• Alibaba’s Qwen3 models now run locally on iPhones, iPads, and Macs via Apple’s MLX framework.
  • The move positions Alibaba inside the US AI hardware stack, amplifying its global influence.

What happened:From GitHub to iPhone: Qwen3 scales Apple’s stack

Alibaba has launched its Qwen3 AI models, now fully optimised for Apple’s proprietary MLX machine learning framework. The models, ranging from compact variants to the open-weight Qwen3-72B, support native deployment across iOS and macOS devices—including iPhones, iPads, and Apple Silicon Macs.

According to Alibaba’s announcement, Qwen3 uses a hybrid reasoning architecture combining dense and MoE (Mixture-of-Experts) layers, delivering low-latency on-device performance. The models support 119 languages, are released under the Apache 2.0 licence, and are hosted openly on GitHub and ModelScope. By offering cross-device reasoning without cloud dependence, Qwen3 challenges Western AI incumbents in mobile accessibility and multilingual reach.

Also read: US probes China’s chip dominance: Tech supply risks
Also read: US sanctions force China chipmakers to focus on older-generation designs

Why it is important

This launch positions Alibaba at the centre of a rare technological handshake between Chinese AI and US hardware. Running Qwen3 locally on Apple MLX devices isn’t just a performance win—it’s a geopolitical statement. Apple, long careful in its China balancing act, now integrates a Chinese LLM into its developer ecosystem, possibly paving the way for Apple Intelligence’s rollout in China, particularly with iOS 18.6 expected soon.

The hybrid reasoning design parallels recent QwQ‑32B and Qwen3 open‑source releases that surged Alibaba’s Hong Kong share price earlier this year. Enterprises and developers now gain on-device access to powerful multilingual models without relying on cloud, reshaping competitive dynamics with U.S. giants like OpenAI and Anthropic.

However, regulatory scrutiny looms — U.S. policymakers may see this as a supply-chain breach, while Chinese authorities could pressure Alibaba to localise data flows or restrict outbound model behaviours. The global AI race is no longer just about accuracy or latency — it’s about whose models live inside whose machines. Alibaba’s integration into Apple’s MLX isn’t just technical; it symbolises a global realignment in AI ecosystems.