In the optical module finished-product segment, Zhongji Innolight rose 10.28% and Eoptolink rose 8.08%, while TFC Optical Communication, which supplies them with optical components and optical engines, rose only 6.79%. Upstream underperformed downstream — is the money still stuck in finished products, or is there a problem at the component end?
- Yuanjie fell 47.92% but rebounded only 0.77%.
- Cumulative gross margin rose from 51.87% to 62.56%, up 10.7 percentage points.
- Xinyisheng's 125.7 yi yuan, up 96.90%.
- In the same quarter, Zhongji's revenue was 222.8 yi yuan, up 174.57% year-on-year
- Net profit attributable to parent 8.79 yi yuan, up 55.30% year-on-year
In the optical module finished-product segment, Zhongji Innolight rose 10.28% and Eoptolink rose 8.08%, while TFC Optical Communication, which supplies them with optical components and optical engines, rose only 6.79%. Upstream underperformed downstream — is the money still stuck in finished products, or is there a problem at the component end?
光模块整机中际旭创涨10.28%、新易盛涨8.08%，给它们供光器件和光引擎的天孚通信只涨6.79%，上游跑输下游，是钱还堵在整机还是器件端出了问题？
The desk answers已核实Walnut, from public filings
The premise of this needs correcting: those three gains are as of the October 28 close, and there was not a single new earnings report intraday that day. Tianfu's Q3 report was discussed at a 19:00 conference call on the 28th and only posted online on the 29th; Zhongji's proposal to cancel repurchased shares only went to the board on the 29th; Xinyisheng's Q3 report was scheduled for the 30th.
Let me finish the conclusion first — it's not that money is stuck at the system-assembly level, nor that the component end is being squeezed on price. Tianfu's single-quarter gross margin was 65.05%, 17 to 18 points higher than the two system-assembly makers; what it actually gives up is payment terms, 92 days, one more month of fronting money than the system-assembly makers.
That day's ranking says nothing about where money is going
Since October, Tianfu fell 22.76%, Zhongji fell 21.29%. The two are almost identical; upstream did not systematically underperform downstream. The real underperformance happened only on the 28th.
That day's ranking was roughly the ranking of declines.
| Company | Since October (%) | October 28 (%) |
|---|---|---|
| Yuanjie Technology | (47.92) | 0.77 |
| Eoptolink | (33.99) | 8.08 |
| T&S Communications | (27.20) | 5.96 |
| Shijia Photons | (25.63) | 7.07 |
| Accelink Technologies | (25.04) | 3.03 |
| TFC Optical Communication | (22.76) | 6.79 |
| Zhongji Innolight | (21.29) | 10.28 |
Xinyisheng fell 33.99% and rebounded 8.08%, Zhongji fell 21.29% and rebounded 10.28%, Tianfu fell 22.76% and rebounded 6.79%; Yuanjie fell 47.92% but rebounded only 0.77%. Within the same "component" label, Shijia Photon rose 7.07%, a 6.3-point gap from Yuanjie.
Explaining prices by "layer" does not hold up in this cross-section. Dispersion within components is wider than between system assembly and components.
The most direct reason Tianfu rebounded the least is simply that it fell the least in October (aside from Zhongji). Zhongji had one more trigger on the 28th: the announcement on the 23rd stated that the board on the 29th would review changing the use of repurchased shares to cancellation.
Tianfu's Q3 report would not be priced until the open on the 29th — on the day this question asks about, the market had not yet seen it.
The component end is not being squeezed on price; the statements are clearer than the prices
In the first three quarters, Tianfu's operating cost was 17.79 yi yuan, down 5.69% year-on-year; operating revenue rose 21.25%. The absolute cost amount is falling while revenue is rising; cumulative gross margin rose from 51.87% to 62.56%, up 10.7 percentage points.
