Industry news — Issue 03

In this issue: what the shop floor noticed

Five things the shop floor noticed — and what they mean for the parts you send us. From the Eternal CNC Strategic Advisory Group. Issued 29 September 2026.

Three numbers worth noting this issue

+29.9%

China 3D-printing equipment output YoY, August 2026 (vs industrial robots +34.6%, Li-ion cells +57.2%)

Source: National Bureau of Statistics, Sep 15 2026

¥2.782 billion

Value of a liquid-cooling equipment supply agreement signed by Shenling on Sep 16 (≈$415M, ~66% of its 2025 revenue)

Source: Shenling filing / Shanghai Securities News, Sep 28 2026

60%–70%

China's humanoid harmonic-reducer localization rate in Q1 2026 (domestic price 40%–60% of Japanese)

Source: CAICT (China Academy of Information and Communications Technology), Q1 2026

Five stories

What happened, and what we think it means

News first, then the line that matters: what changes for a buyer sending parts to a CNC shop.

Robotics / Screws 2026-09

Planetary roller screws: the "tendons" of humanoid robots enter mass production

2026 is widely called the "mass-production year" for humanoid robots, and the part closest to a machine shop — the planetary roller screw — is moving from prototype validation to scaled output. Capacity plans are landing fast: ZW Precision (Wuzhou Xinchun) expects 980,000 planetary roller screws + 2.1 million micro ball screws in annualized capacity by end-2026; Shuanglin's phase-1 100,000 reverse planetary roller screw line began production in June 2026 and is delivering samples to top OEMs; Xinjian Transmission's 1,000,000-unit/year planetary roller screw project has cleared land approval; Beite Technology is investing ¥1.85 billion in an R&D and production base and won a ¥330M BOI project in Thailand for roller screws. The harder technical anchor is Jinwo: its in-house internal-thread process holds 1-micron accuracy at large length-to-diameter ratios, above the best publicly known figure (3 microns); it already supports 50,000 humanoid robots' worth of screw parts a year, with a plan to reach 100,000 by end-2026, 200,000 in 2027 and 1,000,000 long-term, plus a Thailand plant in 2027. Hengli Hydraulics passed Tesla Optimus Gen3 audit in December 2025, makes screws to C3 accuracy (<12 µm error over 300 mm), and brings a Thailand plant online in 2026 sized for 80,000 robots' worth of screws. The capacity build-out is also pulling upstream equipment: Huachen secured an order for 100 thread-grinding machines, while Sijin Intelligent's high-end rolling machines and Xinzuobiao's internal-thread rollers ride the same wave.

What it means for your parts

The screw is the most "machinist-shaped" part of this cycle: the screw shaft is bearing-steel (GCr15-class) finish-turning + grinding, gated by cumulative lead error; the nut is a high aspect-ratio internal thread — thin-walled, prone to chatter, hard to keep tooth-form consistent. The line that matters most for anyone sending us parts: if your shop can finish-turn + finish-machine a C3-class screw shaft, or hold internal threads on nuts with a length-to-diameter ratio above 10:1, this round of orders finds you first. But don't ship a screw as an ordinary shaft — lead error, thread pitch diameter and thread-flank roughness (which directly set transmission efficiency and life) are three hard gates. On volume, the climb from samples to "1,000,000 units a year" is long, and that is exactly the small-batch rapid-prototyping window — before mass production, someone has to make the samples stable first.

Source: All-weather Tech "Planetary Roller Screws: the Joint Lifeline of Humanoids", 2026; Huaxin Securities "2026 Planetary Roller Screw Capacity Build-out"; Sina Finance humanoid supply-chain roundup, 2026; Jinwo investor transcript (via third-party knowledge base), 2026.

