Fabrinet (FN) rose 12.1% in one day. In its fiscal 2027 second-quarter 10-Q, for optical communications revenue tiered by speed, if the 800G-and-above tier no longer rises quarter-on-quarter, will Zhongji Innolight's optical module gross margin in its 2026 annual report decline accordingly?
- The three individual quarters were 46.06%, 46.42%, 46.61% in order.
- Same demand, same period of time, one +78%, one +172%, a 2.2x difference.
- Full-year 2025 overall gross margin was only 42.04%.
- Market data, Nasdaq API, 2026-11-06 intraday: Fabrinet 52-week range 286.07 to 748.89 USD
- On November 5 it fell 14.25% to close at 292.97 dollars
Fabrinet (FN) rose 12.1% in one day. In its fiscal 2027 second-quarter 10-Q, for optical communications revenue tiered by speed, if the 800G-and-above tier no longer rises quarter-on-quarter, will Zhongji Innolight's optical module gross margin in its 2026 annual report decline accordingly?
Fabrinet(FN) 一天涨 12.1%。它 2027 财年第二季 10-Q 里按速度分档的光通信收入，800G 以上那一档如果不再环比上升，中际旭创 2026 年报的光模块毛利率会跟着往下走吗？
The desk answersverifiedWalnut, from public filings
No. Comparing year-on-year with 2025, it definitely will not go down; comparing with the level of the first three quarters, it is a 21-basis-point small bet, and I bet it will not. But more importantly: the document this question asks about does not yet exist.
Fabrinet's 10-Q for the second quarter of fiscal 2027 will not arrive until February 2027; the same-period filing for the prior fiscal year was submitted on February 3, 2026. The company itself, in its November 2 earnings release, wrote the quarter's cutoff date as December 25, 2026—a quarter that has not yet finished cannot have a quarterly report. The latest quarterly report currently available is the fiscal 2027 first quarter filed on November 3, with the reporting period ended September 25.
Even if that 10-Q comes out, "optical module revenue split by speed tier" will not be there. I ran a full-text search across all of Fabrinet's historical SEC filings: "800G" hit 0 times, and "400G" also 0 times. Its revenue is split only along two dimensions: geography (Thailand, United States, Singapore, China...), and three end markets (data center, communications infrastructure, automotive industrial and other). "The tier above 800G" is not something not disclosed this quarter—it has never appeared in this company's filing system. So this question cannot be answered as posed—it is not that I will not answer, it is that the number is not in the filings.
Taking a further step back, even "quarter-on-quarter" cannot be calculated: the new classification was only adopted in fiscal 2026 fourth quarter; the 10-K only restated annual figures, and the quarterly reports for the earlier quarters still use the old classification, so the data center tier has no comparable prior quarter.
By the way, correcting an error: the CIK 1477331 you gave is not Fabrinet. On SEC, this number belongs to a golf course in Washington State called Potholes Reservoir Golf Resort; Fabrinet is 1408710. If you pull by 1477331, you are pulling someone else's filings.
Then use the closest thing available to look at one quarter. Fiscal 2027 first quarter is July to September: Fabrinet data center revenue 7.77 hundred million USD, year-on-year +77.6%; communications infrastructure 4.44 hundred million USD, year-on-year +34.7%; overall 14.74 hundred million USD, year-on-year +50.7%, 3.5% above the upper end of the company's own guidance of 13.75 to 14.25 hundred million. In the same quarter, Zhongji Innolight's single-quarter revenue was 277.69 yi yuan, year-on-year +171.8%. Same demand, same period of time, one +78%, one +172%, a 2.2x difference.
Going one level further down, what Fabrinet assembles cannot change the gross margin of what Innolight sells.
The businesses are not the same kind. Fabrinet's gross margin this quarter was 11.79%, versus 11.90% in the same period last year, actually a bit lower. Over a longer stretch, its gross margin over the past ten quarters has all been between 11.72% and 12.43%; over these three years revenue grew 2.15x, and gross margin did not move. It collects processing fees, assembling customers' materials into finished products; swap what is assembled, and the percentage moves only a few basis points. Innolight sells its own products; in the first three quarters of 2026 its gross margin was 46.39%, and gross margin is selling price minus materials. Which tier Fabrinet assembles cannot change Innolight's selling price, nor can it change Innolight's material cost.