The single quarter is clearer: Q3 gross margin 65.05%, versus 53.69% a year earlier, 11.4 points higher.
| Company | 2026Q1 | 2026Q2 | 2026Q3 | Year-on-year change |
|---|---|---|---|---|
| TFC Optical Communication | 56.60 | 64.65 | 65.05 | 11.37 |
| Zhongji Innolight | 46.06 | 46.42 | Not disclosed | 4.93 |
| Eoptolink | 49.16 | 47.96 | Not disclosed | 1.32 |
Tianfu is 17 to 19 percentage points higher than the two system-assembly makers. The system-assembly makers' gross margins are also rising, just by much less.
In Q3 the company's gross margin remained relatively stable, mainly thanks to the company's continued advancement of lean management in cost control, product optimization and operational efficiency improvement, including continuously raising the level of production automation, optimizing process flows and product design, and improving yield.
Management says "stable"; the statements show it rising. And it rose against the backdrop of a higher share of active business — if optical engines were ramping by conceding price, this number should be moving down.
Q3 single quarter: revenue 19.23 yi yuan, up 31.50% year-on-year, up 28% quarter-on-quarter; net profit attributable to parent 8.79 yi yuan, up 55.30% year-on-year; excluding non-recurring items 8.79 yi yuan, up 55.78% year-on-year, almost entirely from the main business.
The real difference is growth rate; the pit was buried in Q2
In Q2 Tianfu's revenue was 14.98 yi yuan, negative 0.9% year-on-year. In the same quarter, Zhongji's revenue was 222.8 yi yuan, up 174.57% year-on-year; Xinyisheng's 125.7 yi yuan, up 96.90%.
One negative growth, two doubled — this is not a price problem, it is a delivery problem. The difference is not who pushed prices down, but who could deliver goods.
The company's own account is that the active business was affected by upstream core material supply; by Q3, "the tight situation in core material supply had eased markedly compared with earlier," delivery capacity recovered, and only then did quarter-on-quarter growth reach 28%.
So the real constraints at the component end are two: first, revenue elasticity does not move in sync with the unit-price elasticity of modules — when modules move from 800G to 1.6T, unit prices jump, while components are sold by the unit, so volume grows but price does not jump; second, Q2 was choked once by materials.
Money is stuck in payment terms, and the direction is reversed
Tianfu's Q3 accounts receivable were 19.45 yi yuan, up 73.33% from the start of the year, while revenue grew only 21.25% over the same period. Converted by single-quarter revenue, the collection cycle lengthened from 68 days in Q3 last year to 92 days.
By the same method, Zhongji is 61 days, Xinyisheng 55 days. The component end fronts more than a month more than the system-assembly end.
Xinyisheng's half-year report shows accounts payable of 71.25 yi yuan, 19.44% of its total assets. That is the money it holds back from upstream.
Look at cash collection: Tianfu's first-three-quarter operating cash flow was 14.62 yi yuan, 70% of net profit; Zhongji's first-half operating cash flow was 18.00 yi yuan, only 13% of net profit, and down 44% year-on-year; Xinyisheng's first half was 16.16 yi yuan, 21% of net profit.
If "money jammed at system assembly" means cash, it is exactly backwards. The system-assembly makers are the tightest cash link in this chain; their profits have all turned into receivables and inventory — Zhongji at half-year end had receivables of 150.01 yi yuan and inventory of 198.26 yi yuan.
Money is not jammed on anyone's books; the whole chain is fronting money toward the cloud vendors, and the component end is in turn fronting money toward the system-assembly makers. Tianfu's inventory was 12.74 yi yuan, up 178.65% from the start of the year, and the announcement described it as "an increase in raw materials brought by strategic stockpiling"; at half-year end Zhongji had 198.26 yi yuan and Xinyisheng 116.55 yi yuan, also both stockpiling.
All three companies' single-quarter revenue is still moving up quarter-on-quarter — TFC's third quarter was up 28% quarter-on-quarter, Zhongji's second quarter up 14% quarter-on-quarter, and Eoptolink's second quarter up 51% quarter-on-quarter. This is stocking up in line with the 1.6T and CPO production schedules, not unsold inventory piling up.
