---
title: "27,000 Private 5G Networks, 1,260 5G Factories: China Wired the Floor First"
date: 2026-11-19
category: Policy & Governance
site: NeuroAI
canonical: https://neuroai.site/a/na-policy-china-5g-factories-1119b
language: en
---

# 27,000 Private 5G Networks, 1,260 5G Factories: China Wired the Floor First

> China has built 27,000 private 5G networks and 1,260 5G factories, a base that predates the current model cycle and constrains what can be deployed.

The networks were built before anyone could say precisely what would run on them.

That is the plain reading of four figures released in Wuhan at the 2026 China 5G+Industrial Internet Conference. Gao Dongsheng, chief economist at the Ministry of Industry and Information Technology, put the count of industrial 5G private networks at 27,000. The same remarks counted 1,260 5G factories, more than 750 excellent-tier smart factories and 49 at the lighthouse tier, the top of a four-level grading system.

Four numbers, and only one of them describes a network. The other three describe plants.

## Two different units

The 27,000 and the 1,260 get read as a pipeline: networks first, factories later, the second number eventually catching up to the first. That reading is generous.

A private network is a connectivity decision. A 5G factory is an integration decision, and it drags in machines, controllers, vendor software, process engineers and a maintenance contract that outlives the radio equipment. The first can be procured. The second has to be earned, one line at a time.

So the distance between 27,000 and 1,260 is not a lag waiting to be closed on a schedule. It is the price of everything that sits above the physical layer.

## Why the floor asked for its own spectrum

He Shuping, chairman of China Information Communication Technologies Group, told the conference that industrial scenarios carry rigid demand for ultra-low latency, high reliability and compute that can be scheduled on demand.

Rigid is the operative word. An office tolerates a dropped packet. An arm closing on a mold does not.

Public networks are engineered for average behavior: downlink-heavy, best-effort, priced for consumers. A factory needs the opposite shape, uplink-heavy and deterministic, and it needs to own the timing so that a firmware update somewhere upstream cannot change when a press closes. That argument never appears on a coverage map, which is why private spectrum was the precondition rather than the product.

## What was actually on the floor

The exhibits are the more honest disclosure. A 5G smart factory. A base layer that fuses 5G with BeiDou for timing and position. An embodied AI robot dog delivering molds. An embodied AI robot sorting material. And, still under test, a 6G cloud base station that folds sensing, communication and computing into one radio, shown beside 6G simulated terminals.

Read that list as a sequence rather than a display: connectivity, then position, then mobility, then sensing. Each step moves the network closer to the object being made.

The robot dog is the tell. An arm bolted to a frame is a machine-tool problem. A dog walking a mold across a floor is a coverage problem, because the endpoint moves and the link has to hold while it does.

## A 6G station with the correct label

A 6G base station on a conference floor is not a product announcement, and the release does not dress it as one. It says the equipment was under test. That is the right status for 2026.

What the demo does announce is direction. Sensing and compute are moving into the radio itself, so the network can report where a thing is and not only that it connected. Position is the missing variable in most factory automation business cases, which is why putting it in the radio matters more than another throughput record.

Two reference guides were also published at the same event, one for industrial internet in aviation manufacturing and one for agricultural machinery. The pairing is deliberate. Aviation is high-value and low-volume; farm equipment is thin-margin and seasonal. A reference architecture that is offered to both is being pitched as portable, not bespoke.

## Hubei&#x27;s 30 percent

The only demand-side figure in the release came from the host province. Hubei reports AI application penetration above 30 percent among industrial enterprises above designated size.

That is a low number, and it is the most useful one here. It says the installed base of 5G factories is a leading edge, not a description of the sector. Most of the floor has not been instrumented yet, and the gap between the best plants and the median plant is where the next several years of spending sit.

It also reframes the four national figures. Twenty-seven thousand networks is a supply-side count. Thirty percent is a demand-side count. They can both be true while most factories still run on wire.

## What the wiring buys

My judgment is that the 27,000 networks are worth more as an option than as an achievement. Spectrum is allocated, radios are mounted, and the marginal cost of connecting the next machine falls. That is a real asset even while the application layer stays thin.

But an option is not output. The number that would prove the bet is not another network count. It is the 1,260, and specifically whether it rises while the 27,000 stays flat. That combination would mean existing infrastructure is finally being integrated, which is the cheap version of the transition.

The expensive version is the one where both numbers climb together and the bill arrives every year, in radios, in integration and in the engineers who keep the two layers talking. China has already paid the first installment. The second one is decided plant by plant, and no conference release can shortcut it.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-china-5g-factories-1119b
Free to quote with attribution and a link to the original.
