Alibaba announced on June 16, 2025, the launch of its latest Qwen3 AI models, which are now compatible with Apple’s MLX architecture.

According to a statement from the company on WeChat, these new AI models are designed to work on a range of Apple devices, including iPhones, iPads, MacBooks, and Macs.

The Qwen3 model was initially introduced by Alibaba Group on April 29, 2025. It features hybrid reasoning capabilities aimed at enhancing adaptability and efficiency for developers.

Qwen 3 follows Qwen 2.5-Max, which Alibaba released in January 2025 after DeepSeek’s rise.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ Strategic navigation of China’s AI regulatory landscape

The Alibaba-Apple partnership represents a strategic approach to China’s complex AI regulatory environment, where government approval is required for AI models in commercial applications.

Alibaba’s Tongyi Qianwen model has already received regulatory approval in China, making it an ideal partner for Apple to deploy AI features in the Chinese market 1.

This collaboration reflects a broader trend in the tech industry where foreign companies partner with established Chinese firms to navigate local regulations, similar to how Microsoft partnered with Chinese firms for its cloud services.

Market analysts believe this approach could help Apple recover lost ground in China, where its smartphone market share has declined amid competition from local brands like Huawei and Xiaomi that have advanced their AI capabilities 1.

This partnership illustrates how global tech companies must adapt their AI strategies to accommodate regional regulatory requirements, with data localization laws in China requiring different approaches than those used in Western markets 2.

2️⃣ On-device AI capabilities signal shifting competitive dynamics

Alibaba’s deployment of Qwen3 models for Apple’s MLX architecture highlights the growing importance of on-device AI processing as a competitive differentiator in the tech ecosystem.

Alibaba has invested in AI hardware capabilities, demonstrated by their Hanguang 800 chip which can process 78,563 images per second 3.

This partnership addresses a key challenge for Apple, as recent evaluations have shown Apple’s on-device AI models struggling to compete with cloud-based alternatives from competitors like OpenAI and Google 4.