The position is also different. Fabrinet does it for others; Innolight does it itself and also takes orders. That 2.2x growth gap would not appear if high-end really were flattening at the industry level—what the gap reflects is share, not tier. Using one company's product mix to infer another company's price, even the direction could be reversed.
In timing, it is even less "leading." Fabrinet's December quarter is October to December, and Innolight's 2026 fourth quarter is also October to December; the two periods almost overlap. So even if that 10-Q had numbers, it would only be another perspective on the same period of time, not a signal several months ahead. Using it to predict Innolight's gross margin is like using the weather of the same period to infer the temperature of the same period.
On Innolight's side, the annual figures are in fact already 70% locked in by the first three quarters. First three quarters revenue 695.47 yi yuan; estimating the fourth quarter as flat with the third quarter, full year 973.16 yi yuan (estimate), with the first three quarters accounting for 71.5% and the fourth quarter only 28.5%. First three quarters gross margin 46.39% (Innolight does not disclose single-quarter gross margin by product; overall is used instead; optical modules account for 99% of revenue, and in the first half the two differed by 34 basis points); the three individual quarters were 46.06%, 46.42%, 46.61% in order. For the full year to be below the first three quarters' 46.39%, the fourth quarter only needs to fall below 46.39%—between the third quarter and this line there are only 21 basis points. And for every 100 basis points the fourth quarter falls, the full year falls only 28.5 basis points (estimate), transmitting less than 30%.
By product it is clearer. For full-year 2025, the "optical communication transceiver modules" line had a gross margin of 42.61%; for the first half of 2026, this line was 46.59%, 3.98 percentage points higher than last full year and 6.63 percentage points higher than last year's first half (39.96%). Full-year 2025 overall gross margin was only 42.04%. So the statement "the optical module gross margin will trend down in the 2026 annual report" does not hold relative to 2025 — the first half was already nearly four points above last full year, and the second half would have to fall to 39% to 40% (estimate) to catch back down to last year's level. The only thing genuinely still hanging is one question: whether the fourth quarter alone will break below 46.39% and drag the full year down from the first-three-quarters level. That is a 21-basis-point bet.
That +12.1% should not be taken as proof of demand either. Fabrinet's 52-week high was 748.89 dollars (May 14), the low was 286.07 dollars (intraday November 5), and on November 6 it was around 328 dollars intraday, still 56% below that May high. On November 5 it fell 14.25% to close at 292.97 dollars; the same day COHR fell 12.98%, LITE fell 18.68%, AAOI fell 20.89%, CIEN fell 15.22%, MRVL fell 16.43%, and Nvidia fell 9.26%, and the next day everything rebounded. So the 12.1% was one day's bounce off a one-year low, not a new high. On the A-share side the same day was the opposite: Dongshan Precision -10% limit down, with 123 yi in turnover. The two sides gave opposite answers on the same chain.
What can really press down Innolight's gross margin is price, not Fabrinet's tier. Innolight does not disclose shipment volume or unit price, and the statements cannot be broken down into "how much selling price fell, how much cost fell." There is one visible trace: from 36.70% in the first quarter of 2025 all the way up to 46.61% in the third quarter of 2026, six quarter-on-quarter increases in six quarters, with increases of +479, +130, +169, +158, +36, +19 basis points in sequence, the last one the smallest. The slope is already flattening; this has nothing to do with which tier Fabrinet is on, it is the module's own selling-price problem.
The strongest opposing argument: Fabrinet's tier is indeed not a direct variable of Innolight's gross margin, but it may be the first visible reading of "when this round of shortage ends," and Innolight's six quarters of gross margin increases were propped up by shortage pricing in the first place. Once high-end volume at the assembly layer flattens in the fourth quarter, it means supply has caught up with demand, and at the next annual price negotiation module makers will have to cut prices; while lasers, DSP and other genuinely tight materials fall in price slowly. Price falls first, materials fall later, and gross margin gets squeezed from both ends. This argument also has data behind it: Innolight's third-quarter gross margin rose only 19 basis points quarter-on-quarter, the slope is already flattening; on November 5 the whole sector fell 13% to 21%, showing someone is already betting on this. So the conclusion "it cannot transmit through" will fail during the period when the shortage genuinely turns, and cannot be treated as a permanent talisman.
