Alibaba Cloud, the digital technology and intelligence backbone of Alibaba Group, has unveiled significant upgrades to its AI infrastructure to support agentic AI.

"To underscore our long-term commitment to advancing AI, we will progress with our RMB 380 billion investment plan in AI and cloud infrastructure over the next three years," said Eddie Wu, chairman and CEO of Alibaba Cloud Intelligence.
AI infrastructure for agentic AI
Alibaba Cloud's infrastructure upgrades include enhancing its Object Storage Service (OSS) with AI-powered "Vector Bucket," which enables cost-efficient, large-scale vector data storage and retrieval — optimised for RAG and AI apps.
Moreover, HPN8.0, a network specially designed for AI models, is Alibaba Cloud's latest architecture for a high-performance network that claims to facilitate seamless model training, inference, and reinforcement learning (RL) across mixed computational workloads, while supporting ultra-large-scale deployments.
Alibaba Cloud has also added an AI-driven agentic function to its Cloud Threat Detection Response (CTDR) solution to boost detection, analysis, and response capabilities.
For better elasticity, Alibaba Cloud has upgraded its Container Compute Services (ACS) to enhance its auto-scaling capabilities through optimised scheduling and container image cache acceleration technologies.
Upgrading its PolarDB database to optimise for combined data and AI workloads, Alibaba Cloud has introduced a hardware innovation powered by Compute Express Link (CXL) technology, a highly efficient compute-memory interconnect that reduces latency by 72.3%, boosts memory scalability by 16 times, and lays a solid foundation for data and AI workloads.
Lastly, the company's optimisation of Platform for AI (PAI) targets to advance large model development into the agentic AI era. Its novel MoE training acceleration improves Qwen series training by over 300%, while the upgraded DiT training engine reduces Wan series' single-sample training time by 28.1%.