Computing Power Networks: Definitions, Metrics, and Technological Challenges

01. Computing power is the core productive force of the digital economy. The proliferation of computing‑power network infrastructure marks a milestone of the intelligent era, driven by 5G, IoT, AI, and industrial internet. China’s "new infrastructure" policy emphasizes "computing‑power infrastructure", and initiatives such as "East‑Data West‑Compute" accelerate the construction of national computing hubs.

02. Definitions: China Mobile describes a computing‑power network as a new information infrastructure that integrates network, cloud, data, AI, security, edge, terminal, and chain. China Unicom defines it as the efficient allocation of compute and storage resources across cloud‑edge‑terminal via network means. Both aim to make computing as ubiquitous as water and electricity.

03. Computing‑power vs. electricity: While electricity’s ubiquity is achieved by a simple power grid, computing power requires a networked infrastructure that can pool heterogeneous resources (data‑centers, super‑computers, edge nodes) and meet strict latency requirements. The article discusses Nordhaus’s information‑theoretic definition of computing power and the SOPS metric, noting its limited practicality for modern networks.

SOPS = 0.05{[6+log2(memory)+word length]/[(7*addtime+mult time)/8]}

For computing‑power networks, a more pragmatic metric is the maximum number of virtual machines, containers, or bare‑metal servers the system can host, reflecting both resource capacity and user‑perceived performance.

04. Infrastructure challenges: The "three‑horizontal‑one‑vertical" architecture (service layer, network‑control layer, resource layer, and orchestration) currently focuses on integration rather than core chip and data‑transfer technologies. Efficient implementation depends on breakthroughs in underlying processing chips, DPU, and high‑speed optical‑electrical networks.

05. Technological breakthroughs:

Breakthrough 1: Domain‑Specific Architectures (DSA) to boost application‑level computing power.

Breakthrough 2: Data Processing Units (DPU) to offload infrastructure workloads and enable in‑network computing.

Breakthrough 3: Data‑plane proxies powered by DPU to improve serverless and cloud‑native service‑mesh performance.

The article concludes that a successful computing‑power network will act as a new "bottle" for digital productivity, requiring coordinated advances in chips, optics, and network programmability to sustain China’s digital‑economy growth.