Hangzhou, China – Alibaba has announced a series of artificial intelligence (AI) and cloud developments at its annual conference, spanning foundation models, agent platforms, data services and computing infrastructure.
The announcements include updates to the Qwen model family, a new agentic cloud architecture, enterprise tools for managing AI agents and new processors from its chip design unit, T-Head.
Alibaba said the developments are intended to help customers deploy AI at scale across business and consumer applications.
“The theme of this year’s Apsara Conference is ‘Intelligence Goes Beyond’. Over the past few years, AI has continuously expanded our imagination of technological capabilities,” said Joe Tsai, Chairman of Alibaba Group.
“AI possesses vast potential for development–it can be deployed and scaled in real-world scenarios, boosting productivity across thousands of industries. This is the true meaning of ‘Intelligence Goes Beyond’: guiding AI from technological breakthroughs toward value creation,” Tsai continued.
“Today, the total volume of Machine Thinking is less than 3% of all Human Thinking. If that volume eventually scales to 1,000x human capacity, the simple math tells us: Machine Thinking still has an enormous growth runway,” said Eddie Wu, CEO of Alibaba Group.
“As machines are becoming the primary force behind Thinking, turning intelligence into a commodity supplied at scale, the truly groundbreaking products of the Machine Intelligence era have not yet arrived. With this in mind, our target is that by 2032, the global data centre capacity operated by Alibaba Cloud will surpass 20GW, fueling the industry’s exponentially rising demand for AI,” Wu continued.
Qwen models and AI applications
Alibaba confirmed that its next-generation foundation model, Qwen 4, is in training, with Qwen 4.5 and Qwen 5 also on its roadmap. The company projects the series could scale to 5 to 10 trillion parameters.
It also reported progress in recursive self-improvement (RSI), saying Qwen3.8-Max completed 33 automated cycles of pipeline design, data validation, experimentation and error diagnosis. Alibaba said the work helped raise the model’s Artificial Analysis score from 40 to 45.
In a separate chip-design experiment, a model spent more than 60 hours self-improving and made over 10,000 electronic design automation tool calls. Alibaba said it produced production-grade chip bus modules while reducing chip area by 42% without compromising performance.
The company also announced updates to its multimodal AI portfolio, including speech, audio and image-generation models. These include Qwen3.8-LiveTranslate for simultaneous interpretation and Qwen-Audio models for speech recognition, text-to-speech and real-time voice interaction. Qwen-Image 3.1, aimed at creative design and e-commerce marketing, is scheduled to launch later this year.
For smartphones, Alibaba introduced Qwen Intelligence, an agent platform for phone makers that enables AI assistants to carry out complex tasks across applications.
Agentic cloud architecture
Alibaba Cloud outlined an agentic cloud architecture built around three areas: AI Native Cloud for model training and inference, Agent Native Cloud for enterprise agent deployment and operations, and Context Engine for supplying AI systems with real-time information and memory.
Its Platform for AI (PAI) has received optimisations across inference, caching and training. Alibaba said it completed training for a Qwen model in five days during a post-training process.
The company also upgraded its Cloud Parallel File Storage system for AI workloads, reporting throughput in the hundreds of terabytes per second and hundreds of millions of input/output operations per second. It said the changes can reduce enterprise AI storage costs by 69%.
Its HPN 8.0 Pro networking architecture is designed to deliver 100 petabits of bandwidth and support more than 130,000 800G ports in a single cluster. Alibaba said the system’s redundancy features reduce the impact of network upgrades and failures from 50% to 25% compared with the previous generation.
For enterprise agent development, Alibaba Cloud introduced AgentCore, a platform for building, operating and managing AI agents, alongside Agent Security Centre for threat detection and security and compliance controls.
The Context Engine includes Agent Context, which connects documents, business systems, chat records and multimodal data to provide agents with relevant information and long-term memory. Alibaba said it can reduce token usage by up to 67% in applications such as customer service, coding and data analytics.
Alibaba Cloud also upgraded OpenLake into a multimodal data lakehouse that brings different data types together for processing, search, analysis and model training. The company said it can reduce total costs by 38% and query response times by 40% compared with traditional architectures.
New AI computing hardware
T-Head unveiled the Zhenwu V900, a processor for AI training and inference. Alibaba said it delivers three times the performance of the previous Zhenwu M890 and features 216GB of GPU memory and 1,200GB/s of inter-chip bandwidth. Mass production and commercial release are scheduled for the first quarter of 2027.
More than 650 customers across sectors including automotive, finance, energy and manufacturing are using T-Head’s Zhenwu AI chips, according to Alibaba.
The company also introduced an upgraded supernode server that combines the V900 with its networking, storage, and controller technologies. It said the system can support clusters of up to 500,000 cards.
T-Head additionally outlined a roadmap for CPUs designed for agentic AI workloads, with launches planned for 2027. Alibaba said the Yitian 730 is designed to deliver up to 40% higher SPECint2017/GHz performance than the Yitian 710.
