Zhongji Innolight rose 10.28%, with turnover of 235.7 yi; Tianfu Communication, which supplies it with optical components and optical chips, rose only 6.79%, and Shijia Photons rose 7.07%. Module makers are eating so full on orders, why haven't upstream players risen along with them?
- Module maker Zhongji Innolight only +4.6%, Eoptolink -16.5%.
- Active optical device revenue 12.60 yi, year-on-year -19.57%, gross margin 47.03%.
- Shijia +7.07%, 31.6 yi.
- Eoptolink +8.08%, 150.9 yi
- That day Zhongji Innolight +10.28%, turnover 235.7 yi
Zhongji Innolight rose 10.28%, with turnover of 235.7 yi; Tianfu Communication, which supplies it with optical components and optical chips, rose only 6.79%, and Shijia Photons rose 7.07%. Module makers are eating so full on orders, why haven't upstream players risen along with them?
中际旭创涨10.28%、成交235.7亿，给它供光器件和光芯片的天孚通信只涨6.79%、仕佳光子涨7.07%。模块厂订单吃得这么饱，上游为什么没跟着涨？
The desk answers已核实Walnut, from public filings
The ones that rose less on October 28 were not upstream, but module makers. The order of this chain this year is reversed: the closer to optical chips, the more they rose; the closer to complete modules, the less they rose.
That day Zhongji Innolight +10.28%, turnover 235.7 yi; Eoptolink +8.08%, 150.9 yi; TFC +6.79%, 73.7 yi; Shijia +7.07%, 31.6 yi. Turnover differed by three to seven times; the money was going to module makers, and upstream was only dragged along a bit.
Line up the eight companies for comparison, and the direction immediately reverses.
| Company | Position | October 28 | Year-to-date | P/E ratio |
|---|---|---|---|---|
| Zhongji Innolight | Optical modules | +10.28% | +4.6% | 36.6 |
| Eoptolink | Optical modules | +8.08% | -16.5% | 27.3 |
| TFC Optical Communication | Optical components | +6.79% | +39.4% | 83.4 |
| Shijia Photons | Chips/patch cords | +7.07% | +29.7% | 78.9 |
| Accelink Technologies | Components to modules | +3.03% | +76.7% | 67.7 |
| T&S Communications | Optical connections | +5.96% | +34.4% | 116.3 |
| Yuanjie Semiconductor | Optical chips | +0.77% | +89.7% | 139.1 |
| Changguang Huaxin | Optical chips | +3.36% | +23.3% | 515.2 |
The biggest gainers year-to-date are Yuanjie +89.7% and Accelink +76.7%; module maker Zhongji Innolight only +4.6%, Eoptolink -16.5%. This year the money went to upstream first, and module makers have been chasing.
The P/E column is today's closing total market cap divided by net profit attributable to parent for the most recent four quarters; year-to-date gains have been adjusted for bonus shares and transfers, TFC and Eoptolink both did bonus shares and transfers this year, and directly comparing the current price with the price at the start of the year would differ by a large margin.
The so-called "module makers eat their fill, upstream fails to keep up" treats one day's intraday chart as full-year pricing. October 28 was merely a slightly urgent catch-up.
Upstream's money is stuck at its own upstream
TFC is the most typical. In the first half of 2026, passive optical device revenue 15.30 yi, year-on-year +77.40%, gross margin 71.90%; active optical device revenue 12.60 yi, year-on-year -19.57%, gross margin 47.03%.
The active segment was 15.7 yi in the same period last year, accounting for six tenths four of revenue, and fell to four tenths five this year. Those two figures from last year (8.63 yi, 15.67 yi) were back-calculated from the +77.40% and -19.57% written in this half-year report itself.
The half-year report states it very plainly:
During the reporting period, shortages in the supply of certain materials had a certain impact on the company's capacity increase for active optical device products.
That one sentence. TFC's active business needs to purchase chips and core materials externally; when materials are insufficient, the production lines cannot run at full capacity. Zhongji Innolight's orders extend to 2027, and when passed to TFC they become "wants to deliver but cannot."
