What happened
At the 2026 Yunqi Conference, Alibaba made a dense series of announcements spanning AI models, AI chips, AI cloud and AI hardware. The company said its future models will be expanded to a 5-10T parameter scale, and it introduced what it described as China's strongest self-developed AI chip, the Zhenwu V900.
Alibaba also set a target for the scale of global data centers operated by Alibaba Cloud to exceed 20GW by 2032, and launched the AI hardware QwenNote. Group CEO Wu Yongming said in his keynote that AI models, AI chips and AI cloud are the three cornerstones of the machine intelligence era, and that Alibaba will firmly invest in these three infrastructure areas as a long-term strategic choice.
On models, Alibaba presented exploration of recursive self-improvement (RSI), which lets a model identify its own weaknesses from real feedback, design experiments, build data and evaluate results to drive its own evolution. Its Qwen team is pursuing RSI research and plans to train a new model at the 5-10T parameter scale, aiming at more complex, longer-horizon tasks on the path toward ASI. Wu also said integrated multimodal models that combine understanding and generation are another key research direction.
On chips, Pingtouge released the new training-and-inference AI chip Zhenwu V900, which it said has three times the performance of Zhenwu M890 and can meet the training and inference needs of trillion-parameter large models. The chip is based on a self-developed parallel computing architecture, carries 216GB of memory and has 1200GB/s inter-chip interconnect bandwidth, and will go on sale in the first quarter of 2027. Pingtouge also disclosed plans for its Yitian server CPUs, with the Yitian 720 and Yitian 730 to arrive in 2027.
On cloud, Wu said building a global, efficient super network requires joint optimization of GPU, CPU, network, storage, databases and various tools, and that Alibaba is investing fully in AI infrastructure to build an AI cloud for the machine intelligence era, targeting more than 20GW of global data center capacity operated by Alibaba Cloud by 2032.
On devices, Qianwen Office released the AI hardware QwenNote A2 at 1,199 yuan, on sale directly, while the full-stack AI phone solution Qwen Intelligence has also landed, with the Honor Magic9 series to debut it next week.
Why it matters
Alibaba is framing AI not as a single product line but as an industry that requires building plants and generating power. Wu compared AI models, chips and cloud to the three cornerstones of the machine intelligence era, and argued that machine intelligence will eventually be supplied at scale like electricity. That framing suggests the company expects value to accrue to whoever builds the underlying capacity, not only to whoever ships the most visible applications.
The organizational moves described alongside the announcements point in the same direction. Alibaba has repeatedly adjusted its structure this year, including forming the Alibaba Token Hub business group, a group technology committee chaired by Wu, a Token Foundry unit, and merging Cloud Intelligence Group with Pingtouge into an AI cloud and computing services segment while consolidating model and application businesses. The Yunqi keynote lineup, running from model to chip to cloud speakers, mirrors that structure.
Read together, the announcements and the reorganization suggest Alibaba is betting on infrastructure breadth as its differentiator. Whether that bet pays off depends on whether demand for machine intelligence grows as fast as the company assumes, and whether it can keep all three layers advancing at once.
Key facts
Alibaba said future models will be expanded to a 5-10T parameter scale.
Pingtouge released the Zhenwu V900 AI chip, described as having three times the performance of Zhenwu M890, with 216GB memory and 1200GB/s inter-chip bandwidth, going on sale in Q1 2027.
Alibaba targets more than 20GW of global data center capacity operated by Alibaba Cloud by 2032.
Qianwen Office released the AI hardware QwenNote A2 at 1,199 yuan, and the Qwen Intelligence AI phone solution will debut on the Honor Magic9 series next week.
CEO Wu Yongming said AI models, AI chips and AI cloud are the three cornerstones of the machine intelligence era and that Alibaba will firmly invest in them.
What to watch next
The Zhenwu V900 is slated for mass production and sale in the first quarter of 2027, and the Yitian 720 and 730 server CPUs are planned for 2027, so execution against those timelines will be a key test of the chip roadmap.
Alibaba's 20GW data center target extends to 2032, a long horizon that will require sustained capital spending; how the company paces that investment against AI demand will shape the cloud business.
The Qwen team's RSI work and its planned 5-10T parameter model are research bets. Progress on whether models can meaningfully self-improve, and on integrated multimodal capabilities, will indicate how quickly the model layer advances.
