The Zhitong Finance App learned that some media quoted information revealed by people familiar with the matter as reporting that the US consumer electronics giant Apple (AAPL.US) has specially trained a large-scale language model for the Chinese market, which means that the tech giant, which owns the world's most popular consumer electronics brands such as iPhones and iPads, has deviated from the classic strategy of relying on third-party models to support its AI functions in China. This news highlights the “two-track model system” of the Apple Intelligence (Apple Intelligence) architecture implemented by Apple around the world: its own large model is responsible for the core AI application intelligence layer that is autonomous and controllable and deeply integrated, and local cutting-edge models such as Qwen continue to undertake external capacity expansion and China's compliance ecosystem.
For Alibaba, the strategic value brought by this news may be greater than short-term direct AI revenue generation — Apple is equivalent to using one of the most demanding consumer electronics ecosystems in the world to make the world's top commercial endorsement for Qwen's big model+Alibaba Cloud's full-stack AI (“full-stack artificial intelligence”) capabilities.
For Apple, this cooperation is closer to “fixing the strategic growth gap in the Chinese market,” rather than simply adding AI models embedded in smartphones, so it is equally positive for fundamentals and valuations; the Chinese market has become the most competitive battleground for AI phones. Local manufacturers such as Huawei have complete AI application assistants, and Apple's biggest weakness before was that the Chinese iPhone lacked the same level of end-side AI experience as the global version of Apple Intelligence.
Ali helped train its own big model, and the AI on the iPhone end officially entered the “dual-track model era”
People familiar with the matter said that this AI big model (LLM) was developed by Apple in cooperation with Chinese cloud computing and e-commerce giant Alibaba, and training was completed with the technical support of the Chinese tech giant. Since the information is sensitive and has not been made public, these individuals requested anonymity. Official representatives of Apple and Alibaba did not respond to media requests for comment on the latest news.
The news that Apple is training its own AI model for the Chinese market has never been reported before. Previously, Apple preferred to use mature AI models launched by local Chinese partners to introduce generative AI or AI smart models into iPhones and other consumer electronics terminals sold in the Chinese market. In China, AI models created by cutting-edge US technology such as Anthropic's Claude and OpenAI's ChatGPT cannot be used; in the US Apple market, Apple will use these models in combination with its own AI technology.
People familiar with the matter said that Apple Intelligence — a suite of end-side AI applications launched by Apple, is expected to be launched in China in the coming months along with a major update to its iOS operating system.
Having its own AI model created with technology partners specifically for the Chinese market will allow Apple to have greater control over its AI application experience. The Chinese market has long been Apple's most competitive overseas market. Here, Apple has already been robbed of some of its market share by local competitors, including Huawei, and these rivals have quickly taken the lead in the field of smartphones equipped with AI applications.
This proposed launch plan in Apple's plan means that a large foreign technology company will use a unique AI deployment “dual-track AI model/ AI application” strategy in China, which will help Apple cope with strict regulatory barriers between China and the US; these regulatory barriers have previously restricted many other US technology companies from entering the Chinese market.
If implemented, this proposed decision would also make Apple the first foreign technology company to receive Beijing's approval and be able to provide its own proprietary AI models in China. Against the backdrop of the widening trade and diplomatic rift between China and the US surrounding the renewal, iteration and development of AI technology, this also marks a rare cooperation between the US and Chinese technology companies.
Apple has already trained an LLM for the Chinese market with the support of Alibaba, but at the same time, Qwen will still integrate Apple Intelligence into the Chinese version of the iPhone, iPad, Mac, and Vision Pro. Currently, it is unclear whether this Apple self-training model was fine-tuned and distilled from Qwen, or whether it used independent weight training.
Technically, this architecture is very much in line with Apple's own line: the 2026 third-generation Apple Foundation Models already include about 3 billion parameter AFM 3 Core and 20 billion sparse AFM 3 Core Advanced, which only activates about 1 billion to 4 billion parameters each time — although the total parameters are as high as 20 billion, the memory bandwidth and energy consumption cost during operation are similar to the 3 billion parameter model, so they can be successfully deployed on end side devices such as iPhones ; After that, a stronger server model is called through Apple's private cloud Private Cloud Compute.
For Apple, taking the Chinese-specific model into its own hands can more accurately control latency, privacy, Siri/system tool calls, personalized context, and cross-device consistency, while preventing the Chinese version of Apple Intelligence from being locked down by a third-party LLM product route for a long time.
Regulatory hurdles have been removed, but uncertainty remains
The move comes after a lengthy regulatory process. Last month, China's Internet regulator, the State Internet Information Office, completed the official registration of Apple's generative AI/AI smart device technology service, clearing the way for Apple Intelligence to land on the Chinese iPhone for the first time.