What would count as me being wrong: in Innolight's 2026 annual report (disclosed only in 2027), the gross margin of the "optical communication transceiver modules" line is below 46.39%, meaning the fourth quarter alone broke below that line and dragged down the full year; and at the same time Fabrinet's December quarter (10-Q filed in February 2027) data center segment revenue shows zero or negative quarter-on-quarter growth. If both appear together, it means volume turned first and price turned later, the order lines up, and my claim that "Fabrinet's tier cannot transmit through to Innolight's gross margin" is wrong. Either one alone does not count — the full year being dragged down could also be Innolight's own pricing strategy, and Fabrinet flattening could also be share being moved away.
Sources:
- Fabrinet, 8-K, 2026-11-02, accession 0001408710-26-000034, exhibit EX-99.1 filename fn-20261102xex991q127.htm (fiscal 2027 first quarter results, second quarter guidance, second quarter "as of December 25, 2026", first quarter original guidance 13.75 to 14.25 yi USD).
- Fabrinet, 10-Q, filed 2026-11-03, reporting period 2026-09-25, accession 0001408710-26-000041, main document fn-20260925.htm (Note 3 revenue table, income statement, balance sheet inventory 11.93 yi USD).
- Fabrinet, 10-K, filed 2026-08-18, fiscal year ended 2026-06-26, accession 0001408710-26-000028, main document fn-20260626.htm (revenue table for three end markets, fiscal years 2024 to 2026).
- Fabrinet, 8-K, 2026-08-17, accession 0001408710-26-000026, exhibit EX-99.1 filename fn-2026811xex991q426.htm (fourth quarter results and first quarter original guidance).
- SEC EDGAR full-text search, queried 2026-11-07, conditions ciks=0001408710, keywords "800G" and "400G", both returned 0 documents.
- SEC submissions API, queried 2026-11-07: CIK 0001408710 is Fabrinet, 0001477331 is Potholes Reservoir Golf Resort.
- Zhongji Innolight, 2026 third quarter report, 2026 half-year report, 2025 annual report (income statement and main business composition, consolidated figures; tushare 300308.SZ, 2026-11-07).
- Market data, Nasdaq API, 2026-11-06 intraday: Fabrinet 52-week range 286.07 to 748.89 USD; November 5 COHR, LITE, AAOI, CIEN, MRVL, NVIDIA closing prices from the same source.
