1,118 quotes from 456 Chinese CNC factories: what the numbers say, and what they leave out
Fifty-seven minutes. That is the median wait between a buyer sending a machining enquiry and the first Chinese factory answering it, across 1,118 competing quotes gathered between 30 January and 6 February 2026. Half the enquiries were answered inside the hour. Ninety per cent were answered inside six.
The number comes from Haizol, a Shanghai RFQ marketplace, which audited 456 Chinese CNC machining factories and pushed 60 live buyer enquiries through its own platform to see what came back. It is the largest published dataset of its kind. It is also research by a marketplace, into whether marketplaces are fast — which is the first thing worth saying about it, and the last thing the coverage said.
The order book is prototypes, not mass production
Take the marketplace framing out and one finding still reframes the whole picture. Chinese CNC capacity is not primarily serving big runs. It is serving ones and twos.
| Order tier | Share of the 60 RFQs |
|---|---|
| Prototype (1–5 units) | 43.3% |
| Small batch (6–50) | 20.0% |
| Medium batch (51–500) | 21.7% |
| Large batch (501–5,000) | 11.7% |
| Mass production (5,000+) | 3.3% |
Median minimum order quantity: 10 units. At the 25th percentile it was 2. Just under two thirds of all demand in the sample — 63.3% — was for 50 pieces or fewer.
That is the opposite of the story most shops in this region tell themselves about Chinese competition. The comfortable version is that China takes the 10,000-piece jobs and leaves the awkward short runs alone. On this data it takes the awkward short runs too, and it quotes them in under an hour. Haizol says a one-to-five-piece enquiry pulled an average of 19 competing quotes — the same crowd a production order attracts.
The reason offered in Haizol's own video on the report is worth repeating because it is a business argument rather than a technical one: quoting a prototype is a relationship signal. A factory that takes a five-piece job is bidding for the 500-piece job that might follow. Whether that is true across the market or true of factories that have signed up to an RFQ marketplace is exactly the question the dataset cannot answer about itself.
The discount three quarters of buyers never ask for
Only one enquiry in four asked for pricing at multiple quantities — a price at 10, at 100, at 1,000. The ones that did were quoted 37% off at mid volume and 54% off at scale, according to Haizol. The step pricing was already sitting in the suppliers' models. Nobody made them produce it.
The size of the step depends heavily on what the part is cut from:
| Material | Average discount at scale |
|---|---|
| Stainless steel | 43.9% |
| Brass / stainless mix | 38.0% |
| Aluminium | 34.3% |
| Carbon steel | 15.8% |
Carbon steel barely moves, and that makes sense at the machine: cheap stock, short cycle, so there is not much fixed cost per part to spread. Stainless is the reverse — slow cycles, tool wear, more setup discipline, and the fixed share of the job is large enough that volume genuinely changes the unit price. A buyer quoting a stainless part at one quantity is leaving close to half the price on the table. A buyer quoting a carbon steel bracket at one quantity is leaving about a sixth.
What the certification line gives away
Here is the number the write-ups skipped. Of the 456 audited factories, 69.3% say they serve automotive customers. Only 27.9% hold IATF 16949.
That gap is the whole risk in one line. Two thirds of these shops are cutting automotive parts; barely a quarter carry the quality system the automotive supply chain is built on. Forty-three per cent claim aerospace work. Haizol does not publish a matching AS9100 figure, and its absence next to a published IATF number is not an accident of layout.
The capability side is real and should not be waved away. Haizol's audit found 38.8% running five-axis milling on DMG MORI, Mazak and Makino machines, 48.2% offering Swiss machining to ±0.005 mm, and 39.0% running EDM to ±0.002 mm. The equipment is there. The paperwork that lets a buyer prove the equipment was used properly is there in about half as many places.
The comparison that was modelled, not measured
Haizol's headline saving is 42% against a Western supplier after freight and duties, with a best observed case of 83%. The report also runs a scenario against a notional $1,000 Western baseline: $174 best case, $582 typical, $1,257 worst.
Read the last one again. The worst case is a 26% premium — you pay more than you would have paid at home. And Haizol is straight about why the range is that wide: the report states it does not directly measure Western competitor pricing or production lead times. The Chinese half is measured. The Western half is assumed.
That does not make the comparison worthless. It makes it a model, and a model with a published downside case deserves more trust than a marketing number without one. But anyone quoting "42% cheaper" as a fact about the world is quoting an estimate against a baseline nobody sampled.
One tier in the report should be left alone entirely. Haizol's "Huge" complexity band rests on two RFQs and the report flags it as directional. Two data points is an anecdote with a decimal place.
Where the capacity actually sits
Three provinces hold 82.2% of the audited factories: Jiangsu with 138, Guangdong with 132, Zhejiang with 105. Widen it slightly and roughly 90% sits in two coastal clusters, the Yangtze River Delta and the Pearl River Delta.
Quote commitment across the whole sample — quotes honoured at the price given — ran at 98%, or 1,095 of 1,118. Jiangsu came in at 100% on every tracked quote, Guangdong at 99%. On the delivery side Haizol puts prototype parts at 10 to 17 days door to door by air, and 30 to 45 days by sea.
The concentration is the part a buyer should hold onto. "Sourcing from China" is not a country decision. It is a decision about two river deltas, and a shop 1,500 km inland from either is a different proposition with the same passport.
What a shop here should do with this
Fifty-seven minutes is a target, not a talking point. If a machining enquiry lands in a Taxila inbox on Monday morning and is answered on Tuesday afternoon, the competitive gap in this dataset is not tolerance, machine brand or price. It is response time, and it costs nothing to close.
The second thing is the multi-tier quote. Three quarters of buyers in the sample never asked for one, which means most shops never publish one. Sending a price at 10, 100 and 1,000 without being asked answers the buyer's next question before they ask it, and it puts the setup cost argument in your words rather than theirs.
And the certification gap cuts both ways. A shop that can put a current IATF 16949 or AS9100 certificate in front of an automotive or aerospace buyer is competing against a field where, on Haizol's own audit, most of the alternative cannot.
Send us the drawing and the quantity — or three quantities — and we will come back with a price, a tolerance we will actually hold and a date. If we are not the right shop for the part, we will say so on the same day.
Sources
- China CNC Machining Industry Report 2026 — Haizol's own report page, the primary source for the audit of 456 factories and 1,118 quotes. Every figure in this article that is attributed to Haizol comes from this report or from the company's own video summary of it.
- CNC Machining in China: We Analyzed 1,118 Factory Quotes (What You're Getting Wrong) by HAIZOL — the company's seven-minute walk through its own findings, source for the 42% median saving, the 83% best case, the 19-quote average on prototype orders and the province-level commitment rates.
- How to Source Low-Volume CNC Machining (MOQ, Lead Time and Cost Guide) by HAIZOL — on why setup cost dominates short runs and why high-mix job shops and high-volume plants answer the same enquiry differently.
The data window is unambiguous — factory audits during 2026, live RFQs run 30 January to 6 February 2026. The publication date is not: the report has circulated under more than one date this year, so this article dates the data rather than the announcement.