- A modular design cuts construction costs by more than 10% while speeding delivery
- The architecture targets a key bottleneck as AI drives demand for computing infrastructure
Alibaba Cloud says it has cut the delivery time for large artificial intelligence data centers to about 100 days using a fully modular design, as the race to build AI infrastructure puts pressure on both construction speed and cost.
The company said August 10 that its new architecture can deliver data centers in roughly half the time typically required in China and far faster than projects in the US, while reducing overall construction costs by more than 10%.
From construction site to factory
Traditional data centers are typically built in sequence, with power systems installed before cooling and other infrastructure can be added.
Alibaba Cloud’s approach breaks power, cooling, security, automation and fire protection into independent modules that can be manufactured and tested simultaneously in factories before being shipped to the site and assembled.
The process resembles putting together prefabricated building blocks rather than constructing a facility step by step on-site.
Alibaba Cloud Intelligence Vice President and data-center chief Liu Miao (刘淼) said the approach represents more than a change in construction methods.
“Full modularization is not simply an improvement to the construction process, but a restructuring of how data centers are produced — moving on-site construction into factories and turning sequential delivery into parallel delivery,” Liu explained.
He said the company’s CUBE 5.0 architecture is designed to make data-center delivery keep pace with the rapid growth in AI computing demand.
Under the company’s 100-day schedule, factory production and site preparation take about 30 days, installation 50 days and testing another 20 days.
Built for AI workloads
The architecture, known as CUBE 5.0, raises the modularization rate of the five major systems from about 30% in earlier designs to 90%.
It supports high-voltage direct current, both air and liquid cooling, and flexible configurations designed to accommodate different chips and computing architectures across at least three generations.
Alibaba Cloud says the design can reduce construction costs per kilowatt by more than 10%, while first-pass testing rates approach 100%.
Higher space utilization also allows computing density per unit of floor area to reach five to 10 times that of the previous generation.
Power usage effectiveness, or PUE — a standard measure of data-center energy efficiency — is as low as 1.15 for air-cooled systems and 1.10 for liquid-cooled systems.
China’s national data-center energy-efficiency standard defines PUE as a key measure of efficiency, with a lower figure indicating better performance.
The standard has been mandatory nationwide since November 2022, requiring new and expanded data centers to meet a PUE threshold of no more than 1.50.
Taking the model overseas
CUBE 5.0 has already been tested at data centers in Ulanqab and Zhongwei in China and has obtained European CE certification and market access in Southeast Asia.
Alibaba Cloud plans to more than double its global production capacity for modular data centers this year.
The company has operated data centers for more than a decade and currently has infrastructure across 32 regions and 105 availability zones worldwide.
As AI investment accelerates, the ability to build computing capacity faster and at lower cost is becoming a competitive advantage.
Alibaba Cloud’s modular approach is an attempt to turn data-center construction itself into a standardized manufacturing process — potentially giving Chinese suppliers a model they can export to a global market hungry for AI computing capacity.
