Alibaba AI

Alibaba AI refers to the artificial intelligence research, infrastructure and applications of Alibaba Group, the Chinese e-commerce and cloud computing group headquartered in Hangzhou. Its best-known contribution is the Qwen (Tongyi Qianwen) family of large language models, which by 2026 had recorded more than three billion cumulative downloads and over 300,000 derivative models, making it one of the most widely adopted open-weight model series in the world.^[1][5]

Alibaba has pursued what it calls a full-stack AI strategy — combining custom silicon, cloud data centres, foundation models and consumer and enterprise applications — and grew its AI-related cloud business rapidly during 2025 and 2026, with AI product revenue sustaining triple-digit year-on-year growth for twelve consecutive quarters by mid-2026.^[1]

History

Alibaba's AI research efforts were institutionalised with the founding of the DAMO Academy in 2017, a global research institute that became the company's principal vehicle for work in machine learning, speech, vision and natural language processing. The company also established a semiconductor design unit, T-Head, whose chips would later be integrated into its data centre and AI stack.^[1]

The Qwen programme began in April 2023 with the release of Tongyi Qianwen, a large language model that Alibaba quickly supplemented with successive open-weight releases, from the Qwen2 series in 2024 through Qwen2.5, which became one of the most downloaded model families on the Hugging Face platform. In February 2025 Alibaba announced a three-year investment plan of more than RMB 380 billion, then worth about US$53 billion, for cloud and AI infrastructure, and reportedly deepened its partnership with Apple to supply Qwen models for on-device AI features in China.^[1][5]

The release cadence accelerated through 2025 and 2026: the Qwen3 series and the Qwen app for consumers launched in 2025, the open-weight Qwen3.5 followed in February 2026 with a focus on agentic tasks, and in August 2026 Qwen released the multimodal Qwen3.8-Flash alongside open-source weights for a companion model, Qwen3.8-Flash-Next, which the company described as a prototype architecture for its next-generation Qwen4 family.^[2][4]

By its August 2026 quarterly results, Alibaba reported that AI Cloud and Compute Services revenue had reached US$7.1 billion, up 45 per cent year-on-year and the fastest cloud growth in twenty-two quarters, while AI-related product revenue reached US$1.8 billion. Capital expenditure for the quarter rose 75 per cent year-on-year to nearly US$10 billion, and days earlier the group had announced a heavily discounted HK$80 billion share sale to fund its AI ambitions.^[1][2]

Key Technologies

The Qwen model family spans large language models, vision-language models, video generation models such as Wan, coding models and audio systems. Later generations introduced hybrid "thinking" and non-thinking modes that let developers trade reasoning depth for latency and cost, long context windows of 262,144 tokens expandable to one million tokens, and multimodal inputs. Qwen models are published in dense and mixture-of-experts variants at scales ranging from small edge models to frontier systems, many under permissive open licences that allow download, modification and self-hosting.^[2][5]

Alibaba Cloud provides the commercial layer of the stack, offering model-as-a-service through its Model Studio platform, dedicated AI compute, and by mid-2026 a global footprint of 104 availability zones across 32 regions. The group's T-Head unit designs AI processors such as the Zhenwu M890, which the company says had been adopted through Alibaba Cloud services by more than 650 external customers across 20-plus industries. In 2026 Alibaba also began rolling out an agentic services suite, including the AgentRun agent development platform and security products for enterprise AI deployments.^[1][3]

At the application layer, Alibaba consolidated offerings into products including QwenWork, an AI-native workplace agent, and the Qwen app, which introduced paid subscriptions for productivity features in August 2026, while a Qwen shopping assistant was embedded in the Taobao e-commerce app to guide buyers from product discovery to after-sales service.^[1]

Applications and Impact

Qwen models are used for enterprise knowledge management, coding assistance, document processing, customer service and public-sector digitalisation, and are distributed both through Alibaba Cloud and through open-weight downloads that make the family a common foundation for derivative models worldwide. Preview versions of the Qwen3.7 series ranked among the highest-placed Chinese models on independent evaluation leaderboards in 2026, and analysts have described Alibaba as one of the most direct ways for public-market investors to gain exposure to China's AI sector.^[1][6]

The open-weight strategy has been central to Qwen's influence: releasing capable models freely has built developer mindshare that Alibaba then converts into cloud consumption, an approach broadly similar to that of Meta with Llama and Mistral AI in the West. The company's heavy capital expenditure, share placement and rapid model releases have also made it a central participant in the global AI investment race, alongside American hyperscalers and Chinese rivals such as DeepSeek.^[1][2]

>See Also

• Qwen
• Wan
• DeepSeek
• Zhipu AI
• Moonshot AI
• AI Data Centres Malaysia