The market's price for the component end has always been more than double that of the complete-machine end
| Company | Trailing four-quarter net profit attributable to parent (yi CNY) | Market cap on October 28 (yi CNY) | P/E ratio (x) |
|---|---|---|---|
| TFC Optical Communication | 26.35 | 2192 | 83.2 |
| Zhongji Innolight | 173.17 | 7063 | 40.8 |
| Eoptolink | 107.34 | 3222 | 30.0 |
The calculation is the October 28 closing market cap divided by the trailing four quarters of net profit attributable to parent. TFC's trailing profit is the first three quarters' 20.83 yi plus last year's fourth quarter's 5.52 yi; Zhongji's is 136.51 yi plus last year's second half's 36.65 yi; Eoptolink's is 75.29 yi plus last year's fourth quarter's 32.05 yi.
The component end's valuation is 2 to 2.8 times that of the complete-machine end. Saying "the money hasn't reached the component end yet" finds no footing in the valuation — it arrived long ago, and at the most expensive position.
There is also an account for the expensiveness: TFC's August equity incentive set the 2027 to 2029 net profit targets at growth of 150%, 400%, and 700% over 2025; converted at 2025's 20.17 yi yuan, that is 50 yi yuan in 2027 and 101 yi yuan in 2028. An 83x price-to-earnings ratio amounts to calculating more than 40x on the 2027 target profit.
The strongest argument of the opposing side, and three falsification conditions
A smart opponent would say: TFC's high gross margin is bought with scarce part numbers; once CPO splits the optical engine out of the pluggable module and the complete-machine makers and switch makers build the optical engine themselves, it is the link that gets bypassed. TFC's betting its incremental growth from 2027 on CPO-supporting FAU and ELS precisely shows that it too is switching tracks.
This worry has a corresponding line in the accounts: TFC's single-quarter revenue divided by Zhongji's single-quarter revenue was 14.3% in the third quarter of 2025 and 6.7% in the second quarter of 2026. This ratio can only be read as a direction — TFC's customers are not limited to Zhongji alone; but directionally, the value of components per module really is thinning.
There are three signals that I have judged wrong:
- On the 30th the two complete-machine makers have scheduled disclosure of their third-quarter reports; if single-quarter gross margin falls by more than 1.5 points quarter-on-quarter for each and TFC still holds at 65%, then "the complete-machine end takes the pricing power" is wrong;
- If TFC's fourth-quarter single-quarter revenue comes in below 21 yi yuan and quarter-on-quarter growth below 10%, that shows the third quarter's 28% was merely a one-off catch-up delivery after materials recovered;
- If TFC's "receivables divided by single-quarter revenue" continues to rise from the current 101% and breaks above 120%, that is when the component end is truly being squeezed.
As for that day's 3.5 percentage point difference in gains, one trading day does not carry enough information; there is no need to use it as a signal.
Sources: Suzhou TFC Optical Communication Co., Ltd. "2026 Third Quarter Report" 2026-10-29, announcement no. 2026-064; "2026 October 28 TFC Communication Investor Relations Activity Record Form" no. 2026-012; "Announcement on the Company's Recent Production and Operating Conditions" 2026-10-23, announcement no. 2026-057. Zhongji Innolight Co., Ltd. "Announcement on the Company's Recent Production and Operating Conditions" 2026-10-23, announcement no. 2026-100; "2026 Semi-Annual Report" 2026-08-22. Chengdu Eoptolink Technology Inc., Ltd. "Announcement on the Company's Recent Production and Operating Conditions" 2026-10-23, announcement no. 2026-061; "2026 Semi-Annual Report" 2026-08-25. Daily market data taken from the market data interface at the 2026-10-28 close.
