Alibaba unveils new Qwen model, custom chips in bid to become China’s AI factory

Alibaba Group Holding unveiled a sweeping suite of new artificial intelligence products spanning models, cloud infrastructure and chips on Wednesday, positioning itself to become China’s “AI factory” as autonomous agents gain prominence. Speaking at the Alibaba Cloud Summit in Hangzhou, Liu Weiguang, senior vice-president of Alibaba’s cloud computing unit, framed AI as a new form of manufacturing that generates revenue through “training and inference factories”. “What we’re building is China’s AI factory,” Liu said, describing Alibaba as the only AI and cloud company in the country operating “all five layers of the full AI stack”, which bridges chips, agentic cloud, AI models, model service platforms and agentic applications. Alibaba owns the South China Morning Post. During the event, Alibaba announced Qwen3.7-Max, the company’s latest proprietary large language model designed as “a robust foundation” for AI agents. According to Alibaba, the model excels at agentic coding, complex reasoning and “long-horizon tasks” – complex, multi-step missions requiring continuous decision-making over extended periods. The company said Qwen3.7-Max could autonomously operate for up to 35 hours without performance degradation. Zhou Jingren, the former chief technology officer of Alibaba Cloud who was recently appointed as the chief AI architect of the newly formed Alibaba Group Technology Committee, said the model consistently ranked among the top tier on various benchmarks and outperformed all other AI models in China. While preview versions teased on Tuesday showed Qwen3.7-Max overtaking domestic rivals in text and vision capabilities, it still trails top US models from companies like Anthropic, Google and OpenAI. Alibaba made no mention of open-sourcing smaller versions of its new flagship model, as the broader Chinese industry moves away from fully open-sourcing their most powerful models amid intensifying commercialisation pressure. While the company has kept its Qwen3.6-Max and Qwen3.6-Plus models proprietary, it did open-source Qwen3.6-27B – a smaller model Alibaba described as “dense” and one that “packed flagship-level coding power”. On Wednesday, Alibaba also introduced the Zhenwu M890, an AI training and inference chip. The company said the hardware delivered three times the performance of its predecessor, the Zhenwu 810E, making it “exceptionally well-suited for complex agentic AI workloads”. In January, Alibaba’s chip design arm, T-Head, noted that the previous Zhenwu 810E performed on par with Nvidia’s H20 – a graphics processing unit the US chip giant tailored for the Chinese market to comply with US export restrictions. As of April, Alibaba had shipped 560,000 Zhenwu chips to more than 400 clients across 20 industries, including carmakers and financial services, according to T-Head vice-president Gao Hui. Gao said T-Head planned to launch the Zhenwu V900 processor in the third quarter of 2027, followed by the Zhenwu J900 in the third quarter of 2028. Alibaba also launched the Panjiu AL128 Supernode Server equipped with the new Zhenwu M890 chip. First debuted last year with an older chip, the server infrastructure has “dramatically improved the handling of large-scale concurrent requests from agents”, according to the company.