Apple trained a custom large language model for the China market with support from Alibaba, according to a Reuters report citing three people familiar with the matter. The model is designed to power Apple Intelligence on iPhones sold in China, and its regulatory approval would make Apple the first US company cleared to offer a proprietary AI model inside the country. Apple registered the on-device generative AI service with China's cyberspace regulator last month, and the rollout is expected in the coming months after an iOS update.
The partnership is a departure from how Apple has handled generative AI in China until now. In the US and other markets, Apple Intelligence hands off harder queries to OpenAI's ChatGPT, which is not available in mainland China. Rather than build alone or license a fully third-party stack, Apple trained its own model with Alibaba's help — a structure that gives Apple more direct control over the product while satisfying a regulator that requires every publicly deployed AI model to be registered and cleared before launch.
That regulatory clearance is the harder half of the story. China's cyberspace administration reviews training data, safety filters, and content behavior before a generative model can ship to consumers. Domestic firms have moved through the process for two years; foreign firms have not. Apple registering an on-device generative AI service last month is the first time a US company has cleared that bar with its own model, per Reuters.
Key facts
- 01Apple trained a China-specific large language model with support from Alibaba, according to three people familiar with the matter cited by Reuters.
- 02The move would make Apple the first US company approved to offer a proprietary AI model in China.
- 03Apple registered the on-device generative AI service with China's cyberspace regulator last month, clearing a major hurdle for device rollout.
- 04Apple Intelligence is expected to launch in China in the coming months, tied to an iOS update.
- 05Apple previously relied on domestic Chinese models because US systems like OpenAI's ChatGPT are unavailable in the market.
Alibaba's role, as described in the reporting, is that of training partner rather than model vendor. The people familiar with the matter told Reuters the China-focused LLM was developed in partnership with Alibaba and trained with the company's support. Neither Apple nor Alibaba has publicly detailed the split of intellectual property, the base architecture, or how the model compares to Alibaba's Qwen family, which powers much of the company's own consumer and enterprise AI.
For Apple, the commercial stakes in China are direct. The iPhone competes against Huawei, Xiaomi, Oppo, and Vivo, all of which have shipped generative AI features on their domestic handsets over the past 18 months. Apple Intelligence launched in English-speaking markets in late 2024 and has expanded language by language since; China has been the largest and most conspicuous gap. A device sold in Shanghai without on-device AI while a Huawei Mate sitting next to it summarizes emails and edits photos locally is a competitive problem, not a marketing one.
The Alibaba arrangement also gives Apple a domestic anchor for the political optics. Cross-border AI collaboration between US and Chinese firms has narrowed as export controls on chips and model weights have tightened in both directions. A US phone maker training a model inside China with a Chinese cloud partner is the kind of arrangement that reads as a workable middle path — Apple gets a shippable product, Beijing gets a foreign company operating inside its regulatory regime, and Alibaba gets a marquee training relationship.
There are things the reporting does not resolve. The Reuters sources are unnamed, and Apple did not comment. The size of the model, the compute footprint, whether it runs entirely on-device or splits inference with an Alibaba cloud tier, and how much of Apple's global Apple Intelligence codebase carries over are all open questions. Apple has historically shipped a small on-device model paired with a larger private-cloud tier for heavier queries; whether that same architecture holds in China, and whose cloud runs the heavier tier, is not yet public.
It is also worth noting that regulatory registration is not the same as unrestricted deployment. Chinese approval regimes have added obligations after launch — content audits, model retraining requirements, and periodic re-review. Apple's model will be subject to those, and the answers it gives on politically sensitive topics will be different from what the same underlying architecture would produce in California. Apple has navigated content restrictions in China before with the App Store and iCloud; a generative model is a harder surface to manage.
The competitive read is that Apple has moved from AI laggard in China to a first-mover on a specific regulatory question — how a US company ships its own model inside the country. That advantage is narrow and perishable; other US firms will follow if the path Apple has cut proves durable. But for now, the combination of a registered on-device model, a domestic training partner, and a coming iOS update gives Apple something to sell in its second-largest market that competitors from Redmond and Mountain View still cannot.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