这条的前提要修正：那三个涨幅是10月28日收盘的，当天盘中没有一份新财报。天孚的三季报是28日晚上19点的电话会讲的、29日才挂网；中际的回购注销议案29日才上董事会；新易盛的三季报预约在30日。
先把结论说完——不是钱卡在整机，也不是器件端被压价。天孚单季毛利率65.05%，比两家整机厂高十七八个点；它真正让出去的是账期，92天，比整机厂多垫一个月的钱。
那一天的排序，说明不了资金往哪走
十月以来，天孚跌22.76%，中际跌21.29%。两家几乎一样，上游并没有系统性跑输下游。真正跑输只发生在28日那一天。
那天的排序，大致就是跌幅的排序。
| 公司 | 十月以来（%） | 10月28日（%） |
|---|---|---|
| 源杰科技 | (47.92) | 0.77 |
| 新易盛 | (33.99) | 8.08 |
| 太辰光 | (27.20) | 5.96 |
| 仕佳光子 | (25.63) | 7.07 |
| 光迅科技 | (25.04) | 3.03 |
| 天孚通信 | (22.76) | 6.79 |
| 中际旭创 | (21.29) | 10.28 |
新易盛跌33.99%弹8.08%，中际跌21.29%弹10.28%，天孚跌22.76%弹6.79%；源杰跌47.92%却只弹0.77%。同一个"器件"标签里，仕佳光子涨7.07%，跟源杰差6.3个点。
按"层"给价格找理由，在这个截面里对不上。器件内部散得比整机和器件之间还开。
天孚弹得最少，最直接的原因就是它十月跌得最少（除中际外）。中际28日还多一个由头：23日的公告写明，29日董事会要审议把回购股份的用途改为注销。
天孚的三季报要到29日开盘才会被定价——这道题问的那一天，市场还没看过它。
器件端没被压价，报表比价格清楚
前三季天孚营业成本17.79亿元，同比降5.69%；营业收入涨21.25%。成本绝对额在掉、收入在涨，累计毛利率从51.87%抬到62.56%，抬了10.7个百分点。
单季看得更清楚：三季度毛利率65.05%，去年同期53.69%，高了11.4个点。
| 公司 | 2026Q1 | 2026Q2 | 2026Q3 | 同比变化 |
|---|---|---|---|---|
| 天孚通信 | 56.60 | 64.65 | 65.05 | 11.37 |
| 中际旭创 | 46.06 | 46.42 | 未披露 | 4.93 |
| 新易盛 | 49.16 | 47.96 | 未披露 | 1.32 |
天孚比两家整机厂高17到19个百分点。整机厂的毛利率也在涨，只是涨幅小得多。
三季度公司毛利率保持相对稳定，主要得益于公司在成本管控、产品优化及运营效率提升等方面持续推进精益化管理，包括持续提升生产自动化水平、优化工艺流程和产品设计、提高良率等。
管理层说"稳定"，报表是升的。而且是在有源业务占比提升的背景下升的——如果光引擎上量靠的是让价，这个数应该往下走。
三季度单季：收入19.23亿元，同比增31.50%，环比增28%；归母净利润8.79亿元，同比增55.30%；扣非8.79亿元，同比增55.78%，几乎全是主营。
真正的差在增速，坑埋在二季度
二季度天孚收入14.98亿元，同比是负0.9%。同一个季度，中际收入222.8亿元、同比增174.57%；新易盛125.7亿元、同比增96.90%。
一家负增长，两家翻倍——这不是价格问题，是交付问题。差的不是谁把价格压低了，是谁能交出货。
公司自己的说法是有源业务受上游核心物料供应影响；到三季度，"核心物料供应紧张局面已较此前明显缓解"，交付能力恢复，环比才拉出28%。
所以器件端的真实约束是两件：一是收入弹性不跟模块的单价弹性同步——模块从800G换到1.6T，单价是跳着走的，器件按只数卖，量增价不跳；二是二季度被料卡了一次。
钱卡在账期上，而且方向是反的
天孚三季度应收账款19.45亿元，比年初增73.33%，收入同期只增21.25%。按单季收入折算，回款周期从去年三季度的68天拉长到92天。