不会。按年和 2025 年比，一定不会往下走；按跟前三季的水平比，是 21 个基点的小赌，我押它不会。但更要紧的是：这个问题问的那份文件，现在还不存在。
Fabrinet 2027 财年第二季的 10-Q 得等到 2027 年 2 月，上一财年同期那份是 2026 年 2 月 3 日交的。公司自己在 11 月 2 日的业绩稿里把这个季度的截止日写成 2026 年 12 月 25 日——一个还没走完的季度，不可能有季报。目前最新的季报是 11 月 3 日提交的 2027 财年第一季，报告期截至 9 月 25 日。
就算那份 10-Q 出来了，"按速度分档的光模块收入"也不会有。我把 Fabrinet 在 SEC 的全部历史申报做了一遍全文检索，"800G"命中 0 次，"400G"也是 0 次。它的收入只按两个方向拆：地区（泰国、美国、新加坡、中国……），和三个终端市场（数据中心、通信基础设施、汽车工业及其他）。"800G 以上那一档"不是这一季没披露，是这家公司的文件体系里从来就没出现过。所以这个问题没法按原样回答——不是我不答，是数不在文件里。
再退一步，连"环比"都算不出来：新分类是 2026 财年第四季才启用的，10-K 只重述了年度数，前面几季的季报还是老分类，数据中心这一档没有可对照的上季度。
顺手纠个错：你给的 CIK 1477331 不是 Fabrinet。SEC 上这个号属于华盛顿州一家叫 Potholes Reservoir Golf Resort 的高尔夫球场，Fabrinet 是 1408710。照 1477331 去拉，拉的是别人的文件。
那用拿得到的最接近的东西看一季。2027 财年第一季就是 7 月到 9 月：Fabrinet 数据中心收入 7.77 亿美元、同比 +77.6%，通信基础设施 4.44 亿美元、同比 +34.7%，整体 14.74 亿美元、同比 +50.7%，比公司自己 13.75 到 14.25 亿的指引上限还高出 3.5%。同一个季度，中际旭创单季收入 277.69 亿元、同比 +171.8%。同一条需求、同一段时间，一个 +78%，一个 +172%，差 2.2 倍。
再往下一层，Fabrinet 装什么，改不了旭创卖什么的毛利率。
生意就不是一种。Fabrinet 这一季毛利率 11.79%，去年同期 11.90%，还低了一点。拉长看，过去十个季度它的毛利率都在 11.72% 到 12.43% 之间，这三年收入翻了 2.15 倍，毛利率没动过。它收的是加工费，把客户的料装成成品，装的东西换一换，百分比只挪几个基点。旭创是自己卖产品，2026 年前三季毛利率 46.39%，毛利率就是售价减物料。Fabrinet 装的是哪一档，改不了旭创的售价，也改不了旭创的物料成本。
位置也不一样。Fabrinet 是替人做，旭创是自己做还接单。那个 2.2 倍的增速差，如果高端真在行业层面转平就不会出现——差出来的是份额，不是档位。拿一家的产品结构去推另一家的价格，方向都可能是反的。
时间上更不是"领先"。Fabrinet 的 12 月季是 10 月到 12 月，旭创的 2026 年第四季也是 10 月到 12 月，两段几乎是重合的。所以就算那份 10-Q 有数，也只是同一段时间里的另一个视角，不是提前几个月的信号。用它预判旭创的毛利率，等于用同期的天气推同期的温度。
旭创这边的年度数字，其实已经被前三季锁掉七成。前三季收入 695.47 亿元，第四季按与第三季持平估算，全年 973.16 亿元（估算），前三季占 71.5%，第四季只占 28.5%。前三季毛利率 46.39%（旭创不披露单季分产品毛利率，用整体代替；光模块占收入 99%，上半年两者差 34 个基点），三个单季依次是 46.06%、46.42%、46.61%。全年要低于前三季那个 46.39%，只需要第四季单季掉到 46.39% 以下——第三季到这条线之间只有 21 个基点。而第四季每掉 100 个基点，全年只掉 28.5 个基点（估算），传导不到三成。
按产品看更清楚。2025 年全年"光通信收发模块"这一行毛利率 42.61%，2026 年上半年这一行是 46.59%，比去年全年高 3.98 个百分点，比去年上半年（39.96%）高 6.63 个百分点。2025 年全年整体毛利率只有 42.04%。所以"2026 年报的光模块毛利率往下走"这句话，跟 2025 年比是不成立的——上半年就已经比去年全年高出近四个点，下半年要掉到 39% 到 40%（估算）才追得回去年那个水平。真正还悬着的只有一件事：第四季单季会不会跌破 46.39%，把全年从前三季那个水平上拉下来。这是 21 个基点的赌注。
那个 +12.1% 也别当成需求被证实。Fabrinet 52 周最高 748.89 美元（5 月 14 日），最低 286.07 美元（11 月 5 日盘中），11 月 6 日盘中 328 美元附近，离 5 月那个高点还低 56%。11 月 5 日那天它跌 14.25% 收在 292.97 美元，同一天 COHR 跌 12.98%、LITE 跌 18.68%、AAOI 跌 20.89%、CIEN 跌 15.22%、MRVL 跌 16.43%、英伟达跌 9.26%，第二天全体反弹。所以 12.1% 是从一年最低点上弹回来的一天，不是新高。A 股这边同一天是反的：东山精密 -10% 跌停、成交 123 亿元。两边在同一条链上给出相反的答案。
