The market generally infers from Apple's new-model stocking that Dongshan Precision's (002384) fourth-quarter revenue will rise quarter-on-quarter. If the October revenue published by Hon Hai around November 10 turns negative year-on-year, does this inference still hold?
- The company overall was +63.95%, while the part including Apple was +13.29%.
- Last October was NT$895.7 billion.
- Optical modules plus precision components together contributed 88%.
The market generally infers from Apple's new-model stocking that Dongshan Precision's (002384) fourth-quarter revenue will rise quarter-on-quarter. If the October revenue published by Hon Hai around November 10 turns negative year-on-year, does this inference still hold?
市场普遍按苹果新机备货推算东山精密(002384)四季度营收环比走高，11月10日前后鸿海公布的10月营收若同比转负，这个推算还成立吗？
The desk answersverifiedWalnut, from public filings
Conclusion first: even if Hon Hai's October revenue turns negative year-on-year, it cannot overturn the forecast that Dongshan Precision's fourth-quarter revenue will rise quarter-on-quarter. What holds up this projection is not Apple stocking, but optical modules and AI PCB; the Apple line this year is only responsible for not being a drag.
To turn negative, it would have to fall 23% from September
Hon Hai's latest disclosure is September revenue, NT$1,158.6 billion, +38.4% year-on-year. The October figure will have to wait until early November; last October was NT$895.7 billion.
For October to turn negative year-on-year, revenue would have to fall below 895.7 billion, that is, 22.7% less than September.
From 2023 to 2025, Hon Hai's October revenue was 1.12, 1.10, and 1.07 times that year's September, respectively; not one year fell. A real drop of 23% would not be stocking slowing down a bit, but demand breaking off.
Hon Hai's total revenue is no longer a clean read on Apple
In the first three quarters, Hon Hai's revenue was +39.5% year-on-year. On the same Apple chain, Pengding Holdings was +8.7% and Luxshare Precision +29.0%. Apple assembly cannot support this growth rate; the increment is in AI servers.
| Company | Main segment | Q3 single-quarter revenue year-on-year | First three quarters cumulative year-on-year | First three quarters gross margin |
|---|---|---|---|---|
| Dongshan Precision | Flexible boards + AI PCB + optical modules | 73.8% | 67.6% | 21.28% |
| Pengding Holdings | Flexible boards (mainly handsets) | 14.2% | 8.7% | 25.03% |
| Luxshare Precision | Assembly + components | 14.6% | 29.0% | 11.91% |
| Hon Hai | Complete-unit assembly | — (September +38.4%) | 39.5% | — |
Pengding lists mobile phone flexible boards separately as "communication boards," with third-quarter single-quarter 71.40 yi, +10% year-on-year; this is the real speed of this round of Apple stocking. At its October 29 briefing, it attributed third-quarter growth to "AI servers, optical modules and other computing-related businesses."
The part that includes Apple accounts for only 45% of Dongshan
In the 2025 annual report, the largest customer was 186.43 yi, accounting for 46.46% of the full year (the annual report does not name it; in past years it was Apple), roughly equal to 70% of electronic circuit revenue, mainly MFLEX flexible boards.
In the first half of 2026 the structure changed. Electronic circuits were 125.30 yi, accounting for 45.07% of total revenue, with only +13.29% year-on-year.
| Product | 2026H1 Revenue (yi CNY) | Share | Year-on-year | Gross margin |
|---|---|---|---|---|
| Electronic circuits | 125.30 | 45.07% | +13.29% | 18.80% |
| Precision components | 65.81 | 23.67% | +178.63% | 12.13% |
| Optical modules | 53.51 | 19.25% | Newly consolidated | 39.33% |
| Optoelectronic display modules | 27.23 | 9.80% | -10.68% | 5.89% |
Breaking down the increment is more direct: first-half revenue increased 108.43 yi year-on-year, with electronic circuits contributing only 14.70 yi, or 13.6%; optical modules plus precision components together contributed 88%.
The company overall was +63.95%, while the part including Apple was +13.29%. The direction of the fourth quarter quarter-on-quarter is not something this 13% can determine.
