crowded retail sector.
Even so, Apple holds an expansive base of compatible hardware. Data compiled by market research group Omdia indicates that more than 110 million iPhones currently active in mainland China possess the silicon necessary to execute on-device Apple Intelligence tasks. That installed base gives the company immediate distribution once the software leaves developer testing.
The challenge lies in engagement. “Chinese mobile phone manufacturers and internet companies are actively developing AI features and agent ecosystems,” Zhong Xiaolei, a senior analyst at Omdia, told AFP. Domestic handsets already execute multi-step user tasks, such as booking train tickets through WeChat or scanning local delivery receipts, using bespoke agent frameworks that Apple has yet to replicate.
What the trial means for Apple’s bottom line
Greater China, which encompasses the mainland, Hong Kong, and Taiwan, remains central to Apple’s financial health. Revenue from the region expanded by 21 percent during the most recent quarter, providing momentum as executive leadership changes take hold in Cupertino.
With John Ternus taking the helm as chief executive, the company must prove it can defend premium margins in markets where domestic engineering has closed the hardware gap. Delivering functional, localized software tools is widely regarded by investors as essential to sustaining replacement cycles among Chinese iPhone owners.
The current developer beta operates on test builds of iOS, iPadOS, and macOS. Following this limited testing phase, Apple plans to open the features to the wider Chinese public in December through an over-the-air system update, bringing the 18 initial Qwen-backed capabilities to consumer devices before the end of the year.