按同一个算法，中际61天，新易盛55天。器件端比整机端多垫一个多月。
新易盛半年报的应付账款71.25亿元，占它总资产的19.44%。这是它压在上游的钱。
再看收现：天孚前三季经营现金流14.62亿元，是净利润的70%；中际上半年经营现金流18.00亿元，只有净利润的13%，还同比降了44%；新易盛上半年16.16亿元，是净利润的21%。
"钱堵在整机"如果指现金，恰好反了。整机厂是这条链上现金最紧的一环，利润全变成了应收账款和存货——中际半年末应收150.01亿元、存货198.26亿元。
钱没有堵在谁账上，是整条链都在往云厂商那边垫，器件端再往整机厂那边垫。天孚的存货12.74亿元、比年初增178.65%，公告写的是"战略性备货带来的原材料增加"；半年末中际198.26亿元、新易盛116.55亿元，也都在囤。
三家单季收入都还在环比往上走——天孚三季度环比28%、中际二季度环比14%、新易盛二季度环比51%。这是跟着1.6T和CPO的排产在备料，不是卖不动的积压。
市场给器件端的价，本来就是整机的一倍多
| 公司 | 滚动四季归母（亿元） | 10月28日市值（亿元） | 市盈率（倍） |
|---|---|---|---|
| 天孚通信 | 26.35 | 2192 | 83.2 |
| 中际旭创 | 173.17 | 7063 | 40.8 |
| 新易盛 | 107.34 | 3222 | 30.0 |
算法是10月28日收盘市值除以滚动四个季度归母净利润。天孚的滚动利润是前三季20.83亿加去年四季度5.52亿；中际是136.51亿加去年下半年36.65亿；新易盛是75.29亿加去年四季度32.05亿。
器件端的估值是整机端的2到2.8倍。说"钱还没到器件端"，在估值上找不到落点——它早到了，而且是最贵的位置。
贵也有个账：天孚8月的股权激励把2027到2029年的净利润目标定在比2025年增长150%、400%、700%，按2025年20.17亿元折算，就是2027年50亿元、2028年101亿元。83倍的市盈率，等于在按2027年的目标利润算40多倍。
反方最强的话，和三个证伪条件
一个聪明的反对者会说：天孚的高毛利是料号稀缺换来的，等CPO把光引擎从可插拔模块里拆出去、整机厂和交换机厂自己做光引擎，它就是被绕开的那一环。天孚把2027年起的增量押在CPO配套的FAU、ELS上，恰恰说明它自己也在换赛道。
这个担心在账面上有对应：天孚单季收入除以中际单季收入，2025年三季度是14.3%，2026年二季度是6.7%。这个比值只能当方向看，天孚的客户不止中际一家；但方向上，每块模块里的器件价值量确实在变薄。
我判断错的信号有三个：
- 30日两家整机厂预约披露三季报，如果单季毛利率环比各降1.5个点以上、天孚还守在65%，那"整机拿走定价权"就错了；
- 天孚四季度单季收入如果不到21亿元、环比不到10%，说明三季度的28%只是物料恢复后的一次性补交付；
- 天孚"应收除以单季收入"如果从现在的101%继续升破120%，那才是器件端真的在被压。
至于那天3.5个百分点的涨幅差，信息量不够一个交易日，不必当信号用。
来源：苏州天孚光通信股份有限公司《2026年第三季度报告》2026-10-29，公告编号2026-064；《2026年10月28日天孚通信投资者关系活动记录表》编号2026-012；《关于公司近期生产经营情况的公告》2026-10-23，公告编号2026-057。中际旭创股份有限公司《关于公司近期生产经营情况的公告》2026-10-23，公告编号2026-100；《2026年半年度报告》2026-08-22。成都新易盛通信技术股份有限公司《关于公司近期生产经营情况的公告》2026-10-23，公告编号2026-061；《2026年半年度报告》2026-08-25。日行情数据取自行情接口2026-10-28收盘。