For the entire first half, TFC revenue was 28.28 yi, year-on-year +15.15%, while in the same period Zhongji Innolight was 417.78 yi, year-on-year +182.49%. The same phrase "full order book" lands on the two companies' statements as a difference of 15% and 182% — it is not that demand was not passed through, but that capacity could not catch it.
The third-quarter report makes up part of this gap: single-quarter revenue 19.23 yi, year-on-year +31.50%, net profit attributable to parent 8.79 yi, year-on-year +55.30%, single-quarter gross margin 65.05%, even higher than the second quarter's 64.66%.
But the announcement date of this third-quarter report is October 29; at the October 28 close the market did not yet have it.
Two other figures in the statement prove they are scrambling for materials: prepayments 0.92 yi, +343.75% versus the start of the year; inventory 12.74 yi, +178.65% versus the start of the year, and the company itself writes the reason as "increase in raw materials brought by strategic stockpiling." Scrambling for materials shows capacity is the bottleneck, not orders.
What Shijia sells is actually jumpers
Shijia is called an "optical chip" company, but opening the revenue breakdown in the half-year report, of main business 14.78 yi, fiber optic connectors were 7.62 yi, accounting for 51.5%; AWG 3.83 yi; indoor optical cable 0.98 yi; cable materials 1.37 yi; the DFB that is truly called a chip was only 0.49 yi, accounting for 3.3%, and PLC 0.25 yi, accounting for 1.7%.
The half-year report also clearly states the positioning of its own chips: DFB "has achieved stable volume supply in the access network field," continuous-wave DFB used in data center silicon photonics "some products have achieved volume supply," and EML "is currently in the customer verification stage." The high-speed laser chip that is most scarce in 1.6T modules — it has not yet begun to sell it.
Its third-quarter report is very strong: single-quarter revenue 12.27 yi, year-on-year +116.06%, net profit attributable to parent 2.71 yi, year-on-year +226.19%, much faster than TFC's +31.50%. But the base is there — single-quarter 12 yi versus Zhongji Innolight's 223 yi.
Cash also comes back slowly: operating cash flow for the first three quarters -4.39 yi, net profit 5.86 yi, receivables 11.78 yi, +128% versus the start of the year, inventory 10.10 yi, doubled versus the start of the year. The company's stated reasons are "increased inventory stockpiling to meet product order demand; revenue grew quickly, and payment terms for part of goods sold have not yet come due" — stockpiling is for orders, but the money is indeed still with customers and in the warehouse.
Module makers have locked in upstream's prices
In Zhongji Innolight's October 23 announcement there are two sentences, one about locking volume, one about locking price:
The company will continuously strengthen supply chain assurance capabilities through measures such as providing diversified product solutions, strategic binding with core suppliers, and signing long-term supply agreements.
Currently downstream demand is growing rapidly, manufacturers with large-scale delivery capability are limited, the market price system is healthy, and product quotes to customers are relatively rational.
Translated: when supply is tight, module makers use long-term agreements to lock both upstream volume and price. What upstream gets is a price-locked long-term order, with volume growing linearly with module makers' shipments, but no pricing upside. Module makers sell it to cloud providers, quotes are "rational," and the gross margin goes to themselves.
Innolight's own Q3 report only goes to the board today; the most recent complete financial report is the interim report dated August 22. The denominator of its 36.6x P/E is still the old one; if Q3 profit continues to grow quarter-on-quarter, once the report is swapped the multiple will clearly move lower. This is also the reason funds dare to buy before the report comes out.
Another piece of corroborating evidence is where the stockpiling sits. Innolight's inventory at the end of H1 was 198.26 yi, and at the end of Q3 2025 it was 109.02 yi, up 81.9%; prepayments rose from 0.80 yi at the end of last year's Q3 to 9.54 yi. Materials across the whole chain are being stockpiled by the module maker, and it also sets the production schedule; upstream delivers on its rhythm.
Look at one more layer of structure: TFC's active gross margin is 47.03% and passive is 71.90%, a gap of 24.9 percentage points in between. A rising active share will itself drag on blended gross margin; in Q3, while active was recovering, blended gross margin still rose from 64.66% to 65.05%, showing that pricing and mix in the passive segment are not broken. The module maker does not carry this burden—what it sells is the module itself.
