Why Intelligent Computing Centers Are the Backbone of China’s AI Boom

Definition

Intelligent Computing Centers (智算中心) have emerged as a new type of public infrastructure that provides AI‑required compute, data, and algorithm services. They differ from traditional supercomputing or cloud data centers in purpose, technical standards, functions, application domains, and the "investment‑construction‑operation" model.

Industry Chain Analysis

The upstream of the industry chain includes civil engineering (construction, cooling, power distribution, telecom) and IT infrastructure (AI servers, networking, storage, data‑center management systems). The mid‑stream comprises intelligent computing services, IDC services, and cloud services. Downstream demand comes from sectors such as internet, finance, telecom, transportation, autonomous driving, robotics, metaverse, smart healthcare, and entertainment creation.

Leading AI Computing Architecture

According to the "Guidelines for Innovative Development of Intelligent Computing Centers," the core technologies are AI chips, AI servers, and AI clusters, with large AI models driving algorithmic advances. An operating system acts as the "neural hub" to manage and schedule compute resources efficiently, while a robust software ecosystem ensures usability.

Development Outlook and Forecast

Data volumes and algorithm complexity are exploding, pushing AI compute demand upward. The guide predicts that by 2028 China’s intelligent compute capacity will approach 2,800 EFLOPS, with a 47.58% CAGR from 2022‑2026. This growth will support the transition of 80% of AI scenarios to dedicated compute resources in intelligent centers.

Geographic Distribution

Existing and under‑construction centers typically offer around 100 PFLOPS. Large‑scale nodes (>1,000 PFLOPS) are concentrated in the Beijing‑Tianjin‑Hebei, Yangtze River Delta, and Pearl River Delta regions, targeting 300‑1,000 PFLOPS to meet big‑model workloads. Smaller nodes (40‑200 PFLOPS) are being built in cities and districts to complement the large nodes and serve diverse industry needs.

Service Models

Leasing services: Enterprises can rent compute resources either through full‑equipment leasing (building dedicated resource pools) or cloud‑style on‑demand services, enabling flexible scaling and reducing upfront investment.

Data services: Offer data governance, operation, and model training/fine‑tuning, as well as vertical AI solutions derived from model‑data incubation.

Operation services: Include government‑run, public‑private partnership, or fully private operation models, aiming to boost industrial innovation, scientific research, and public services.

Talent cultivation services: Provide training for data processing, labeling, and model development to help SMEs bridge technical gaps and fully leverage intelligent compute.