Alibaba Cloud has announced an extensive set of new AI technologies and infrastructure upgrades at its annual Apsara Conference 2025 in Hangzhou, China. The company reaffirmed its plan to invest RMB 380 billion (US$53 billion) over the next three years to strengthen AI and cloud computing capabilities, with a focus on supporting the next generation of large models and agent-based AI applications.
Eddie Wu, Chairman and CEO of Alibaba Cloud Intelligence, said the company aims to make large AI models function like operating systems for the AI era. “In the future, large AI models will be deeply integrated into a wide range of devices, functioning like operating systems — equipped with persistent memory, seamless cloud-edge coordination, and the ability to continuously evolve. We remain committed to open-sourcing Qwen and shaping it into the ‘operating system of the AI era,’ empowering developers around the world to build transformative AI applications,” he said.
Wu added that Alibaba Cloud is positioning itself as a full-stack AI service provider. “We will progress with our RMB 380 billion investment plan in AI and cloud infrastructure over the next three years,” he said.
Since launching the first generation of its Qwen model in 2023, Alibaba has released more than 300 AI models built on its Qwen large language and Wan visual generation models. These have been downloaded over 600 million times, with more than 170,000 derivative models created. Over one million companies and developers have used Qwen on Alibaba’s Model Studio platform.
Next-generation infrastructure for AI
At the event, Alibaba Cloud introduced significant upgrades to support the growth of agentic AI systems. The company enhanced its Object Storage Service (OSS) with a new “Vector Bucket” feature, which allows cost-efficient storage and retrieval of large vector datasets. This improvement aims to simplify the development of retrieval-augmented generation (RAG) applications and reduce complexity for businesses managing both raw and vector data.
The firm also presented its new high-performance network architecture, HPN8.0, designed to handle large-scale AI training and inference. With 800 Gbps network throughput, the upgrade doubles previous capacity and supports complex, mixed workloads for reinforcement learning and model deployment.
Security was another key focus. Alibaba Cloud enhanced its Cloud Threat Detection Response (CTDR) platform with AI-driven agents powered by Qwen. These agents automate threat detection, analysis, and response, improving automated incident investigation success from 59% to 74% and managing 70% of response actions without human intervention.
Improved computing and development tools
The company has also upgraded its container services, boosting auto-scaling and isolation features to handle up to 15,000 pods per minute. These improvements aim to make running large-scale, highly concurrent AI agent requests more efficient and secure.
Its PolarDB database has been optimised for data and AI workloads, introducing faster memory interconnect technology that cuts latency by 72% and increases scalability 16-fold. A new Lakebase architecture integrates support for open-data formats such as Lance, Iceberg and Apache Hudi, helping reduce storage costs and streamline multimodal data management.
On the development side, Alibaba’s Platform for AI (PAI) now includes optimisations that speed up training for its Qwen models by over 300% and reduce single-sample training time for its Wan models by 28.1%. Inference efficiency has also improved significantly, with higher throughput and lower latency, allowing faster scaling for infrastructure supporting large models.
These announcements underline Alibaba Cloud’s strategy to lead in the global AI infrastructure market while making its technology more open and accessible to developers and enterprises.