## 毛利率不按上下游排，按卡不卡排
把这条链摊开，结论比"上游跌得少、下游涨得多"复杂得多。最上游的光芯片毛利82.06%，同在无源器件这一档的光纤连接器只有33.27%；下游的模块整机反而是46%到48%。
| 环节 | 2026Q2毛利率（%） | 2026Q2收入同比（%） | 合同负债/单季收入（%） |
|---|---|---|---|
| 光芯片 源杰科技 | 82.06 | 372.73 | 27.1 |
| 光芯片/PLC 仕佳光子 | 37.47 | 65.15 | 1.5 |
| 光纤连接器 太辰光 | 33.27 | 44.16 | 0.03 |
| 无源器件+光引擎 天孚通信 | 64.66 | 15.15 | 7.5 |
| 模块整机 中际旭创 | 46.42 | 182.49 | 0.09 |
| 模块整机 新易盛 | 47.96 | 100.34 | 1.73 |
"层"不是解释变量，一比就露。四家挂着"光器件"标签的同行公司做的是四种生意：源杰卖EML和DFB光芯片，仕佳卖PLC分路器加一部分芯片，太辰光卖MT插芯和光纤跳线，天孚卖无源器件加光引擎代工。它们的毛利率从33.27%到82.06%，中间跨了48.8个百分点——比"上游对下游"那3.5个点的涨幅差大一个数量级。
同一天的涨幅也是这么散的：源杰0.77%，新易盛8.08%，中间隔7.3个点。用"上游"给价格归类，在这个截面里找不到落点。
同一层里的节奏也不同步。太辰光一季度单季收入同比还是负8.04%，二季度才回到正44.16%；仕佳一季度32.18%、二季度65.15%、三季度116.06%，一路往上。连接器是先掉后弹，光芯片是一路加速——这不是同一门生意的脉搏。
## 卡点在源杰那一层，不在器件
源杰二季度单季收入5.70亿元、同比增372.73%，毛利率从去年同期的51.71%抬到82.06%。量翻近五倍、毛利率抬了30.35个百分点，还带27.1%的客户预付款——这是产能最紧的那一环才会同时出现的三件事。
同一季度，天孚收入同比只有15.15%。中际182.49%，新易盛100.34%。差在哪一环，天孚自己说得很清楚。
> 有源业务随着上游核心物料供应情况较前期逐步改善，公司有源光器件交付能力提升，带动业务收入增长。
到三季度，这句话变成"核心物料供应紧张局面已较此前明显缓解"。
所以二季度那个15.15%不是被客户压价，是天孚被自己的上游卡了料。这是卡点传导最标准的形态：卡点不在你这一环，但你的产出被别人卡住，报表上就显示成"你跑输"。
链条从下往上排是这样的：云厂商资本开支 → 1.6T模块（中际、新易盛）→ 光引擎与无源器件（天孚、太辰光）→ 光芯片与关键物料（源杰、仕佳，以及未上市的DSP和海外EML）。越往前走越紧。
紧的那一环拿议价权，议价权最后落在毛利率上——这条推理在这条链上可以逐环验：源杰82.06%、天孚64.66%、中际46.42%，正好是离卡点由近到远。
还有一层时间差被那天的涨幅盖住了。天孚二季度被料卡住，三季度的料才缓解、收入环比才拉出28%——器件的报表比芯片的报表晚一个季度。拿同一天的涨跌去比上下游，比的是两个不同季度的现实。
天孚夹在中间，两头都在抢，但抢的方向不一样：下游抢它的产能，它抢上游的料。
## 客户先给它付钱，方向是反的
如果器件端真被压价，第一件不会发生的事就是客户先付钱。天孚三季度合同负债3.66亿元，比年初的1.61亿元增128.14%，季报的原话是"预收客户款项增加"。
```chart
type: bar
title: 天孚通信合同负债（客户预付），二季度见底、三季度跳三倍
unit: 亿
labels: 2025Q4, 2026Q1, 2026Q2, 2026Q3
values: 1.61, 1.47, 1.12, 3.66
```
按单季收入折算，天孚三季度是19.0%，源杰二季度27.1%，新易盛1.73%，中际0.09%。整机厂几乎收不到预付款，因为它的客户是云厂商，账期从来不是它定的。越靠近卡点的一环，越能先收到钱——这条规律在表里是连贯的，没有一个例外。