真正能压住旭创毛利率的东西是价格，不是 Fabrinet 的档位。旭创不披露出货量和单价，报表上拆不出"售价降了多少、成本降了多少"。有一条看得见的痕：从 2025 年第一季的 36.70% 一路涨到 2026 年第三季的 46.61%，六个季度六次环比上涨，涨幅依次是 +479、+130、+169、+158、+36、+19 个基点，末一次最小。斜率已经在平，这跟 Fabrinet 装哪一档没关系，是模块自己的卖价问题。
反方最强的一条：Fabrinet 的档位确实不是旭创毛利率的直接变量，但它可能是"这轮紧缺什么时候结束"的第一个可见读数，而旭创这六个季度的毛利率上涨本来就是紧缺价在托着。组装层的高端量一旦在第四季转平，说明供给追上了需求，下一轮年度议价模块厂要降价；而激光器、DSP 这些真正紧的料降价慢。价先掉、料后掉，毛利率就是被两头夹下来的。这条还有数据撑腰：旭创第三季毛利率环比只加 19 个基点，斜率已经在平；11 月 5 日整个板块跌 13% 到 21%，说明已经有人在为这件事下注。所以"传不过来"这个结论，在紧缺真正转向的那一段会失效，不能当成永远的护身符。
什么情况算我看错：旭创 2026 年年报（2027 年才披露）里，"光通信收发模块"这一行的毛利率低于 46.39%，也就是第四季单季跌破这条线、把全年拉了下来；同时 Fabrinet 12 月季（2027 年 2 月提交的 10-Q）数据中心分部收入环比零增长或负增长。这两个同时出现，就说明量先转、价后转，顺序对上了，我这条"Fabrinet 的档位传不到旭创的毛利率"就是错的。单独一个出现不算——全年被拉低也可能是旭创自己的价格策略，Fabrinet 转平也可能是份额挪走。
- 01Fabrinet，8-K，2026-11-02，accession 0001408710-26-000034，附件 EX-99.1 文件名 fn-20261102xex991q127.htm（2027 财年第一季业绩、第二季指引、第二季"截至 2026 年 12 月 25 日"、第一季原指引 13.75 到 14.25 亿美元）
- 02Fabrinet，10-Q，2026-11-03 提交、报告期 2026-09-25，accession 0001408710-26-000041，正文 fn-20260925.htm（附注 3 收入表、利润表、资产负债表存货 11.93 亿美元）
- 03Fabrinet，10-K，2026-08-18 提交、财年截至 2026-06-26，accession 0001408710-26-000028，正文 fn-20260626.htm（三个终端市场 2024 至 2026 年度收入表）
- 04Fabrinet，8-K，2026-08-17，accession 0001408710-26-000026，附件 EX-99.1 文件名 fn-2026811xex991q426.htm（第四季业绩与第一季原指引）
- 05SEC EDGAR 全文检索，2026-11-07 查询，条件 ciks=0001408710、关键词"800G"与"400G"，两条都命中 0 份文件
- 06SEC submissions 接口，2026-11-07 查询：CIK 0001408710 为 Fabrinet，0001477331 为 Potholes Reservoir Golf Resort
- 07中际旭创，2026 年第三季度报告、2026 年半年度报告、2025 年年度报告（利润表与主营业务构成，合并数
- 08tushare 300308.SZ，2026-11-07）
- 09行情数据，Nasdaq 接口，2026-11-06 盘中：Fabrinet 52 周区间 286.07 至 748.89 美元
- 1011 月 5 日 COHR、LITE、AAOI、CIEN、MRVL、英伟达收盘价同源
前提一句话交代完：Fabrinet 2027 财年第二季的 10-Q 要到 2027 年 2 月才有，这家公司全部历史申报里"800G"出现 0 次，所以"按速度分档"这个前提不成立，下面谈的是拿得到的最近一期。
## 19 个基点是摊薄出来的，不是卖价挣出来的
Q3 单季收入 277.69 亿元、营业成本 148.26 亿元、毛利 129.43 亿元，毛利率 46.61%。Q2 是 222.81、119.38、103.44、46.42%；Q1 是 194.96、105.17、89.80、46.06%。
拿环比看：Q3 收入比 Q2 涨 24.63%，成本涨 24.19%，毛利额涨 25.13%。收入比成本多涨 0.44 个百分点，这一差换成了 19 个基点的毛利率。
这不是提价的样子。真提价，毛利额增速会明显快过收入增速，这里只快 0.5 个点。Q2 那 36 个基点也是同一回事：收入 +14.29%、成本 +13.51%，差 0.78 个点。