The strongest counterargument
TFC's Q3 gross margin of 65.05% was higher than Q2, with active recovering; the FAU and ELS supporting CPO are already in stable mass production, and the company says CPO products will gradually ramp from 2027. If active continues to scale in Q4, the upstream segment's report will catch up and the share price will catch up too—so it is only a timing gap, not a logic gap.
I accept half of this. The timing gap is real, but the money has already been paid: TFC 83.4x, Shijia 78.9x, Innolight 36.6x.
TFC also gave a marker—the equity incentive assessment requires net profit in 2027, 2028, and 2029 to grow 150%, 400%, and 700% versus 2025; based on 2025 net profit attributable to parent of 20.17 yi, that implies 50.4 yi in 2027 and 161.4 yi in 2029. Targets set this high mean, on the other side, that any quarter that falls a bit short kills the valuation, not the earnings.
Under what circumstances this judgment would be wrong
- Innolight's board meets today (October 29) to review the Q3 report; if single-quarter gross margin falls below 45% (Q2 was 46.42%), it means the module maker is also being squeezed on price, and "full orders equal module-maker profitability" no longer holds.
- If TFC's Q4 single-quarter revenue is less than 22 yi (only 14% growth over Q3's 19.23 yi), it means the materials relief was one-off, and the upstream capacity story needs to be re-rated.
- If Shijia's fiber optic connector share keeps rising while DFB stays at the 0.5 yi level, then it is a connector company, and the 78.9x P/E has no corresponding business.
The real sequence in this chain is: chips tighten first, components tighten next, modules ramp last; the share price also rises in chips first, components follow, modules catch up. What happened on October 28 was the last leg.
What to watch in the next leg is not whether orders are full—Innolight's orders are already booked out to 2027, and this information has long been public—but whether TFC's active can scale quarter-on-quarter for two consecutive quarters, turning "cannot handle it" into "can handle it."
