Home · Technology · Sep 24 archive
Alibaba Unveils Full-Stack AI Strategy With New Qwen Models and Chips
Confirmed
In Short: At the Apsara Conference, Alibaba Cloud unveiled a full-stack artificial intelligence (AI) strategy, encompassing new Qwen models, proprietary chips, and advanced cloud infrastructure.
The company's cloud computing arm introduced the Zhenwu V900 AI training and inference processor, developed by its chip design unit T-Head. This processor is designed to deliver three times the performance of the Zhenwu M890, with 216 GB of GPU memory and 1,200 GB/s of inter-chip bandwidth.
Alibaba also announced a roadmap for its next-generation Yitian 720 and Yitian 730 CPUs, scheduled for launch in 2027. These CPUs are designed for agentic AI tasks and offer improved single-core performance, core density, and energy efficiency.
The company's full-stack AI strategy includes the Qwen Intelligence platform, a full-stack agentic solution for smartphones that enables device manufacturers to build next-generation AI smartphones capable of handling complex tasks across multiple applications.
Alibaba is targeting over 20GW of global data center capacity by 2032, driven by the escalating demand from AI workloads. This expansion is part of a comprehensive full-stack AI strategy, integrating proprietary chips, cloud infrastructure, and models.
Alibaba's next-generation Qwen 4 model is currently in training, with plans to scale to between 5 trillion and 10 trillion parameters. The Qwen 4 model conducted more than 60 hours of self-improvement across the design lifecycle and made more than 10,000 electronic design automation tool calls.
The company also announced updates to its speech recognition, text-to-speech, and real-time interaction models, alongside Qwen-Image 3.1 for creative design and e-commerce applications.
Alibaba's Qwen3.8-Max autonomously carried out an entire process spanning pipeline design, data validation, iterative experiments, and error diagnosis over a month, completing 33 improvement cycles.
The Qwen3.8-LiveTranslate simultaneous interpretation model reduced latency by about 20% to 2.3 seconds from 2.8 seconds, enabling more natural and responsive real-time interpretation.
Alibaba also announced an AI-powered music generation model, HappyShrimp 1.1, and updates to multimodal models spanning speech, audio, vision, and world models.
The company's AgentCore platform, part of the Agent Native Cloud, enables enterprises to build, operate, and manage AI agents throughout their lifecycle.
Alibaba said its next-generation Qwen models can reduce token usage by up to 67% in knowledge-intensive applications such as customer service, AI coding, and data analytics.
What this adds
Alibaba's full-stack AI strategy aims to expand the use of AI across various industries, including automotive, finance, large language models, embodied intelligence, energy, and manufacturing.
What's confirmed
- In a chip design experiment, Alibaba said the model conducted more than 60 hours of self-improvement across the design lifecycle and made more than 10,000 electronic design automation tool calls.
What's still developing
- ETDatacenters Published On Sep 24, 2026 at 07:55 AM IST Select ETDatacenters as Preferred Choice on Google Share Icon Copy Link Share on Whatsapp Share on Linkedin Share on X Share on Telegram Share on Facebook Bookmark Comment Highlights Alibaba Cloud targets over 20GW global data center capacity by 2032, driven by AI demand.
- The announcements were made at the Apsara Conference, Alibaba Cloud’s annual technology event.
- The Qwen-Audio-3.1-TTS-Next audio generation model can combine dialogue and ambient sounds from a text script in a single process to generate rich, cinematic audio, Alibaba said.