A flexible board has to go through three steps before it reaches Dongshan
Where it comes from is Apple's October-to-November finished-product production scheduling. Who it passes through: assembly plant shipment confirmation, with module plants in between, and only then the flexible board plant. Where it lands: Dongshan's "electronic circuit products" line item.
The timing is later than one thinks. Flexible boards ship three to six weeks before assembly; boards installed in October were booked into Dongshan's revenue in July or August. Hon Hai's October figure cannot change Q3, and for Q4 it can only change the November-to-December stretch.
What moves first is not revenue, but accounts receivable and inventory turnover; at the end of September these two items were 126.87 yi and 123.67 yi.
To overturn the quarter-on-quarter rise, it would have to fall 20%
For the fourth quarter not to rise quarter-on-quarter, revenue would have to be below the third quarter's 175.82 yi. Over the past three years, Q4 versus Q3 was at its lowest +4.7% and at its highest +29%.
Calculated at 1.047 times, the worst year, the fourth quarter would be 184.1 yi. To fall below 175.82 yi, the Apple part would have to fall more than 20% quarter-on-quarter while optical modules and AI PCB do not increase at all.
The company answers "not convenient to disclose" to all questions on fourth-quarter optical module shipments, so this bit of confidence can only come from capacity line items: construction in progress 54.67 yi, +133.04% from the beginning of the year; the third-quarter report states this is expansion of optical modules and AI PCB. It is capacity, not orders.
The strongest opposing argument
The opposing side will say: the third quarter already shipped out all of Apple's goods. Dongshan's Q3 single-quarter gross margin was 23.07%, the highest in history, with a high share of high-priced models; the goods may all be piled into Q3, and the fourth quarter is an air gap.
This argument holds up halfway. In 2024, Q4 was only 4.7% higher than Q3, a year when the Apple chain was weak and new businesses had not yet filled the gap.
What is different this year is capacity. The second quarter was 146.60 yi and the third quarter 175.82 yi, +19.9% quarter-on-quarter; construction in progress rose from 23.46 yi at the beginning of the year to 54.67 yi, all pointing to optical modules and AI PCB. This line is ramping quarter by quarter and does not depend on fourth-quarter seasonality.
When I would be wrong: when the 2027 April annual report comes out, if 2026 fourth-quarter single-quarter revenue (full year minus 453.79 yi) is below 175.82 yi; or if Hon Hai's October decline exceeds 15% year-on-year rather than barely turning negative—the latter would mean what is being cut is volume, and "the magnitude is not enough" would not hold.
Source: Suzhou Dongshan Precision Manufacturing Co., Ltd. "2026 Third Quarter Report" 2026-10-31, announcement no. 2026-061; "2026 Semi-Annual Report" 2026-08-22 (Cninfo annoId 1225491182); "2025 Annual Report" 2026-04-22 (annoId 1225139318); "Investor Relations Activity Record" 2026-08-24 (annoId 1225497036); Pengding Holdings "2026 Third Quarter Report" 2026-10-29 (annoId 1225635653) and "Investor Relations Activity Record" 2026-10-29 (annoId 1225642145); Luxshare Precision "2026 Third Quarter Report" 2026-10-30 (annoId 1225645369); Hon Hai Precision September 2026 consolidated revenue, Taiwan Stock Exchange Market Observation Post System "Listed Company Monthly Revenue" September 2026.