Sources: Zhongji Innolight "Announcement on the Company's Recent Production and Operating Conditions" 2026-10-23 (2026-100); Zhongji Innolight "2026 Interim Report" 2026-08-22 (2026-081); Zhongji Innolight "H-Share Announcement (Date of Board Meeting)" 2026-10-15; TFC Communication "Announcement on the Company's Recent Production and Operating Conditions" 2026-10-23 (2026-057); TFC Communication "2026 Interim Report" 2026-08-19 (2026-040); TFC Communication "2026 Third Quarter Report" 2026-10-29 (2026-064); TFC Communication "Record of Investor Relations Activity on October 28, 2026" (2026-012); Shijia Photons "2026 Interim Report" 2026-08-01; Shijia Photons "2026 Third Quarter Report" 2026-10-22.
10月28日涨得少的那一头不是上游，是模块厂。这条链今年的顺序是反的：越靠近光芯片涨得越多，越靠近整机模块涨得越少。
那天旭创+10.28%、成交235.7亿，新易盛+8.08%、150.9亿，天孚+6.79%、73.7亿，仕佳+7.07%、31.6亿。成交额差三到七倍，钱是往模块厂走的，上游只是被带了一下。
把八家公司排开对比，方向立刻反过来。
| 公司 | 位置 | 10月28日 | 年内 | 市盈率 |
|---|---|---|---|---|
| 中际旭创 | 光模块 | +10.28% | +4.6% | 36.6 |
| 新易盛 | 光模块 | +8.08% | -16.5% | 27.3 |
| 天孚通信 | 光器件 | +6.79% | +39.4% | 83.4 |
| 仕佳光子 | 芯片/跳线 | +7.07% | +29.7% | 78.9 |
| 光迅科技 | 器件到模块 | +3.03% | +76.7% | 67.7 |
| 太辰光 | 光连接 | +5.96% | +34.4% | 116.3 |
| 源杰科技 | 光芯片 | +0.77% | +89.7% | 139.1 |
| 长光华芯 | 光芯片 | +3.36% | +23.3% | 515.2 |
年内涨得最多的是源杰+89.7%、光迅+76.7%，模块厂旭创只有+4.6%、新易盛-16.5%。钱今年先给了上游，模块厂一直在追。
市盈率一栏是今天收盘总市值除以最近四个季度的归母净利润；年内涨幅已经还原送转，天孚和新易盛今年都做过送转，直接拿现价跟年初价比会差出一大截。
所谓"模块厂吃饱、上游没跟上"，是把一天的分时图当成了全年的定价。10月28日不过是追赶急了一点。
上游的钱卡在它自己的上游
天孚最典型。2026年上半年，无源光器件收入15.30亿、同比+77.40%、毛利率71.90%；有源光器件收入12.60亿、同比-19.57%、毛利率47.03%。
有源这块去年同期是15.7亿、占收入六成四，今年掉到四成五。去年那两个数（8.63亿、15.67亿）是按这份半年报自己写的+77.40%和-19.57%反推的。
半年报里写得很直白：
报告期内个别物料供给紧缺对公司有源光器件产品提产带来一定影响。
就是这一句。天孚的有源业务要外购芯片和核心物料，料不够，产线就开不满。旭创的订单下到2027年，传到天孚这里变成想交但交不出。
整个上半年天孚收入28.28亿、同比+15.15%，同期旭创417.78亿、同比+182.49%。同一句"订单饱满"，落在两家报表上是15%和182%的差别——不是需求没传过去，是产能接不住。
三季报把这个缺口补回来一部分：单季收入19.23亿、同比+31.50%，归母净利8.79亿、同比+55.30%，单季毛利率65.05%，比二季度的64.66%还高。
