Hangzhou: TECHz Magazine – News Desk
At its annual Apsara Conference in Hangzhou, Alibaba unveiled a comprehensive full-stack artificial intelligence roadmap spanning cloud infrastructure to localized smart devices. Emphasizing the theme “Intelligence Goes Beyond,” CEO Eddie Wu projected Alibaba Cloud’s global data center capacity will surpass 20 gigawatts by 2032 to meet rising AI demand.
Alibaba announced its Qwen 4 foundation model is currently in training, with future iterations (Qwen 4.5 and 5) scaling between 5 trillion and 10 trillion parameters. Showcasing autonomous capabilities, Qwen 3.8-Max recently ran a month-long self-improvement loop across 33 cycles, raising its Artificial Analysis Intelligence Index score from 40 to 45.
In a separate chip design experiment using over 10,000 electronic design automation calls, the model reduced a production-grade bus module’s area by 42% without performance loss. Multimodal upgrades include Qwen-Audio-3.1-TTS-Next for cinematic soundscapes, Qwen-Image 3.1 for advanced image generation, the HappyOyster 2.0 Preview world model, and Qwen Intelligence, an on-device agentic solution for smartphones.
On the hardware front, Alibaba’s chip unit, T-Head, introduced the Zhenwu V900 AI processor, launching in Q1 2027. Delivering triple the performance of the M890, the V900 features 216 GB of GPU memory, 1,200 GB/s inter-chip bandwidth, and native FP8/FP4 support.
It will power a new supernode server capable of scaling to clusters of 500,000 accelerator cards. T-Head also detailed its CPU roadmap: the Yitian 720 arrives in 2027, followed by the Yitian 750 in 2029, which will feature direct interconnections with Zhenwu AI chips for seamless hardware integration.
Finally, Alibaba detailed its three-layer Agentic Cloud architecture designed to build and manage AI agents. The AI Native Cloud layer features a Cloud Parallel File Storage system delivering 100 TB/s throughput (cutting enterprise AI storage costs by 69%) and the HPN 8.0 Pro networking architecture supporting 100-petabit bandwidth and 130,000 800G ports.
The Agent Native Cloud layer introduces AgentCore and the Agent Security Center for lifecycle management, human-agent collaboration, and compliance. Rounding out the stack, the Context Engine unifies corporate and multimodal data, granting AI agents continuous learning capabilities while reducing token usage by up to 67%.
