China’s Cyberspace Administration (CAC) officially registered Apple Intelligence on July 15, clearing Apple’s AI feature suite for use across iOS, iPadOS, macOS, and visionOS in the country. The approval ends a wait stretching back to 2024, when Apple Intelligence launched globally but was blocked from China due to local rules requiring separate registration and certification of large language models.

Rather than trying to get a single model approved, Apple took a different path: two separate Chinese AI partners, each handling different functions.

Alibaba’s Qwen model handles the core generative AI work: text summarisation, email drafting, document editing, and image generation. These are the features Apple users in the US get from OpenAI’s models, but in China, Qwen takes that role.

Baidu powers specific features including Visual Intelligence, the camera-based AI that can identify objects, translate text in real time, and answer questions about what you’re looking at. Baidu also handles search-related functions, mirroring what Google’s Gemini and OpenAI handle in other markets.

Both partnerships were required to pass CAC review, which scrutinises how AI models handle content, data storage, and access to sensitive topics.

Why Apple Needed Two Partners

The dual-partner setup reflects the structure of China’s AI regulatory landscape more than it reflects Apple’s preferences. CAC approval is granted per model, not per product, so Apple had to identify which models could pass the registration process and align those models with the specific features they’d power.

Alibaba’s Qwen has already cleared multiple CAC certifications for other use cases, making it a practical choice for the generative AI workload. Baidu’s long history of compliance in China, including its registered Ernie model, made it the natural partner for visual and search functions.

A Two-Year Gap Closed

Apple shipped Apple Intelligence features to users in the US, UK, Australia, Canada, and most of Europe starting in late 2024. Chinese users, representing Apple’s second-largest market by revenue, have been left out entirely.

That matters commercially. Apple’s China revenue has been under pressure from local Android manufacturers offering comparable hardware at lower prices. One of the few reasons customers pay Apple’s premium in any market is software quality, and Apple Intelligence is now a central part of that pitch. Without it available in China, Apple was effectively selling a premium device with a feature gap.

The approval does not include a specific rollout date. CAC registration clears the legal path; Apple still has to complete technical integration, testing, and a local launch announcement.

What This Means for Business

The Apple Intelligence story is really a story about how global AI deployment works now, and it has direct implications for any business thinking about AI at scale.

Regulatory fragmentation is the baseline. The same AI feature can use different underlying models depending on which country the user is in. That is not a bug in Apple’s implementation; it is the compliance requirement. In China, you use Qwen and Baidu. In the EU, you may face different transparency and data residency rules. In the US, there is no unified framework yet. Building for global AI means building for regulatory variation.

Approved local models are becoming a procurement category. Businesses with operations in China, or planning expansion there, need to understand which AI models have CAC approval for their use case. It is no longer enough to pick a model based on benchmark scores. Procurement decisions now include regulatory status by jurisdiction.

The localized AI model approach will scale. Apple’s dual-partner setup is an early, visible example of what many enterprise deployments will look like: a product with a consistent user experience but different AI infrastructure depending on where the user is located. Data leaders who build AI workflows today should ask whether their stack can support this kind of model swapping by region, or whether they are building something that will create compliance problems as they grow internationally.

Data training and storage still matter. CAC approval requires AI providers to meet specific rules about how training data was sourced and how user data is stored. Qwen and Baidu meet those rules for China. Any business building AI tools for regulated markets will face equivalent questions. Working with a provider who has already cleared those approvals in your target market is significantly easier than trying to certify a foreign model.

For data teams and business leaders watching the AI space, the Apple-China story is worth tracking not because Apple is interesting, but because it is the clearest current example of what enterprise AI governance looks like in a market with actual enforcement. The playbook that Apple is following, partner locally, get regulatory sign-off per model, maintain consistent product experience, is the playbook that global enterprises will need to follow too.

Enterprise DNA works with teams navigating exactly this kind of complexity: how to build AI capability that holds up under real operational and compliance requirements, not just in a lab environment. If your team is assessing how to build AI into your workflows in a way that will scale without creating regulatory exposure, the Omni Advisory service is a practical starting point for that conversation.