但这份三季报的公告日期是10月29日，10月28日收盘时市场手上还没有它。
报表里另有两个数能证明是在抢料：预付款0.92亿，比年初+343.75%；存货12.74亿，比年初+178.65%，公司自己写的原因是"战略性备货带来的原材料增加"。抢料说明产能是瓶颈，不是订单。
仕佳卖的其实是跳线
仕佳被叫"光芯片"，但翻开半年报的收入拆分，主营14.78亿里，光纤连接器7.62亿、占51.5%；AWG 3.83亿；室内光缆0.98亿；线缆材料1.37亿；真正叫芯片的DFB只有0.49亿、占3.3%，PLC 0.25亿、占1.7%。
半年报对自家芯片的定位也写清楚了：DFB"已在接入网领域实现稳定批量供货"，数据中心硅光用的连续波DFB"部分产品已实现批量供货"，EML"目前正处于客户验证阶段"。1.6T模块里最缺的那颗高速激光器芯片，它还没开始卖。
它的三季报很猛：单季收入12.27亿、同比+116.06%，归母净利2.71亿、同比+226.19%，比天孚的+31.50%快得多。但基数摆在那里——单季12亿对旭创的223亿。
现金也回得慢：前三季经营现金流-4.39亿，净利5.86亿，应收11.78亿比年初+128%，存货10.10亿比年初翻倍。公司给的原因是"为满足产品订单需求的库存备货增加；营业收入增长快，销售商品部分账期未到"——备货是为订单备的，但钱确实还在客户和仓库里。
模块厂把上游的价锁住了
旭创10月23日那份公告里有两句话，一句讲锁量，一句讲锁价：
公司将通过提供多元化产品方案，与核心供应商战略绑定、签订长期供应协议等措施持续增强供应链保障能力。
目前下游需求快速增长，具备大规模交付能力的厂商有限，市场价格体系健康，给客户的产品报价相对理性。
翻译过来：供给紧的时候，模块厂用长协把上游的量和价一起锁了。上游拿到的是锁价长单，量随模块厂出货线性增长，但没有涨价弹性。模块厂拿去卖给云厂商，报价"理性"，毛利率就归自己。
旭创自己的三季报今天才上董事会，最近一份完整财报是8月22日的半年报。它那36.6倍市盈率的分母还是旧的；如果三季度利润继续环比增长，报表一换，这个倍数会明显往下走。这也是资金敢在报表出来之前先买的理由。
另一个佐证是囤料的位置。旭创上半年末存货198.26亿，2025年三季度末是109.02亿，涨了81.9%；预付款从去年三季度末的0.80亿涨到9.54亿。整条链的料是模块厂在囤，排产也是它定，上游按它的节奏交货。
再看一层结构：天孚有源毛利率47.03%、无源71.90%，中间差24.9个百分点。有源占比往上走，本身就会拖住混合毛利率；三季度在有源修复的同时，混合毛利率还能从64.66%升到65.05%，说明无源那块的价格和结构没有坏。模块厂没有这个包袱，它卖的就是模块本身。
反方最强的一击
天孚三季报毛利率65.05%高于二季度，有源在修复；CPO配套的FAU、ELS已经稳定量产，公司说CPO产品2027年起逐步提量。如果四季度有源继续放量，上游这一段报表会补上，股价也会补，那就只是时间差，不是逻辑差。
这一条我认一半。时间差是真的，但钱已经先付了：天孚83.4倍、仕佳78.9倍，旭创36.6倍。
天孚还给了个坐标——股权激励考核2027、2028、2029年净利润比2025年增长150%、400%、700%，按2025年归母20.17亿折算，2027年要50.4亿、2029年要161.4亿。目标定得这么高，另一面就是：任何一个季度差一点，杀的是估值，不是业绩。
什么情况下这套判断就错了
- 旭创今天（10月29日）开董事会审三季报，单季毛利率若跌破45%（二季度46.42%），说明模块厂也开始被压价，"订单饱就等于模块厂赚钱"这条就不成立。
- 天孚四季度单季收入若不到22亿（比三季度的19.23亿只增14%），说明物料缓解是一次性的，上游的产能故事要重估。
- 仕佳光纤连接器占比继续升、DFB还停在0.5亿量级，那它就是一家连接器公司，78.9倍市盈率没有对应的业务。
这条链的真实顺序是：芯片先紧、器件跟着紧、模块最后放量；股价也是芯片先涨、器件跟上、模块补涨。10月28日发生的是最后一段。
下一段看什么，不是看订单饱不饱——旭创的订单已经排到2027年，这个信息早就是公开的——而是看天孚的有源能不能连着两个季度环比放量，把"接不住"变成"接得住"。
