Next‑Generation Cloud Infrastructure: Alibaba Cloud’s High‑Performance “Cloud Infrastructure As A Computer”

Today, data centers have entered the cloud computing era, and under pressures such as digital transformation, East‑to‑West data migration, and carbon‑neutral goals, the definition of next‑generation cloud infrastructure is needed.

At the 2022 Yunqi Conference technical forum, Alibaba Group Vice President and Alibaba Cloud Infrastructure Head Zhou Ming answered that the next‑generation infrastructure is a high‑performance “Cloud Infrastructure As A Computer”.

Alibaba Cloud now operates in 28 countries and regions across four continents, with 85 availability zones, over 2,800 network nodes, and more than one million servers, fundamentally reshaping IT hardware, data centers, chips, and the entire industry chain. The evolution moves from single machines to clusters to data centers, culminating in “cloud infrastructure as a computer” that powers diverse cloud services.

Definition: Cloud Infrastructure As A Computer – a high‑performance infrastructure.

Features:

Integrated Architecture Design: A unified stack comprising high‑performance data centers, servers, networks, and intelligent operation platforms, with the PredFabric low‑latency, high‑availability network and modular Cube DC design optimizing overall utilization.

High Performance: From single‑server performance to cluster‑wide and data‑center‑wide performance, the infrastructure is defined as “high‑performance”.

System‑Level Integration: Combines custom servers, switches, high‑performance NICs with software platforms such as the intelligent operation brain and heterogeneous computing platforms, delivering elastic, native heterogeneous compute, predictable networking, and intelligent operations.

High‑Performance Network PredFabric: Reduces tail latency by 90%, achieves sub‑2 µs latency, and scales linearly with bandwidth up to 1.6 Tbps using the self‑developed Solar‑RDMA protocol, greatly benefiting AI workloads.

High‑Performance Computing Capabilities: Supports heterogeneous computing, resource pooling, and acceleration engines. Technologies like Zhenduan heterogeneous compilation, vODLA native compute, and ACCL communication eliminate network congestion and boost distributed training efficiency.

The infrastructure underpins elastic cloud resources, network and traffic elasticity, and large‑scale pooling through memory expansion, sharing, and pooling, expanding hardware resource pools tenfold.

Green computing is a core focus: Alibaba Cloud employs single‑phase immersion cooling (PUE 1.09), reduces server power consumption by 10%, and sources clean energy, achieving a 77.4 kt CO₂ reduction and over 50% clean‑energy usage across its five major data centers.

Overall, the next‑generation “Cloud Infrastructure As A Computer” provides a high‑performance, sustainable, and cost‑effective foundation for cloud services, driving digital transformation across industries.