Alibaba Group Shifts AI Model Strategy, Releases Third Proprietary Model

Alibaba Group has broken with China's dominant open-source AI strategy by releasing its third proprietary model, signaling potential retreat from the approach that fueled the nation's rapid AI ascent. The company's Qwen model family previously represented a cornerstone of this ecosystem, capturing more than half of worldwide open-source downloads and serving as the preferred base layer for researchers building fine-tuned derivatives. This shift comes as the enormous costs of training and running AI models—estimated at billions for top-tier systems—create financial pressure despite the prestige and strategic advantages of open-sourcing.

China's embrace of open-source AI has created a powerful innovation flywheel, with labs publishing model weights that let others iterate and distill them into new products. This approach has made low-cost Chinese AI not just dominant in Global South markets but increasingly prevalent in Silicon Valley, where an Andreessen Horowitz partner estimates 80% of companies pitching with open-source AI use Chinese models. The strategy has also served as a talent pipeline, with GitHub remaining accessible behind China's firewall and the country producing approximately half of the world's AI researchers.

The tension between open and proprietary models presents a fundamental challenge for the global AI landscape. While open models typically trail proprietary ones by about six months, this gap has remained surprisingly narrow, raising questions about how companies like OpenAI justify their massive valuations when Chinese rivals offer nearly comparable technology for free. Beijing has officially blessed the open-source approach as central to its tech ambitions, making a full-scale abandonment unlikely, but Alibaba's move suggests Chinese labs may increasingly adopt hybrid models that mix open and proprietary releases.