The Pearl River Delta, measured

The seven entities

Every row links to a profile built the same way: registry extract, audited accounts, and a direct measurement of the company’s site. Headcount is the registry figure except where only the insured-staff filing is published, which is noted in the profile.

What the cluster can actually do

Between them these seven cover the whole ladder, which is unusual for a sample this small. Prototype and quick-turn: Xunjiexing runs a 24-hour option in Shajing and a 60,000 m²/month smart prototype plant in Zhuhai that started producing in the first half of 2025. Volume and breadth: Champion Asia’s eight plants span 1 to 108 layers and include SMT and finished-product assembly, so a buyer can hand over a boxed product rather than a bare board. High-layer difficulty: Victory Giant and Shengyi are the two plants here whose 2024–2025 results were transformed by AI-server and switch demand.

What the cluster does not publish is a common vocabulary. Only two of the seven state a layer range at all (Champion Asia, 1–108; Suntak’s Thai line, 2–78 planned). Only Suntak publishes a monthly area capacity per factory; the other six do not. None states an HDI order (1+N+1, 2+N+2, any-layer), a maximum panel size, or a copper-weight ceiling.

The margin column is the most useful row in this table, because it is arithmetic rather than marketing. Two plants of similar revenue — Victory Giant at ¥10.7bn and Shengyi at ¥9.5bn — sit at 22.7% and 31.2%, and the difference is mix: Shengyi’s book is concentrated in high-layer communications and server work with a laminate advantage through its parent, while Victory Giant’s is broader across automotive, terminals and medical. Xunjiexing at 14.4% is a different business model again: prototype engineering sold at prices meant to win the volume that follows. Kinwong and Suntak sit between at 21.6% and 21.1% — the same margin with opposite balance sheets, Kinwong earning 10.35% on equity and Suntak 4.19% while it pays for a tenth factory. None of these margins is a quality signal. Each tells you which band your order would sit in.

Six wrong industry codes

This is the finding that matters most if your sourcing file is built from a database rather than from company visits. Six of these seven have a registry industry code recorded in this directory, and none of the six is a PCB code — worse, no two agree on what industry they are in.

The consequence is mechanical: a buyer, or a buyer’s assistant, filtering a supplier database for “printed circuit board manufacturing” returns none of these six. They appear only if you already know their names. Glorysky makes the point sharply — its registry record says vehicle repair while the commercial aggregator carrying the same record labels the company “PCB manufacturing” in its own industry field. One database, two answers, same plant.

The digital audit

Measured on 1 November 2026, from one network outside China. Every cell below is a direct observation, not a claim.

Four failures are worth naming individually, because each one is cheap to fix and expensive to keep.

What no factory here publishes

Zero of seven publish a certificate number. That sentence needs care, because all seven claim certification, and some of the claims are detailed: Champion Asia’s 2025 annual report names ISO 9001:2015, ISO 14001:2015, ISO 45001:2018, QC 080000:2017 and IATF 16949:2016 in its own pages; Xunjiexing claims IATF 16949, ISO 13485 and GJB 9001C, the military standard; Glorysky holds six national designations including 专精特新“小巨人”; Shengyi says multiple management systems are implemented without naming one; Kinwong and Suntak state nothing at all. Not one of them publishes a certificate number, a certification body, a scope statement or an expiry date. A buyer who needs to check whether a plant is certified has to ask, and then check the number in the issuing registry.

The same is true of the commercial basics. No plant here publishes a minimum order quantity. One publishes a lead time for prototypes (Xunjiexing, 24 hours). None publishes a stack-up library, an impedance tolerance, or IPC class. For seven factories with ¥47.5bn of revenue between them, that is not an accident of web design — it is a deliberate choice to keep the sales conversation private, and it is the reason a directory that records the absence is useful.

What a buyer should take from this

How this was compiled

Seven profiles, each from four source classes: the enterprise registry extract (name, USCC, capital, address and its change history, industry code, headcount, shareholder schedule); the audited annual report or published results; the company’s own website in both languages; and a direct measurement of that website on 1 November 2026 — DNS, HTTP and HTTPS behaviour, response size, the lang attribute, JSON-LD and hreflang counts, robots.txt and sitemap.xml.

Three limits worth stating. Headcount definitions differ — registry figure for six plants, insured-staff filing for Glorysky — so the 62,271 total is indicative, not exact. Revenue years differ, as flagged above. And nothing here was checked against customs records, court dockets or the IATF and UL registries; where a profile records a risk count from an aggregator, it says so and does not draw a conclusion from it.

This page measures publication, not quality. A plant that publishes nothing may build better boards than one that publishes everything — and in this cluster, the plant with the least machine-readable identity among the top three by revenue is the one that exports the most.