结论先给：鸿海10月营收就算同比转负，也推不翻东山精密四季度营收环比走高。撑住这条推算的不是苹果备货，是光模块和AI PCB；苹果那条线今年只负责不拖后腿。
要转负，得比9月掉两成三
鸿海最新公布的是9月营收，NT$1兆1,586亿，同比+38.4%。10月的数字要等11月上旬，去年10月是NT$8,957亿。
10月要同比转负，营收得掉到8,957亿以下，也就是比9月少22.7%。
2023到2025年，鸿海10月营收分别是当年9月的1.12、1.10、1.07倍，没有一年是跌的。真跌两成三，那不是备货缓一缓，是需求断了。
鸿海的总营收，已经不是苹果的干净读数
前三季鸿海营收同比+39.5%。同一条苹果链上，鹏鼎控股+8.7%、立讯精密+29.0%。苹果组装撑不起这个增速，增量在AI服务器。
| 公司 | 主要环节 | Q3单季营收同比 | 前三季累计同比 | 前三季毛利率 |
|---|---|---|---|---|
| 东山精密 | 软板＋AI PCB＋光模块 | 73.8% | 67.6% | 21.28% |
| 鹏鼎控股 | 软板（手机为主） | 14.2% | 8.7% | 25.03% |
| 立讯精密 | 组装＋零部件 | 14.6% | 29.0% | 11.91% |
| 鸿海 | 整机组装 | —（9月+38.4%） | 39.5% | — |
鹏鼎把手机软板单列为「通讯用板」，三季度单季71.40亿、同比+10%，这是苹果这一轮备货的真实速度。它在10月29日说明会上把三季度增长归给「AI服务器、光模块等算力相关业务」。
含苹果的那块，只占东山四成半
2025年报里第一大客户186.43亿、占全年46.46%（年报不具名，历年是苹果），约等于电子电路收入的七成，主要是MFLEX软板。
2026年上半年结构换了。电子电路125.30亿，占总收入45.07%，同比只有+13.29%。
| 产品 | 2026H1收入（亿元） | 占比 | 同比 | 毛利率 |
|---|---|---|---|---|
| 电子电路 | 125.30 | 45.07% | +13.29% | 18.80% |
| 精密组件 | 65.81 | 23.67% | +178.63% | 12.13% |
| 光模块 | 53.51 | 19.25% | 并表新增 | 39.33% |
| 光电显示模组 | 27.23 | 9.80% | -10.68% | 5.89% |
拆增量更直白：上半年收入同比增加108.43亿，电子电路只贡献14.70亿，占13.6%；光模块加精密组件合起来贡献88%。
公司整体+63.95%，含苹果那块+13.29%。四季度环比的方向，不是这13%定得了的。
一条软板要走三步才轮到东山
从哪来是苹果10到11月的整机排产。经过谁：组装厂出货确认，中间隔着模组厂，最后才轮到软板厂。落到哪：东山「电子电路产品」这个科目。
时点比想的靠后。软板在组装前三到六周发货，10月装机的板子，钱七八月就记进东山了。鸿海10月的数改不了Q3，Q4也只能改11到12月那一截。
先动的不是收入，是应收账款和存货的周转，9月末这两项是126.87亿和123.67亿。
要翻掉环比走高，得掉两成
四季度要环比不涨，营收得低于三季度的175.82亿。过去三年Q4对Q3最低一次也有+4.7%，最高+29%。
按最差那年的1.047倍算，四季度是184.1亿。要掉到175.82亿以下，苹果那块得环比掉两成以上，同时光模块和AI PCB一点不增。
公司对四季度光模块出货量一律答「不便披露」，所以这点底气只能来自产能科目：在建工程54.67亿、比年初+133.04%，三季报写明是光模块和AI PCB扩产。是产能，不是订单。
反方最强的一条
反方会说：三季度已经把苹果的货打完了。东山Q3单季毛利率23.07%，历史最高，高价机型占比高，货可能全堆在Q3，四季度是空档。
这条站得住一半。2024年Q4只比Q3高4.7%，就是苹果链弱、新业务又没顶上的一年。
今年不一样的地方在产能。二季度146.60亿、三季度175.82亿，环比+19.9%；在建工程从年初23.46亿加到54.67亿，全部指向光模块和AI PCB。这条线是逐季放量，不靠四季度性。