来源：中际旭创《关于公司近期生产经营情况的公告》2026-10-23（2026-100）；中际旭创《2026年半年度报告》2026-08-22（2026-081）；中际旭创《H股公告（董事会会议日期）》2026-10-15；天孚通信《关于公司近期生产经营情况的公告》2026-10-23（2026-057）；天孚通信《2026年半年度报告》2026-08-19（2026-040）；天孚通信《2026年第三季度报告》2026-10-29（2026-064）；天孚通信《2026年10月28日投资者关系活动记录表》（2026-012）；仕佳光子《2026年半年度报告》2026-08-01；仕佳光子《2026年第三季度报告》2026-10-22。
10月28日那3.5个点的差距，其实到29日开盘就已经作废了。天孚28日晚上19:00到19:45开了一场线上通讯会，会上直接把三季度报了：收入19.23亿、同比增31.50%，净利8.79亿、同比增55.30%。正式的三季报文本29日才挂网，数字在前一晚就公开了。
"模块厂吃饱、上游没跟上"这句话的有效期只有四个小时，从15:00收盘到19:00开会。
## 那天比的是谁先说
28日盘中，这条链上手里捏着三季度新数的只有两家：仕佳10月22日披露，光迅10月26日披露。长光华芯是28日当天，天孚要等收盘之后，旭创要等29日董事会。
四家公司处在四个不同的披露阶段，涨跌幅却被放进同一张表里比。这天比的不是基本面，是信息的新鲜度。
旭创那天能涨10.28%，手里多一样东西：10月23日那份生产经营公告——订单能见度、供应链怎么绑定，都写在纸上了，而且还没被三季度报表检验过。资金买的是"已经写下来、还没被算进价里"的东西，这是最省力的一笔。
天孚不在这个位置。它的新数要等到晚上七点才开始讲。
## 一个模块的钱，上游只拿得到四分之一
旭创二季度单季收入比一季度多27.85亿。天孚三季度环比多4.25亿，仕佳多3.08亿——两家加起来7.33亿，是旭创增量的26%，而且还晚了一个季度。
同一个订单，模块厂报表上多100元，上游只能多26元。
体量上更直白：上游三家（天孚、仕佳、太辰光）上半年收入合计53.2亿，模块两家（旭创、新易盛）626.9亿，前者是后者的8.5%。旭创一个季度222.8亿，是天孚整季19.2亿的11.6倍。
所以"订单吃饱"传不到上游报表上。不是上游没受益，是它在一个模块里的价值量本来就这么大。模块厂多卖一倍，上游最多多卖四分之一，还要晚三个月才落进利润表。
对股价的含义很直接：同一条"订单又超预期"的消息，落在模块厂报表上是全额，落在上游只剩四分之一。
那上游到底靠什么赚钱？把一份单季毛利拆开就清楚了。天孚三季度单季毛利额12.51亿，二季度9.68亿，多了2.82亿。
拆开看：收入多了4.25亿，按二季度64.66%的毛利率算，这部分多带来2.75亿；毛利率抬了0.38个百分点，只多带来0.07亿。
2.82亿的毛利增量里，97%是"多交了货"，3%是"卖得更贵"。
仕佳是同一个形状：毛利增量1.48亿，其中1.15亿来自多卖，0.33亿来自毛利率从37.47%抬到40.13%。
```chart
type: group
title: 三季度毛利增量，多半是多交的货挣来的
unit: 亿
labels: 天孚, 仕佳
values: 2.75, 1.15
values2: 0.07, 0.33
legend: 多交的货|毛利率抬升
```
这两组数合起来是一句话：模块厂锁住的是涨价的时点，不是价格的水平。上游拿到的是"量随你出货走、价不因紧缺上浮"的长单，它在报表上的指纹就是——毛利增量几乎全部由量解释，毛利率只靠产品结构和良率一点点往上蹭。
## 议价权不在位置，在独家
把二季度单季拆开横着排，台阶不是按"越上游越高"来的。
| 公司 | 环节 | 单季收入 | 单季毛利率 | 环比Q1 |
|---|---|---|---|---|
| 源杰科技 | 光芯片 | 5.70 | 82.1 | 4.3 |
| 天孚通信 | 光器件 | 14.98 | 64.7 | 8.1 |
| 新易盛 | 光模块 | 125.72 | 48.0 | (1.2) |
| 中际旭创 | 光模块 | 222.81 | 46.4 | 0.4 |
| 仕佳光子 | 连接与芯片 | 9.19 | 37.5 | 3.3 |
| 太辰光 | 光连接 | 6.60 | 33.3 | (4.5) |
| 光迅科技 | 器件加模块 | 38.36 | 24.6 | (2.3) |
| 长光华芯 | 激光芯片 | 2.61 | 22.6 | (7.5) |
最后一列是毛利率相对一季度的变化，单位是百分点；收入单位是亿元。
最高的源杰82.1%，是链条上最紧的那一环：上半年数据中心收入7.74亿、同比增640.29%，占总收入83.7%，70mW和100mW的CW光源在批量出货。
最低的两家里，一家是"光芯片"长光华芯（22.6%），一家是"器件带模块什么都有"的光迅（24.6%）。