According to this arrangement, Alibaba's large Qwen model will be integrated into the Chinese market and integrated into the Apple Intelligence version provided along with the final compatible iPhone, iPad, Mac, and Vision Pro models.
Last week, Apple released a Chinese version of the AI guide explaining how eligible Mac users in mainland China can connect Qwen to their Siri AI voice digital assistant and “writing tool” function. This is a Mac-specific arrangement that may greatly help Apple's exclusive PCs compete in the Chinese AI PC market. Since then, Apple has removed this Chinese version of the AI guide without providing any explanation.
After the announcement of registration news related to Apple's AI function in July, Alibaba's shares listed and traded in the US rose by about 4% before the market, highlighting that investors agree that cooperation between the two sides may reshape China's AI application ecosystem.
Apple hopes to use cutting-edge AI technology to boost sales in the Chinese market, while Ali receives “technical endorsement” from the world's top terminal ecosystem
The partnership between Apple and Alibaba was first confirmed in February 2025. At the time, Alibaba's board chairman Joe Tsai (Joe Tsai) said at the World Government Summit in Dubai that Apple will use Alibaba's exclusive AI technology to provide AI model support for its iPhone smartphone in China.
Cai Chongxin said at the time, “They have had negotiations with many Chinese companies. In the end, they chose to work with us.” Subsequently, the launch plan was continuously delayed due to Apple's need to adjust its relevant features to meet Chinese regulatory requirements.
The Chinese market is still critical to Apple's business and performance growth, and even its stock price performance, accounting for a significant share of its total revenue. The lack of AI features in the Chinese version of the iPhone was previously thought to be one of the factors dragging down sales in Greater China, as consumers are turning to local smartphone brands that already provide built-in AI application assistants.
Currently, it is unclear what role the models trained by Apple with Alibaba's technical support will play outside of the third-party Chinese AI model, and how these models will work together.
The core factor why Alibaba has become a highly strategic partner in this system is undoubtedly that Qwen already has cutting-edge modeling capabilities, engineering capabilities, cloud infrastructure, and regulatory implementation capabilities at the same time. Qwen3.8-Max, which was just released on August 3, has 2.4 trillion total parameters, about 95 billion activation parameters, and a maximum of 1 million token context windows. Using sparse hybrid expert architecture (Sparse MoE) + mixed attention, Ali officially claims that it ranked fifth in Text Arena and second in Vision Arena, and highlights code, long-range agents, multi-modality, and real-world workload capabilities; compared to small AI startups with only model capabilities, Ali can also provide training computing power and models Studio, team of engineers, enterprise-grade cloud, and local Chinese compliance system.
These factors also enable Apple to combine “Apple's own model control layer+Ali Qwen capability layer+China's local computational power/compliance layer”: not only maintaining Apple Intelligence's product sovereignty, but also using Ali to solve China's LLM ecosystem and engineering deployment problems. More importantly, Chinese regulators completed registration for the Apple Intelligence AI service in July. This kind of local cooperation actually addresses the structural obstacle that US Claude/ChatGPT is unable to directly replicate global solutions in China.
Alibaba Cloud's quarterly revenue up to March has increased 38% year over year to 41.63 billion yuan. AI-related products account for about 30% of external cloud revenue. Management expects this ratio to exceed 50% in about a year's time, and the company is even preparing to invest more than 380 billion yuan in AI over the previous three years. Apple's collaboration may further increase the consumption of AI model training, AI inference workloads, cloud computing resources, and development ecosystems, and increase the trust of other large enterprises using Qwen. Therefore, what really matters to Alibaba's stock price and basic outlook is not “how much money does Alibaba make for every iPhone sold,” but rather that Qwen is moving from being a Chinese open source model champion to an enterprise-level AI infrastructure standard layer.
The Chinese market has become the fiercest battleground for AI smartphones. Local manufacturers such as Huawei have complete AI assistants, and Apple's biggest weakness before was that the Chinese iPhone lacked an AI service experience equivalent to the global version of Apple Intelligence; although Apple's revenue in Greater China increased sharply by 22.4% year-on-year to US$18.82 billion in the latest quarter, it is still lower than Visible Alpha's estimate of about US$19.67 billion, which is enough to show that the recovery in Chinese iPhone demand has not completely lifted competitive pressure.
Therefore, working with Ali to create its own model exclusive to China allows Apple to embed Siri, writing tools, agentic AI, and personal context deeper into the iOS operating system, and drastically reduce its strategic dependence on any single Chinese model supplier; if it ultimately brings higher iPhone switching rates, service usage rates, and user retention, its significance to Apple's stock price and fundamental expansion level is undoubtedly to reduce the “backward AI discount in the Chinese market” and extend the cash flow period for services related to Apple brand hardware+AI applications such as iPhones for a long time.