Chinese AI giants pivot toward proprietary models to drive revenue, performance

The pivot

It has been reported that Chinese AI firms including Alibaba Cloud (阿里云) and Zhipu AI (智谱AI) are choosing not to open‑source some of their newest large models, instead serving them exclusively through official cloud platforms and chatbot sites. The move marks a visible shift from the more open posture seen in parts of China’s AI community, though companies say they are not abandoning open‑source strategies entirely. Who benefits from keeping the biggest models behind paywalls — cloud operators, enterprise customers or the firms themselves?

Alibaba’s new releases

This week Alibaba Cloud released three proprietary Qwen models that are accessible only via its cloud or chatbot interface — notably Qwen3.6‑Plus, billed as having enhanced coding capabilities, and Qwen3.5‑Omni, a multimodal engine that reportedly processes text, audio, images and video. The previous generation, Qwen3‑Omni, was open‑sourced. According to an Alibaba Cloud spokesperson, the latest Omni incarnation will not be open‑sourced because the Omni series was less popular among developers based on download figures on open‑source platform Hugging Face. Alibaba Cloud is the AI and cloud unit of Alibaba Group Holding (阿里巴巴集团), the owner of the South China Morning Post.

Why now

There are practical and commercial reasons for the turn. Flagship models are ballooning in size and compute needs, making local hosting difficult and costly; serving them from centralized cloud infrastructure makes performance and monetization easier to control. Geopolitics also matters: export controls on advanced chips and rising regulatory scrutiny have pushed firms to centralize deployment and manage distribution of powerful models. Reportedly, companies are keen to “capture the full value” of model usage via subscription, API fees and cloud services rather than rely solely on open‑source diffusion.

Consequences

For developers and the open‑source community the change narrows access to the most capable systems and could accelerate a two‑tier ecosystem: smaller, open models for experimentation and large, proprietary models for production and revenue. For Western observers, the shift underscores how commercial incentives, technical constraints and geopolitical pressure are reshaping AI development in China — and why control over model access is becoming as strategically important as model capability.