Alibaba Cloud is expanding its full-stack AI strategy with new foundation models, multimodal AI capabilities, proprietary chips, agentic cloud infrastructure, and tools designed to bring AI deeper into everyday business and consumer applications.
The updates were announced at this year’s Apsara Conference, under the theme “Intelligence Goes Beyond.” Alibaba Group Chairman Joe Tsai said the company’s focus is on moving AI beyond technological breakthroughs and toward practical value creation across industries.
Alibaba Group CEO Eddie Wu also outlined the company’s long-term infrastructure ambitions, targeting more than 20GW of global data center capacity operated by Alibaba Cloud by 2032 as demand for AI computing continues to grow.
Qwen Models Push AI Development Forward
Alibaba revealed that its next-generation Qwen 4 model is currently in training, alongside a roadmap for Qwen 4.5 and Qwen 5, with future models projected to scale to between 5 and 10 trillion parameters.
The company also highlighted progress in Recursive Self-Improvement (RSI), where AI models use feedback from their own performance to improve training and experimentation. Alibaba said an updated Qwen3.8-Max model improved its Artificial Analysis score from 40 to 45 after more than a month of automated development cycles.
The company also showcased a chip-design experiment in which a model conducted more than 10,000 electronic design automation tool calls over 60 hours, resulting in a reported 42% reduction in chip area without compromising performance.
Alibaba is also expanding its multimodal AI lineup with models for real-time translation, speech recognition, text-to-speech, audio generation, and image creation. Qwen3.8-LiveTranslate is designed for simultaneous interpretation, while Qwen-Audio-3.1-TTS-Next can generate cinematic soundscapes combining dialogue and ambient audio from text.
Meanwhile, Qwen-Image 3.1 is set to launch later this year with tools aimed at creative design and e-commerce marketing, including transparent-background generation and image editing.
New Chips And An Agentic Cloud
Alibaba’s chip design unit T-Head also introduced the Zhenwu V900, an AI processor designed for training and inference. The company says it delivers three times the performance of its predecessor, with 216GB of GPU memory and 1,200GB/s of inter-chip bandwidth. Commercial release is scheduled for Q1 2027.
T-Head also announced its next-generation Yitian 720 and Yitian 730 CPUs, designed with agentic AI workloads in mind.
On the cloud side, Alibaba introduced upgrades built around three areas: AI Native Cloud, Agent Native Cloud, and Context Engine. These are designed to support large-scale AI model development, enterprise AI agents, and the data and memory required for agents to operate effectively.
Among the new offerings is AgentCore, an enterprise platform for building, managing, and monitoring AI agents, alongside the Agent Security Center for security and compliance. Agent Context provides AI agents with access to real-time information and long-term memory, while Alibaba’s upgraded OpenLake brings different types of enterprise data into a unified data lakehouse.
Alibaba says these developments are part of a broader strategy to make AI more scalable, secure, and useful across business operations—from application development and customer service to data analytics, automation, and AI-powered devices.