天孚同时还占着上游的钱：应付账款7.07亿元、比年初增75.75%，应付票据2.84亿元、增207.20%。它自己也往上预付材料款0.92亿元、增343.75%，季报写的是"预付材料款增加"。
下游收预付、上游占款、中间把12.74亿元压在存货里。钱没有"堵"在某一环，它是被换成了存货，三家都在干：中际半年末198.26亿元、新易盛116.55亿元。
所谓"钱还堵在整机"，在报表上找不到对应的科目。整机厂的现金并不比器件端宽裕，天孚还在同期拿出3000万元自有资金参设私募基金、公告了1亿到1.5亿元的回购——它的现金也没紧到要省。
## 它跑输，是因为身份换了、货还没交够
先分清一件事：市场不是没给它价。器件端的市盈率一直是整机端的两到三倍，它拿的是这条链上最贵的位置。所以跑输不是没人认它的新身份，是新身份还没交出足够的货。
- 备货不是为这个季度，是为CPO：存货比年初增178.65%，原话"战略性备货带来的原材料增加"；记录表说得更白，"为CPO等新兴技术路线配套逐步上量做好物料准备"。苏州、江西、泰国三地扩产也在推进。
- 新故事要等2027年：CPO配套产品"目前已进入批量交付阶段"，但"预计CPO产品自2027年起逐步提量"；NPO还在"研发推进、客户送样和验证阶段"，"预计2027年也有望实现规模量产"。今年剩下两个季度，增量还是现在这套无源加有源。
- 回购注销这件事两家都有：天孚10月23日董事会通过、24日公告，1亿到1.5亿元、价格上限350元/股、全部注销减资；中际是23日公告要上董事会审议。催化剂基本对称，天孚的方案还更具体，它照样少涨3.5个点。
那3.5个百分点落不到任何一份文件上，只能当噪音。
## 反方最强的话：65%是定制价，不是稀缺价
一个不同意这套解释的人会说：天孚的高毛利不是产能稀缺换来的，是给少数客户做定制化光引擎换来的，那是"设计加代工"的价。等CPO把光引擎从可插拔模块里拆出去，模块厂自己建线做，这十六到十八个点就该还回去。
这个反驳有力，因为动机在报表里看得见：中际46.42%、新易盛47.96%，自己做光引擎能省十几个点，它们有足够理由自建。天孚把2027年的增量押在CPO配套的FAU、ELS上，本身也说明它知道可插拔模块里的光引擎会被收回去。
我给它一个限定：三季报不披露前五大客户占比，客户集中度这一条我核不到，只能当风险提，不能当结论。
## 什么情况下我错了：三个证伪条件
- 源杰、太辰光的三季报还没出。如果它们的单季毛利率出来低于二季度，"卡点在最上游"就错了。已经出的仕佳三季度40.13%、环比升2.66个点，光迅25.51%、环比升0.93个点，中游还在往上走。
- 天孚四季度合同负债如果掉回2亿元以下，3.66亿元就只是一次性预收，"客户预付等于锁产能"不成立。
- 存货除以单季收入，天孚现在是66%（12.74除以19.23）。如果四季度比值继续升、收入环比又不到10%，备货的性质就从"为CPO准备"变成"卖不动"。
来源：苏州天孚光通信股份有限公司《2026年第三季度报告》2026-10-29，公告编号2026-064；《2026年10月28日天孚通信投资者关系活动记录表》编号2026-012；《关于回购股份方案的公告》2026-10-24，公告编号2026-059；《关于与专业投资机构共同设立私募基金的公告》2026-10-26，公告编号2026-061。中际旭创股份有限公司《关于公司近期生产经营情况的公告》2026-10-23，公告编号2026-100；《2026年半年度报告》2026-08-22。成都新易盛通信技术股份有限公司《关于公司近期生产经营情况的公告》2026-10-23，公告编号2026-061；《2026年半年度报告》2026-08-25。陕西源杰半导体科技股份有限公司、河南仕佳光子科技股份有限公司、深圳太辰光通信股份有限公司、武汉光迅科技股份有限公司2026年单季数据取自各自定期报告累计数轧差（本期累计减上期累计），报告期2026-06-30及2026-09-30。