Crazy Rich China | Data Centres, AI Computing Power and the National Computing Network

Explore this series: Crazy Rich China article directory · Crazy Rich World — all countries and learning routes.

Crazy Rich China | Data Centres, AI Computing Power and the National Computing Network begins with a question that sounds almost like electricity: where does computing power come from, and how do you move it to the people who need it?

Did you know? By the end of June 2026, China’s intelligent computing capacity reached 2,185 EFLOPS, according to the Ministry of Industry and Information Technology. National computing-facility utilisation reached 71.4%, and 52 large intelligent-computing facilities had been built with more than 10,000 accelerator cards each.

More than 70 high-speed computing interconnection links had also been built around national computing hubs, and 17 regional nodes for computing interoperability had been approved by September 2026.

That makes China data centres, AI computing power, East Data West Computing, cloud infrastructure China, GPU data centres, computing network and AI infrastructure part of one giant systems story: chips, electricity, cooling, fibre, software and geography becoming a new kind of national utility.

Did You Know? 2,185 EFLOPS Is Intelligent Computing, Not All Computing

An EFLOP represents one quintillion floating-point operations per second.

The 2,185-EFLOPS figure refers specifically to intelligent-computing capacity, the kind of accelerated computing used heavily for AI workloads.

It should not be treated as a measure of every computer in China.

Precise categories matter because computing statistics can otherwise sound larger or smaller than they really are.

A Data Centre Is a Factory for Computation

Traditional factories take physical inputs and transform them into physical outputs.

Data centres take electricity, data and instructions and transform them into computation.

Servers perform calculations.

Networks move data.

Cooling removes waste heat.

Software allocates work.

The product is invisible, but the infrastructure is intensely physical.

AI Changed the Demand Curve

Large AI models require enormous quantities of accelerator computing.

Training consumes large bursts of computation.

Inference consumes computing each time users ask models to generate text, images, code or decisions.

Government digital-economy reporting in 2026 said token usage by major Chinese models had risen more than 1,400-fold over two years.

AI therefore turns software demand into electricity and chip demand.

Why the East Wants Computing but the West Has Energy and Land

Many AI companies and internet firms are concentrated in eastern China.

Large western regions often have more abundant land and renewable-energy resources.

The East Data, West Computing initiative tries to connect those complementary advantages.

Some workloads can be processed farther from the user if network delay is acceptable.

Geography becomes part of cloud architecture.

The Eight National Computing Hubs

China’s national computing strategy is organised around eight national hub nodes and related data-centre clusters.

Five of the major hubs are in western regions.

The idea is not that every eastern server should move west.

It is that suitable workloads can be matched with suitable locations through a coordinated national network.

Latency Decides What Can Move

Not every computing job can be sent thousands of kilometres away.

Real-time control, interactive services and some financial or industrial applications need low latency.

Batch AI training, backup and some data processing can tolerate more distance.

The network therefore needs hierarchy.

Edge, regional and national computing locations solve different timing problems.

Why Fibre Is as Important as the GPU

A powerful accelerator is useless if data cannot reach it efficiently.

High-speed fibre links connect computing hubs with users and other data centres.

By mid-2026, more than 70 high-speed interconnection links had been built around China’s national hubs.

Computing power becomes more valuable when it can be scheduled across a network rather than trapped inside one building.

Utilisation Is the New Metric

By June 2026, national computing-facility utilisation was 71.4%.

That number matters because the industry is moving from “How many servers can we build?” toward “How efficiently can available computing be used?”

An idle accelerator still consumes capital.

The next phase of competition is scheduling, interoperability and workload matching.

Why 52 Giant AI Facilities Matter

MIIT reported 52 intelligent-computing facilities with more than 10,000 accelerator cards each by mid-2026.

That is an industrial scale far beyond a conventional enterprise server room.

These facilities support model training, inference and large scientific or commercial workloads.

The data centre becomes strategic infrastructure.

Cooling Is a First-Class Engineering Problem

Computers turn most electrical energy into heat.

As AI accelerators become denser, cooling becomes harder.

Air cooling may be sufficient for some racks.

High-density systems increasingly use liquid cooling or other advanced thermal-management approaches.

A data centre’s performance is partly determined by how effectively it removes heat.

Water Use Depends on Design

Some cooling systems use substantial water; others use more air, closed loops or different cooling technologies.

Climate also matters.

A data centre in a cool, dry western region faces different thermal conditions from one in a humid coastal city.

There is no single “data centre water footprint” that applies to every facility.

Engineering choices matter.

