Tech & Inventions | Artificial Intelligence

Apple’s China AI partner has had a rough ride

7 November 2026 4 min read هدهد تك | Hudhud Tech

On November 4, Apple Intelligence commenced beta testing in China, incorporating Alibaba’s Qwen into selected Siri, Writing Tools, and Visual Intelligence experiences. This partnership highlights the rapid rise of Qwen, which was launched in 2023. According to Hugging Face, Qwen achieved over 2 billion downloads on its Hub within the first seven months of 2026, surpassing competitors like Google and Meta. However, Qwen’s journey has not been without challenges, facing internal skepticism, intense competition, and significant personnel changes.

Initial Challenges in the Market

Qwen’s development did not follow a straightforward path to success. Alibaba introduced Tongyi Qianwen in April 2023, shortly after the launch of ChatGPT, which sparked a race in China to develop large language models. The company began releasing model weights publicly, making open source a central aspect of Qwen’s strategy.

Internally, Qwen was not the preferred choice within Alibaba. Some teams continued to use Meta’s Llama into 2024, while others experimented with DeepSeek’s R1 model. Qwen needed to gain acceptance within its own company before it could compete in the broader market.

Open source contributions became crucial for Qwen’s development. Developers fine-tuned the model and created derivatives, which led to improvements that were incorporated into Qwen2.5, enhancing its capabilities in language, coding, and mathematics.

By late 2024, Qwen2.5 began to attract developers in both China and the U.S., as well as more teams within Alibaba. The introduction of DeepSeek’s R1 model raised the competitive stakes, pushing rivals to enhance their offerings.

Alibaba founder Jack Ma frequently sought updates on Qwen’s progress. During the final development phase, employees worked long hours, with some even sleeping at their desks to meet deadlines.

In April 2025, Alibaba released Qwen3, which introduced eight open models and hybrid reasoning capabilities, allowing for a balance between complex task processing and faster responses for simpler queries. The model utilized a mixture-of-experts architecture, activating only a portion of its parameters to reduce inference costs. Reuters described this launch as part of an escalating AI competition in China.

Internal Team Dynamics and Leadership Changes

As Qwen expanded its model family and developer outreach, internal pressures began to mount. In March 2026, technical leader Junyang Lin announced his departure, stating simply, “Bye my beloved Qwen.” This was followed by the exit of post-training chief Bowen Yu and senior researcher Binyuan Hui.

Alibaba did not disclose specific reasons for Lin’s departure, but reports indicated tensions related to proposed organizational changes, team boundaries, and resource allocation.

During an emergency meeting, researchers expressed frustration over the disparity in computing resources, questioning why paying customers of Alibaba Cloud received priority over Qwen’s own team.

Management acknowledged communication issues but emphasized that the changes were intended to strengthen Qwen. This situation highlighted a broader challenge faced by leading AI labs: the need for centralized resources while maintaining the autonomy of research teams.

In response, Alibaba tightened coordination efforts and established the Token Foundry in June, consolidating key development teams under CEO Eddie Wu. By September, Dayiheng Liu, a long-time Qwen researcher, was appointed head of the Qwen large language model project, providing clarity after several rounds of reorganization.

Expansion Beyond Model Development

Qwen has evolved from a model family into a comprehensive suite of offerings, including foundation models for language, coding, and multimodal tasks, as well as Alibaba Cloud’s Model Studio for developers and consumer products like the Qwen App.

This breadth of capabilities was significant for Apple, which evaluated various Chinese AI providers, including Baidu, Tencent, ByteDance, Alibaba, and DeepSeek, before selecting Alibaba as its partner for Apple Intelligence in China.

By August 2026, Apple had trained a China-specific model with Alibaba’s assistance, integrating Qwen into its approved setup. Alibaba is supporting this initiative at a group level, having pledged at least 380 billion yuan ($53 billion) over three years for AI and cloud infrastructure, significantly more than its previous decade’s investment in these areas.

Alibaba’s broader AI investment strategy encompasses foundation models, AI-native applications, and the integration of AI across existing businesses. By May 2026, the company announced that its spending would exceed the initial 380 billion yuan plan.

The partnership with Apple represents a new phase for Qwen, which now faces the challenge of maintaining its agility and innovation as its operational scale expands.