Alibaba presented a broad expansion of its artificial intelligence infrastructure that combines new Qwen models, its own accelerators, supernodes and the international growth of Alibaba Cloud. The company has set a goal of surpassing 20 GW of global data center capacity by 2032, according to the roadmap announced at the Apsara Conference this week.
The plan places different layers of AI infrastructure under the same strategy. In addition to physical computing capacity, the company is developing chips, networks, storage, models and services for AI agents within its own cloud platform.
The next generation of the company's main model, Qwen 4, is already in training. Alibaba also stated that the Qwen 4.5 and Qwen 5 series could reach a scale of 5 trillion to 10 trillion parameters, although it has not yet disclosed a launch schedule for these versions.
Zhenwu V900 arrives in 2027
In the hardware layer, T-Head, Alibaba's semiconductor unit, introduced the Zhenwu V900, an accelerator intended for training and inference of AI models. According to the company, the chip offers three times the performance of the Zhenwu M890 and has 216 GB of memory and 1,200 GB/s of inter-chip bandwidth.
Mass production and commercial launch are scheduled for the first quarter of 2027. Alibaba states that Zhenwu chips already serve more than 650 customers in sectors such as automotive, finance, energy and manufacturing.
The company also introduced a new supernode architecture that integrates the Zhenwu V900 with its own networking and storage components. The project was designed to support clusters of up to 500,000 accelerators, expanding the possible scale of training and inference within the company's infrastructure.
Alibaba Cloud expands international presence
The physical expansion accompanies investment in chips and models. Alibaba Cloud plans to open its first cloud regions in Turkey, Finland and the Netherlands over the next 12 months, in addition to expanding its capacity in Malaysia, Germany, the United Arab Emirates, France and Hong Kong. Currently, the company operates 107 availability zones in 31 regions.
The roadmap also includes infrastructure specific to AI agents, with services for training and inference of models, execution of enterprise agents and management of context and memory. The strategy brings together the Qwen models, its own chips, data center infrastructure and cloud services on the same platform.
The goal of more than 20 GW by 2032 gives a physical dimension to this expansion. Instead of concentrating investment only on models, Alibaba is simultaneously increasing computing capacity and control over the different layers needed to develop and operate AI systems at large scale.