什么情况下我错了：2027年4月年报出来，2026年四季度单季营收（全年减453.79亿）低于175.82亿；或者鸿海10月同比下降超过15%而不是勉强转负——后者说明砍的是量，「幅度不够」就不成立。
- 01苏州东山精密制造股份有限公司《2026年第三季度报告》2026-10-31，公告编号2026-061
- 02《2026年半年度报告》2026-08-22（巨潮annoId 1225491182）
- 03《2025年年度报告》2026-04-22（annoId 1225139318）
- 04《投资者关系活动记录表》2026-08-24（annoId 1225497036）
- 05鹏鼎控股《2026年三季度报告》2026-10-29（annoId 1225635653）及《投资者关系活动记录表》2026-10-29（annoId 1225642145）
- 06立讯精密《2026年三季度报告》2026-10-30（annoId 1225645369）
- 07鸿海精密2026年9月合并营收，台湾证券交易所公开资讯观测站「上市公司每月营业收入」2026年9月
## 鸿海这根尺子，刻度已经换了
鸿海2026年第二季度云端网络产品营收占比首次过半（中央社、经济日报，2026-08-12），苹果组装所在的消费智能产品退到第二位。10月初台湾媒体已经在用「AI营收超车iPhone」做标题。这条变化决定了10月营收这个指标的性质：它是AI机柜、苹果组装、笔记本、元件的混合读数，苹果那一块已经不到一半。
鸿海自己的季度数也印证了这一点。三季度营收同比+47%，9月单月同比+38.4%。9月是苹果新机起量的月份，增速反而低于季度平均，增量发生在7、8月——那是AI机柜交付的节奏。
「10月同比转负」最可能的成因因此有两个：一是去年10月的基数，8,957亿，当时的单月历史新高，同比+11.2%、月增7%（经济日报，2025-11-05）；二是AI机柜在季度之间的交付分布。两个原因都跟苹果砍单无关。
拿这个指标去证伪东山，误差比信号大。
## 同行三季度：环比不分腿，毛利率分腿
六家公司三季度单季环比都在两成上下，东山19.9%甚至是六家里最低的——因为它二季度的基数已经被光模块抬过一轮。所以环比这个维度，在苹果和AI同时拉货的三季度，区分不出两条腿。
| 公司 | Q3单季环比 | 2025年Q4对Q3 | 单季毛利率（2025Q3→2026Q3） |
|---|---|---|---|
| 东山精密 | 19.9% | 1.29 | 14.11% → 23.07% |
| 沪电股份 | 27.6% | 1.08 | 35.84% → 44.02% |
| 工业富联 | 25.0% | 1.23 | 7.00% → 6.99% |
| 鹏鼎控股 | 29.7% | 1.17 | 23.08% → 26.54% |
| 立讯精密 | 22.0% | 1.16 | 12.84% → 12.11% |
| 蓝思科技 | 29.3% | 1.00 | 19.09% → 16.78% |
分腿的是同比。蓝思科技三季度单季收入同比-8.1%，是六家里唯一负增长的；沪电+90.0%、工业富联+57.7%、东山+73.8%。中间没有过渡带，不是一个行业景气度能解释的分布。
毛利率更直白。东山单季毛利率一年抬了8.96个百分点，沪电抬了8.18个百分点，幅度几乎一样；同期蓝思掉2.31个百分点、立讯掉0.73个百分点。抬的是同一件事——高毛利的AI板、光模块占比上来了。同一个「苹果备货」的故事，解释不了这8.96个百分点。
## 不到两成的收入，四成的毛利
上半年光模块收入53.51亿，占19.25%，毛利率39.33%，折出毛利额21.04亿；同期毛利总额56.02亿，光模块占37.6%。电子电路收入占45.07%，毛利额23.56亿，占42.1%。
收入和毛利是两套权重。四季度环比在收入端由电子电路和光模块一起定，在利润端几乎由光模块单独定。而这套结构是今年才有的，去年这时候东山还是一家软板公司。
```chart
type: line
title: 东山精密单季毛利率
unit: %
labels: 2025Q3, 2025Q4, 2026Q1, 2026Q2, 2026Q3
values: 14.11, 14.71, 19.33, 20.89, 23.07
```