位置在这张表里没有解释力。解释力在"这一环里你有没有替代品"。
光迅是最清楚的反面教材：芯片、器件、模块它都能做，每一段都进得去，每一段都不是独家。代价是存货121.10亿（年初57.46亿，增110.76%）、预付9.76亿（增818.72%）、前三季经营现金流-18.81亿，而净利是正的12.84亿。
什么都有，等于什么都不够。
标签也会骗人。28日长光华芯只涨3.36%，很容易被读成"光芯片落队"。
它三季度单季收入2.42亿、同比增92.86%，扣非净利0.13亿，毛利率24.5%。半年报写得很清楚，主业是半导体激光芯片——高功率单管、高功率巴条、VCSEL，用在激光加工、泵浦源、激光雷达，光通信芯片只是其中一条产品线。
源杰做的是数通：CW光源、EML，客户是光模块厂。
两家都挂着"光芯片"三个字，毛利率差57个百分点，验证周期、客户结构、周期位置全都不同。把它们塞进同一张涨幅表里比，结论必然是错的。
## 全链同时垫钱，这是供给约束的形状
旭创上半年净利136.51亿，经营现金流只有18.00亿，同比减44.08%。应收账款150.01亿（年初62.77亿），存货198.26亿（年初126.81亿），预付9.54亿（年初1.34亿），而合同负债只有0.21亿——客户不预付一分钱。
收现比也分得清：天孚上半年销售收现26.64亿对收入28.28亿，94.2%；旭创328.52亿对417.78亿，78.6%。上游交货基本就收钱，模块厂每100元收入里有21元还挂在账上。
仕佳前三季经营现金流-4.39亿，光迅-18.81亿。四家公司、四种生意，同一个动作：存货和预付同时大增、现金流追不上净利、收入还在加速。
这是供给约束下的备料形态。需求见顶的指纹正好相反——收入先减速，存货和应收后堆。这两件事在报表上的先后顺序完全不同，很容易看混。
## 反方最强的一击
有人会说：上游毛利率全在往上走，天孚从一季度的56.60%升到二季度的64.66%、三季度再抬0.38个百分点，仕佳从34.13%到37.47%再到40.13%，这叫被锁价？
这条我认。价格确实没有被压下去。
但它也没有往上走。天孚的毛利率抬升来自无源占比和良率，仕佳来自产品结构——两样都有天花板，而且都不写在合同价里。单价的贡献，在上面那个拆解里只占3%和22%。
准确的说法是：模块厂锁的是量的确定性，上游让出的是涨价的弹性。上游换到的是产能利用率，让出去的是周期顶部的超额利润。
还有一层更硬的反对：上游的好，市场早算过账了。天孚的市盈率是旭创的2.3倍。同样一句"订单饱满"，一个按36倍卖，一个按83倍卖。差距不在28日那3.5个点里，在过去十一个月的定价里。
## 什么情况下这套判断就错了
- 天孚四季度单季毛利率跌破63%、收入还在增，说明量增价跌，"锁的是时点"这个修正被推翻，变成真的压价。
- 旭创三季度存货再增50亿以上、单季收入环比增速掉到10%以下，备货就从主动变成积压。
- 光迅四季度存货不再增、经营现金流转正，说明"什么都有"开始有议价权，"卡点才定价"这条要改。
把这三条倒过来看，就是接下来该盯的东西。
28日发生的事一句话能说完：市场先给了"看得见订单"的那一头，后给了"还没报数"的这一头。
29日开盘起，天孚这个信息差就没有了，数字已经在桌上；旭创的三季报最快也要等29日董事会开完。
接下来的定价不看成交额，看交付：上游能不能连着两个季度把量交出来，毛利率还能不能靠结构撑住，现金流能不能追上净利。
这三件只要兑现两件，"上游被压"这套说法就该丢掉；一件都不兑现，83倍和79倍就是纯粹的预付。
来源：天孚通信《投资者关系活动记录表》2026-012（2026-10-28）；《2026年第三季度报告》2026-064（2026-10-29）；《2026年半年度报告》2026-040（2026-08-19）。中际旭创《关于公司近期生产经营情况的公告》2026-100（2026-10-23）；《2026年半年度报告》2026-081（2026-08-22）。仕佳光子《2026年第三季度报告》（2026-10-22）；《2026年半年度报告》（2026-08-01）。新易盛《2026年半年度报告》（2026-08-25）。太辰光《2026年半年度报告》（2026-08-15）。源杰科技《2026年半年度报告》（2026-08-29）。长光华芯《2026年第三季度报告》（2026-10-28）；《2026年半年度报告》（2026-08-31）。光迅科技《2026年第三季度报告》（公告编号2026-061，2026-10-26）。单季数据均由本期累计减上期累计得出，一季报直接取累计值；毛利率为营业收入减营业成本后除以营业收入。