Alibaba Cloud Reduces Data Center Assembly Time by 50% Using Modular Architecture

Alibaba Cloud presented at its annual Apsara conference a modular data center architecture called “CUBE DC 5.0”, which, according to the company, will help it meet the demand for AI infrastructure by increasing productivity and reducing the time to build new objects, writes The Register.

According to Alibaba Cloud, the CUBE DC 5.0 architecture is based on “ready-made modular designs” and “advanced proprietary technologies such as hybrid wind-liquid cooling, DC power distribution architecture and intelligent control system.”

Image source: Alibaba Cloud

Without disclosing details of the technologies used, Alibaba Cloud said the modular approach reduces data center deployment time by up to 50% compared to traditional construction methods.

The Register asked Alibaba to explain the “wind-liquid hybrid cooling system.” Presumably, we are talking about a cold plate liquid cooling system – a method in which thin reservoirs of cooled liquid are placed on the equipment, and cooling is achieved by circulating the liquid and/or blowing air across the plates.

Alibaba Cloud Intelligence CEO Eddie Wu said the company is “investing heavily in building the AI ​​infrastructure of the future.” The company announced at the conference a scheduler that improves the management of hardware resources, increasing utilization up to 90%.

Also unveiled was the ninth-generation Enterprise Elastic Compute Service (ECS) IaaS offering, which received improvements to handle AI applications. In particular, the speed of the recommendation engine was increased by 30% and the number of requests to read/write to the database per second by 17%.

In addition, Alibaba Cloud announced the Open Lake data utility, which brings together several large data engines for use by generative AI applications. Another new offering from the Chinese company is DMS: OneMeta+OneOps, which apparently aggregates and manages metadata from 40 different data sources.

Alibaba Cloud also announced the release of new open-source AI models of the Qwen 2.5 family and a new AI model of the Tongyi Wanxiang family for converting text to video with support for Chinese and English.