单季毛利率连续五个季度单边向上。这条线本身就是「增量来自高毛利新业务」的证据。
## 环比引擎是在苹果淡季换掉的
2026年一季度131.38亿、二季度146.60亿、三季度175.82亿，环比+11.6%、+19.9%。一季度和二季度是苹果的淡季，新机连影子都没有。
东山在这两个季度连续环比两位数增长，说明环比引擎在苹果不动的时候就已经换掉了。四季度的环比，不需要苹果先动。
反过来说，历史数据确实站在市场那边——2025年东山的四季度做到三季度的1.29倍，是六家里最高的一档，这个倍数就是软板时代留下的。问题是，今年一季度到三季度，已经把「不等苹果」这件事演了一遍。
## 时间差：10月的货，记在三季度的账上
软板在整机组装前三到六周发货。鸿海10月装的机器，对应软板在9月上中旬到10月初就出了，大头记在东山三季度。能落到四季度的，只有10月发货的那一小截，加上11、12两个月。
苹果新机的组装峰值在10到11月，12月自然回落，这一段本来就不指望环比。所以鸿海10月这个数就算转负，能改的也只是四季度里一个月左右的发货量。
落到报表上是「营业收入」和「应收账款」，不是存货。三季度末应收账款126.87亿，对应单季收入0.72个季度、约65天，年初是0.75个季度。账期没有放宽，说明三季度的收入不是靠压货压出来的。
## 反方最强的一条：卡在物料，不在苹果
真正能打掉四季度环比的，是光模块自己。公司在10月8日分析师会上讲：「光模块行业整体需求保持旺盛，上游核心物料供应整体偏紧」；被问到DSP，回答是「供应整体可控」——可控不等于充足。同一场会还说800G是主流、1.6T「正逐步起量」，迭代期本身就带着物料切换的风险。
产能也没落地。三季度末在建工程54.67亿、其他非流动资产27.07亿（预付的工程设备款），两项都指向光模块和AI PCB，但一个还在「在建」、一个还在「预付」，四季度变不成产量。能用的还是现在已投产的产线。
能证明「四季度有货可交」的只有预付款项：11.31亿，比年初2.74亿增312.21%，公司注明是索尔思预付的材料款。这是备货，不是订单。合同负债5.54亿，比年初增16.7%，只相当于单季收入的3.2%——订单是PO驱动，客户不预付，看不到锁量。
存货123.67亿、比年初增38.5%，公司归因索尔思销量增长；但同期资产减值损失2.78亿，主要是存货跌价准备增加。备的货里有一部分转得慢。
所以最聪明的反驳不是「苹果的货已经在三季度打完」，而是：苹果腿按季节性回落、光模块又被物料卡住交付，两件事同时发生，四季度环比贴着三季度走平。这个反驳不依赖鸿海，也不依赖苹果，它打在承重墙上。
还有一条容易被漏掉：东山上半年外销占83.20%，收入以美元结算。四季度人民币若走强，折算回人民币的收入会被压掉一截；三季度财务费用8.24亿、同比增1034%，其中就含汇兑损失。它只影响报表上的收入数字。
## 什么数字出来，这条判断就错了
- 2027年4月年报：2026年四季度单季收入低于175.82亿。
- 鸿海10月营收同比降幅超过15%，而不是勉强转负——那说明砍的是量，不是基数。
- 鸿海四季度法说会（11月中）对云端网络产品的四季度指引环比下滑，AI这条腿也熄火。
- 年报分产品表：光模块四季度收入低于三季度，环比引擎停转。
- 12月鸿海月营收连续两个月同比为负，同时鹏鼎四季度通讯用板环比下滑，苹果这条腿确认走弱。
来源：东山精密《2026年第三季度报告》2026-10-31（1225646907）、《2026年半年度报告》2026-08-22（1225491182）、《投资者关系活动记录表》2026-10-08（1225598207）；沪电股份《2026年三季度报告》2026-10-28（1225627804）；工业富联《2026年第三季度报告》2026-10-30（1225639532）；鹏鼎控股《2026年三季度报告》2026-10-29（1225635653）；立讯精密《2026年三季度报告》2026-10-30（1225645369）；蓝思科技《2026年三季度报告》2026-10-26（1225620808）；鸿海精密2026年9月合并营收（钜亨网2026-10-05）、2025年10月合并营收及同比（经济日报2025-11-05）、2026年二季度云端网络产品占比（中央社2026-08-12）。