Compute / Liquid cooling 2026-09-28

Liquid-cooling orders booked into 2027: quick connectors and manifolds are the real bottleneck

Issue 02 gave liquid cooling its "ruler" (GB/T 48023 / 48097). This issue is the other end of that ruler — demand has exploded. Economic Daily (via United Daily News) reported on Sep 7 that liquid-cooling chain makers are disclosing bulk orders, with several firms booked through year-end and lines running two shifts. Shanghai Securities News on Sep 28 confirmed "booked into 2027": JinFu's orders run to 2027, with staff working overtime through the Mid-Autumn and National Day holidays; on Sep 16 Shenling signed a $415M (≈¥2.782 billion) cooling-equipment supply deal with an overseas customer — about 66% of its 2025 revenue — and H1 new orders in its data-service segment were ~2.3× the prior year, with ~¥1.8 billion liquid-cooling-related. Xingrui's AI-server hardware revenue hit ¥60.5M, +221% YoY in H1, with nearly ¥1 billion in new orders YTD. The worst supply-demand mismatch is at the UQD quick connector — highest reliability barrier, few qualified suppliers, expansion lagging data-center delivery; manifolds and high-end pumps/valves are also tight on engineering precision. CICC estimates the 2026 global liquid-cooling market for smart computing centers at about ¥114.7 billion, +273% YoY.

What it means for your parts

This is the "implementation" chapter of Issue 02's standard. The standard wrote ±10% flow resistance and interchangeability into the rules; demand has now written the capacity gap into reality — the tightest piece isn't the cold plate (Taiwanese makers are already at million-plates-a-month) but the UQD quick connector and the manifold. For anyone sending us parts: it's no longer "should I make liquid-cooling parts," it's "capacity can't keep up and the big players are fighting over stable suppliers." If your connector body / manifold holds ±10% flow resistance batch to batch and stays interchangeable, that is exactly the class being fought over. On the drawing: cold-plate substrates / lids (flatness, flow channels), manifolds (cavities, ports) and UQD bodies (turning + seal faces) are 5-axis milling + pressure-leak-test work, typically 10–500 pieces — marking seal-face roughness, leak rate and flow resistance before you send the drawing matters more than anything.

Source: Economic Daily (via United Daily News) "Liquid-cooling makers booked through year-end, two shifts", Sep 7 2026; Shanghai Securities News "Booked into 2027! Liquid cooling is hot", Sep 28 2026; Cailian Press liquid-cooling chain coverage, Sep 2026; CICC smart-computing liquid-cooling market estimate, 2026.

CNC industry 2026-09-15

Machine-tool H1 report, re-read: cutting-tool exports +23% beat the machines; domestic five-axis is taking import's ground

The China Machine Tool & Tool Builders' Association (CMTBA) first-half figures (disclosed via World Metals on Sep 15) deserve a closer look: industry revenue ¥530.9 billion, +7.6%; metal-cutting machine tool output 432,000 units, +6.9% (NBS); import-export total $17.2 billion, +5.9%, with a $6.49 billion trade surplus. Splitting "machines" from "consumables" is the interesting part — Sina Finance's monthly high-end-manufacturing export report (Jan–Aug 2026) shows machine tool exports at $3.95 billion, -3%, but cutting-tool exports at $2.32 billion, +23% (Asia 40% / Europe 37% / North America 13%). While whole-machine exports are still grinding, precision tools are taking overseas market share. On imports, metal-working machine tool imports were $2.61 billion, -2.6% — the import decline exceeds the export decline, meaning import substitution on high-end machines is accelerating.

What it means for your parts

Two inferences that hit you directly: (1) cutting-tool exports +23% means fiercer competition among domestic precision toolmakers, so your consumable cost is coming down — the same end mill is cheaper this year than last, and that saving lands in your per-piece cost. (2) machine imports falling while domestic five-axis takes share means more downstream shops are running domestic five-axis — shops like ours that can both program domestic five-axis and manage fixtures become more attractive. The "profit doubled but margin only 3.7%" note from Issue 01 is still the footnote here: if the machine builders themselves barely make money, the price war isn't over, and CNC jobbing can't win on price hikes — pricing power still lives in "consistency + lead time + small batch."

Source: CMTBA 2026 H1 operating data, via World Metals, Sep 15 2026; NBS metal-cutting machine tool output; GACC / Sina Finance "High-end Manufacturing Customs Export Monthly (Jan–Aug 2026)".

