Why Computing Power Networks Are the Next Backbone of Digital Infrastructure

What Is Computing Power?

Computing power, or "算力", refers to the processing capability of devices ranging from smartphones and PCs to supercomputers. It is the foundation for all software and hardware applications, with higher CPU, GPU, and memory configurations generally delivering greater computing power.

Key Characteristics of a Computing Power Network

The network must provide a high‑quality connection featuring ultra‑large bandwidth, ultra‑low latency, massive connectivity, and multi‑service support. Six essential characteristics are:

Elasticity : Bandwidth and duration can be adjusted on demand, such as providing a custom‑duration high‑bandwidth link for weather‑center calculations.

Agility : Users should obtain compute resources without worrying about underlying resource distribution, ensuring rapid and seamless access.

Lossless : Even a 0.1% packet loss can cause a 50% loss of computing efficiency, so lossless transmission within data centers is critical.

Security : Data must be securely delivered to compute nodes and back, covering storage encryption, tenant isolation, and protection against attacks.

Perception : The network must sense application requirements and provide differentiated SLA guarantees, as well as performance monitoring.

Visibility : A digital network map should model the relationship between applications, compute, and network, enabling transparent topology, fault tracing, and intelligent routing.

Measuring Computing Power

Common metrics include MIPS, DMIPS, OPS, FLOPS, and Hash/s. For example, a typical PC operates at the GFLOPS level, while China’s Sunway TaihuLight reaches 93.015 PFLOPS. In cryptocurrency mining, hash rates (e.g., GHash/s for PCs, 200 EHash/s for the Bitcoin network) indicate the probability of successful mining.

Classification of Computing Power

Computing power can be divided into two broad categories:

General computing power – small workloads and routine applications.

High‑Performance Computing (HPC) – large‑scale workloads that require clusters of tightly coupled servers. HPC can be further split into scientific, engineering, and AI computing.

With the rise of AI, computing power is often categorized as either general or AI‑oriented.

Why a Dedicated Network Is Needed

Data‑driven digital economies generate massive data that must be processed across cloud, edge, and end devices. A computing power network allocates and schedules compute, storage, and network resources on demand, improving efficiency for intelligent‑world applications that rely on data + computing + algorithms.

Cloud‑Edge‑End Architecture

The architecture evolves from a centralized cloud to a pervasive three‑tier model:

Cloud : Centralized data‑center resources offering public, private, or hybrid clouds.

Edge : Distributed edge‑computing nodes that process data locally to reduce latency and bandwidth usage.

End : Devices such as PCs, smartphones, TVs, and IoT meters that contribute idle compute resources, forming a shared‑compute pool.

Effective scheduling across these tiers requires a network that can interconnect all compute nodes.

Core Functions of a Computing Power Network

Routing : The network senses compute demand and provides optimal compute‑aware routes.

Scheduling : An intelligent "brain" orchestrates elastic, global allocation of compute resources.

Trading : Blockchain‑based platforms enable trusted compute‑resource transactions.

Enabling Technologies

Key technologies that satisfy the six characteristics include:

SRv6 : Replaces MPLS‑based manual provisioning with automated, minute‑level service activation, supporting pervasive access and agile onboarding.

Network Slicing : Creates isolated virtual networks ("专网") on a single physical infrastructure, guaranteeing lossless transmission and security for diverse services such as meteorology or academia.

IFIT (In‑Flow Inspection) : Inserts colored bits into live traffic to pinpoint packet loss, measure per‑hop latency/jitter, and reconstruct paths for real‑time monitoring.

Cloud‑Network Integrated Security : Deploys a unified security brain (e.g., QianKun Cloud) and edge security boundaries, leveraging zero‑trust, quantum‑grade encryption, and adaptive policies to protect compute, data, and network layers.

National Initiatives and Future Directions

China’s "East Data West Compute" program directs massive compute demand from the east to data‑center clusters in the west, establishing a nationwide data‑center network. Future research focuses on:

Pervasive intelligent compute for smart‑city, smart‑factory, and autonomous driving.

Data‑centric heterogeneous architectures (GPU, FPGA, AI chips, DPU/IPU) that break the von Neumann bottleneck.

All‑optical transport networks with ultra‑high bandwidth and low latency.

Deterministic low‑latency networks for industrial digitalization.

Deep integration of compute and network (NFV/SDN) enabling compute‑aware routing and in‑network processing.

Intelligent "brain" that fuses network, cloud, data, AI, security, edge, and blockchain into a unified information infrastructure.

Trusted, blockchain‑based compute sharing platforms that democratize idle compute resources.

Green, low‑carbon designs that align with national carbon‑neutral goals.

Conclusion

Computing power networks represent a paradigm shift from human‑centric communication networks to machine‑centric infrastructures that dynamically allocate compute, storage, and network resources. By integrating advanced protocols (SRv6), slicing, real‑time monitoring, and unified security, they enable scalable, secure, and efficient services for the emerging AI‑driven digital society.