Electricity Is the Largest Hidden Input

AI computing turns electricity directly into digital work.

That makes power availability, grid reliability and price critical.

Western hubs can benefit from abundant wind, solar and other energy resources.

The computing economy therefore connects directly to the energy economy.

Read Data Centres beside Solar and Nuclear

China is expanding solar, storage, grids and nuclear at the same time that AI demand rises.

Those systems increasingly interact.

Read Crazy Rich China | Solar Power, Renewable Energy, Grid Scale and Battery Storage and Crazy Rich China | Nuclear Power, Hualong One and the Clean-Energy Engineering Economy.

Computing Power Is Becoming Schedulable

China is developing regional nodes designed to make computing resources easier to discover and allocate.

This moves the system toward a utility-like model.

A company may care less about exactly which building performs a workload and more about price, performance, data rules and latency.

Abstraction turns hardware into a service.

Cloud Computing Is the Commercial Interface

Most businesses do not want to buy and operate every server they need.

Cloud providers package computing, storage, databases and AI services into on-demand products.

That spreads infrastructure cost across many users.

The cloud is not somewhere in the sky.

It is someone else’s data centre presented through software.

AI Makes Chips Strategic

Intelligent-computing facilities depend heavily on GPUs and other accelerators.

That links data centres to semiconductor manufacturing, packaging, memory and high-speed networking.

Read Crazy Rich China | Semiconductors, Integrated Circuits and the Electronics Manufacturing Machine.

Compute demand creates pressure across the chip supply chain.

Why Data Centres Create Local Economic Questions

A data centre can bring investment, construction and technical jobs.

But it also consumes land and electricity and may use water depending on cooling design.

The local economic question is therefore not simply how many megawatts arrive.

Communities need to ask what jobs, taxes, infrastructure and environmental trade-offs accompany the facility.

What Students Can Learn from Data Centres

Physics — heat transfer and power;
Mathematics — utilisation, capacity and scheduling;
Computing — parallel processing, networks and cloud architecture;
Geography — why workloads and energy are distributed across regions;
Economics — capital utilisation and utility pricing; and
Civilisation — how intelligence becomes dependent on energy and infrastructure.

China and Singapore: Computing under Different Constraints

Singapore has strong regional data-centre demand but severe land, energy and carbon constraints.

China has continental scale and can distribute facilities across regions with very different energy and climate profiles.

The contrast is useful.

Read Making Singapore Rich | Digital Economy, AI and Data Centres.

Ten Vocabulary Words for Reading Computing Infrastructure

EFLOPS — quintillions of floating-point operations per second;
accelerator — specialised processor optimised for workloads such as AI;
rack — frame holding servers and networking equipment;
utilisation — proportion of available capacity actually being used;
latency — delay between request and response;
interconnect — high-speed link between computing systems;
inference — running a trained AI model to produce outputs;
training — adjusting an AI model using data and computation;
liquid cooling — removing heat using a circulating liquid system; and
orchestration — software coordination of workloads across computing resources.

Frequently Asked Questions

How much intelligent computing capacity did China have in June 2026?

MIIT reported 2,185 EFLOPS.

What was data-centre utilisation?

Overall computing-facility utilisation was 71.4% nationwide at the end of June 2026.

What is East Data, West Computing?

It is China’s strategy for coordinating computing demand in eastern regions with data-centre and energy resources in western regions through national hub nodes and network links.

Why are data centres important for AI?

Large AI models need enormous accelerator computing for training and inference, plus storage, networking, cooling and electricity.

Are all workloads suitable for remote western data centres?

No. Workloads with strict latency requirements may need to remain closer to users, while batch or delay-tolerant workloads can be placed farther away.

Do data centres use a lot of electricity?

High-performance computing is energy intensive. Actual power use depends on facility size, hardware, utilisation and efficiency.

Helpful Reading Across the China and Singapore Graph

References and Current Sources

State Council Information Office / MIIT, China’s intelligent computing capacity reaches 2,185 EFLOPS.
State Council Information Office, Briefing on industry and information technology in H1 2026.
Digital China Summit, China’s computing infrastructure expands in H1 2026.
Digital China Summit, A rapidly expanding computing network.
State Council Information Office, Nation’s smart computing scale surges.

Computing Power Is Becoming a Utility for Intelligence

Electricity made machines easier to deploy because factories no longer needed their own power source for every device.

Cloud computing did something similar for software.

Did you know? China’s national computing network is attempting the next abstraction: make vast pools of AI computation easier to discover, schedule and use across distance.