The Zhitong Finance App learned that on August 10, based on the pioneering fully modular design architecture, Alibaba Cloud shortened the delivery period for large-scale AIDC data centers to 100 days, leading the world. For the same construction period, the traditional domestic delivery cycle is 6 to 12 months, and the US is 12 to 18 months. While the construction period was drastically shortened, the overall construction cost of the Alibaba Cloud data center was reduced by more than 10%. This year, Alibaba Cloud is planning to more than double the global production capacity of modular data centers.
Demand for AI computing power has exploded, and data center delivery speed requirements have increased. The traditional construction period is serial construction, and each step depends on the completion of the previous step, such as installing the power supply and distribution first, and then connecting to the air conditioner. Alibaba Cloud created a fully modular design concept. Each process is produced in parallel. After the factory is manufactured, it is pre-transferred to a container, then delivered to the data center, and then hoisted like a building block on site.
CUBE 5.0 is a next-generation architecture developed by Alibaba Cloud for AIDC. Through a fully modular design, the overall modularity rate of the five major systems of power supply, cooling, security, intelligence, and fire protection has increased from 30% in the early years to 90%. At the same time, using high-voltage DC, weather-liquid compatibility technology and ultra-high elastic design, it can maximize compatibility with mainstream heterogeneous chips and computing architectures, and support at least three generations of chip iterations.
Based on CUBE 5.0, the data center's 100-day delivery cycle begins with the basic plant or stacked containers and ends before servers and network equipment enter. Meanwhile, after 30 days of factory production and site preparation, 50 days of on-site installation, and 20 days of commissioning and testing, the five major systems of power supply, cooling, security, intelligence, and fire protection were deployed. Similar construction periods, the traditional domestic delivery cycle is 6 to 12 months, and the US is 12 to 18 months.
The fully modular data center not only shortened the construction period to 100 days, and the relevant construction costs per kilowatt were reduced by 10% compared to the previous generation. The one-time product test pass rate was close to 100%, and delivery, cost, and quality were improved at the same time, solving the “impossible triangle” in project management. Due to the significant increase in space utilization, the computing power density per unit area has reached 5-10 times that of the previous generation. The equipment is also more intensive. For example, the number of transformers has been reduced by almost half, air cooling and liquid cooling can be freely switched, air cooled PUE ≤ 1.15, and liquid cooled PUE ≤ 1.10. Furthermore, since the classification is more clear, space is reserved for later expansion.
Since the first line of Flying Sky's code was written in 2009, Alibaba Cloud has been building, operating and maintaining data centers for 17 years, with 32 regions and 105 availability zones around the world. Alibaba Cloud's data center architecture solutions have also gone through 17 years of iteration. Facing the AIDC era, Alibaba Cloud and leading manufacturers in the industry have been jointly developed over a long period of time to transform drawings into BOM (Bill of Materials, Bill of Materials) that can be mass-produced, independently design component layouts and even screw specifications, and rely on China's complete industrial chain to standardize production in factories.
Currently, the CUBE 5.0 architecture has been piloted in data centers such as Alibaba Cloud's Ulanqab and Zhongwei, and has obtained European CE certification based on EN standards and Southeast Asian product approval based on IEC standards. This year, Alibaba Cloud plans to more than double the global production capacity of modular data centers.