Robotics / Reducers 2026

Domestic precision reducers hit "ten-thousand-unit monthly delivery": joint cores move from chokepoint to substitution

CAICT's Q1 2026 industry monitor shows China's humanoid harmonic-reducer localization rate has reached 60%–70%, with domestic product priced at 40%–60% of Japanese imports, cutting whole joint transmission cost by 30%–50% and pulling domestic humanoid unit cost down to the ¥100,000 class. On capacity, leaders have fully switched from "sample delivery" to "annual-framework bulk supply": GreenHarmony's humanoid-lightweight P-series harmonic hits 5 arc-seconds (matching Harmonic Drive's high end), is the sole domestic Tier-1 harmonic supplier to Tesla Optimus, and its 5,000,000-unit humanoid-dedicated line went fully online in July 2026 with orders booked through end-2027; Shuanghuan's subsidiary Huandong holds RV reducer orders exceeding ¥23 billion, deliverable to 2028; Giant Wheel's XT cycloidal reducer passed 3,000-hour endurance testing and is 50% lighter; Quanzhibo and Laifu Harmonic run automated integrated-joint lines at 90 seconds per set, >96% yield. Industry data shows leading makers' monthly shipments have cleared 10,000 sets.

What it means for your parts

The "invisible but most expensive" reducer parts — the flexspline (thin-walled cup, easy to distort, hard to turn and mill), the circular spline (internal gear), the flexible-bearing ring, the output flange — are all precision machined parts. Rising localization plus ten-thousand-unit monthly delivery means their bulk machining demand is moving from "samples" to "annual frameworks." For anyone sending us parts: reducer makers are hunting stable suppliers for flexsplines / circular splines / bearing rings; if your thin-walled parts don't distort and your internal gear form is consistent, that is a steadier, longer contract than robot structure work. But remember Issue 02 story 1 — a reducer ships as a four-part set (flexspline / circular spline / bearing / flange); one dimension drift scraps the set, so tolerances are extreme and what's competed on is batch consistency, not a pretty single part.

Source: CAICT (China Academy of Information and Communications Technology) Q1 2026 industry monitor; Hangzhou International Humanoid Robot Exhibition / AndroidShow industry coverage, 2026; Sina Finance core-component roundup, 2026.

Process 2026-09-15

3D-printing equipment +29.9%: additive isn't taking CNC's work, it's backfilling it

In the NBS August industrial data, one item gets upstaged by robots and Li-ion cells: 3D-printing equipment output rose +29.9% YoY (vs Li-ion +57.2%, industrial robots +34.6%, high-tech manufacturing value-added +16.7%). Metal additive equipment also shows a small export line (~$8M in August). Behind the growth are parts with conformal cooling, lightweight lattices and complex internal cavities — the "subtractive can't do, or can't do cheaply" class — and they overlap heavily with this issue's story 2, the liquid-cooling cold plate (conformal channels).

What it means for your parts

One clarification: additive doesn't replace CNC, it backfills CNC. Complex cavities, conformal cooling and lattice light-weighting form fast by printing, but almost every printed part then goes back to turn / mill to create datums, threads and assembly features — printing owns "forming," CNC owns "fits, seals and measures." For anyone sending us parts: if your part has conformal cooling or a deep cavity, we can evaluate a "3D printing + follow-up CNC finish" hybrid route, which is often faster and cheaper than pure five-axis roughing. Not a gimmick — one more option to pull your lead time down.

Source: NBS "Main Industrial Production Data, August 2026", Sep 15 2026 (Xinhua, CCTV News coverage); GACC / Sina Finance high-end manufacturing export monthly (metal 3D-printing equipment), Jan–Aug 2026.

How we write this page

Three questions we get

How often is this page updated?

Weekly. We do not run a news ticker — we run a reading list, and we only include items that change what a buyer of CNC parts should do this quarter.

Where do the numbers come from?

Primary sources only: National Bureau of Statistics, General Administration of Customs, CMTBA, CAICT, company filings, and named press. We do not republish analyst notes as facts.

Why a shop-floor view and not a market report?

Because a market report tells you where the wind is blowing. A shop view tells you whether the work has actually landed. Every story here has a "what it means" line — that is the part we are responsible for.

Keep reading

Related on the blog

Next issue

12 October 2026

Storylines we are following: whether embodied AI shown at the industry fair reaches real production lines; whether AI full-inspection tightens the tolerances on the drawings we receive; and whether the roller-screw nut's internal thread becomes the next chokepoint.

Send us your next drawing

Have a part in mind?

Send the STEP and the PDF. We will quote within 2 hours on a working day, and we will tell you whether your tolerances are what they should be.