How the Emerging Computing Power Internet Will Transform AI and Data Services

Background

In May 2023, the China Academy of Information and Communications Technology, together with China Telecom, Mobile and Unicom, launched the "Computing Power Internet Test Network" and released the "Computing Power Internet Architecture 1.0", sparking nationwide attention. Shortly after, the Ministry of Industry and Information Technology issued the "Computing Power Interconnection Action Plan", emphasizing the construction of a unified national market for computing power services.

Definition

According to Wang Zhiqin, deputy director of the Academy, the computing power internet is not a new network but an extension of the existing Internet that connects distributed computing resources through standardized identifiers and protocol interfaces, enabling intelligent perception, real‑time discovery and on‑demand acquisition of heterogeneous computing power.

Motivation

Computing power has become a strategic resource for AI, industry and government. Rapid growth of AI models creates an exponential demand for compute, while existing resources are unevenly distributed, leading to waste and low utilization. The computing power internet aims to solve these problems by interconnecting and sharing resources.

Challenges

Building the computing power internet faces three major challenges: (1) establishing physical connectivity and sufficient bandwidth for massive data flows; (2) creating unified standards, identifiers and protocols to link heterogeneous resources; (3) ensuring security, data protection and reliable operation of the large‑scale network.

Key Tasks and Roadmap

The Action Plan outlines six areas and sixteen tasks, including standards development, technology trials, ecosystem co‑construction, and the creation of a national computing‑power trading platform. By 2026 the goal is a complete standards and rule system; by 2028 nationwide standardized interconnection with intelligent perception, real‑time discovery and on‑demand access.

Application Scenarios

Intra‑center (算内) Scenario

High‑speed interconnect within data‑center clusters, supporting AIGC model training with bandwidth up to 800 Gb/s and intelligent load‑balancing.

Inter‑center (算间) Scenario

Network linking multiple intelligent‑computing clusters across regions, requiring long‑distance, high‑bandwidth, zero‑loss transmission and coordinated scheduling.

Ingress (入算) Scenario

Connection from end‑side users to intelligent‑computing centers, handling massive “elephant‑flow” data streams with elastic bandwidth (10 Mbps–100 Gbps) and flow‑level scheduling.

Data‑Communication Role

New data‑communication technologies provide the necessary high bandwidth, low latency, intelligent scheduling, and zero‑packet‑loss capabilities to support all three scenarios, turning compute resources into on‑demand services similar to water or electricity.

Conclusion

China has already identified 131 enterprises and 499 resource pools, aggregating 111.3 EFLOPS of intelligent compute. Continued effort on the Action Plan’s tasks will accelerate the realization of a nationwide computing‑power internet, driving digital transformation across industries.