两个季度连起来，差额从 0.78 点缩到 0.44 点。规模摊薄在减弱，这是这条毛利率线上唯一让人不安的地方——不是价掉了，是"省"快省完了。
口径交一句：旭创不披露单季分产品毛利率，也不按速率分档披露，季度只能用整体代替；今年上半年产品 46.59%、整体 46.25%，差 34 个基点，方向一致。
## 存货 239 亿、预付款 15.67 亿，这两个数才是答案
存货从 2025Q3 的 109.02 亿元涨到 2026Q3 的 238.78 亿元，一年多 1.19 倍；同期收入涨 1.78 倍。存货涨得比收入慢，所以不是积压。
真正反常的是预付款：2025Q3 只有 0.80 亿元，2026Q1 跳到 14.88 亿元，Q3 是 15.67 亿元，二十倍。旭创不披露预付给谁，但对象只可能是料、设备和产能。
15.67 亿元相当于它一个季度营业成本（148.26 亿元）的 10.6%。提前付掉一成成本的钱，为的是把明年的料和设备按住。
```chart
type: line
title: 中际旭创单季毛利率，六个季度涨完之后只剩 21 个基点余量
unit: %
labels: 2025Q1, 2025Q2, 2025Q3, 2025Q4, 2026Q1, 2026Q2, 2026Q3
values: 36.70, 41.49, 42.79, 44.48, 46.06, 46.42, 46.61
```
## 同行六家：三家在掉，三家在涨
| 公司 | 2026Q1 | 2026Q2 | 2026Q3 | 两季变动 |
|---|---|---|---|---|
| 中际旭创 | 46.06 | 46.42 | 46.61 | 55 |
| 新易盛 | 49.16 | 47.96 | 46.27 | (289) |
| 太辰光 | 37.80 | 33.27 | 26.94 | (1086) |
| 光迅科技 | 26.83 | 24.58 | 25.51 | (132) |
| 天孚通信 | 56.60 | 64.66 | 65.05 | 845 |
| 源杰科技 | 77.81 | 82.06 | 83.42 | 561 |
太辰光 Q3 单季收入 10.07 亿元、同比 +161%，毛利率从 33.27% 掉到 26.94%，两个季度掉 10.86 个点。收入翻了一倍多，毛利率掉了一成——量在涨、单位毛利在掉，这就是"800G 以上不再环比上升"落到报表上具体长什么样：不是收入不涨，是收入涨、毛利率塌。
新易盛两季掉 2.89 个点。这几家里它和旭创生意最像、体量最接近，它已经在让价了。
旭创同期还涨 55 个基点。同一段时间、同一条链、同一批料，方向不一样——所以现在决定毛利率的不是"行业处在哪一档"，是各家手里还剩多少筹码。
除旭创自己，涨的那两家都在上游：源杰 2026Q3 单季毛利率 83.42%（Q1 是 77.81%，2025 全年 58.11%），单季收入 6.84 亿元；天孚 65.05%（Q1 是 56.60%）。这两家卖的正是模块厂的料和件。
涨的是料和件，掉的是模块：利润正从模块厂这一层往上游挪。这一层里能同时守住量和价的，现在只剩旭创一家。
## Fabrinet 的档位，进不了旭创的利润表
Fabrinet 收的是加工费，这一季毛利率 11.79%，去年同期 11.90%，过去十个季度都在 11.72% 到 12.43% 之间。收入三年翻 2.15 倍，毛利率一动没动——它报价是成本加成，装什么不改百分比。
旭创 46.61%，是自己卖产品，毛利率是营业收入减营业成本再除收入。
所以 Fabrinet 的档位结构只进它自己的营业收入（加工费那一栏）。要进旭创的报表，只有一条路：终端需求强度 → 模块订单的量与价 → 旭创的营业收入。它进不了营业成本，进不了税金及附加，也进不了资产减值。落到哪个科目这件事，决定了它根本不是同一个变量。
时点上也不领先：Fabrinet 的 12 月季是 10 到 12 月，旭创的 2026 年第四季也是 10 到 12 月，同期对同期。
能进旭创营业成本的读数在别处：源杰的 83.42%、天孚的 65.05%，还有旭创自己那 15.67 亿元预付款。三个数指向同一件事——料和件在涨价，而且已经被提前锁掉。
## 反方最强的一条，我认方向不认时点
比我懂这行的人会说：旭创六个季度毛利率从 36.70% 涨到 46.61%，涨的本来就是紧缺价，不是本事。上游已经在分走议价权——源杰单季毛利率从去年同期的 61.62% 涨到 83.42%，一年涨 21.8 个点；旭创同期只涨 3.8 个点。上游涨得比下游快五倍，议价权正在往上游移。
800G 以上那一档一旦不再环比上升，1.6T 的稀缺溢价先没；而料价因为有长协和预付款是滞后的。价先掉、料后掉，两头一夹，毛利率就是被挤下来的。这条逻辑我接受，而且它比"Fabrinet 传不传导"这个说法硬得多。
我不接受的是时点。这里是我的推断：料价进营业成本要滞后一到两个季度；而 2026 年报里第四季只占 28.5%（前三季占 71.5%，按第四季与第三季持平估全年 973.16 亿元），Q4 单季每掉 100 个基点，全年只掉 28.5 个基点。
要 2026 年报往下走，得 Q4 单季从 46.61% 掉到 46.39% 以下（整体口径），只有 21 个基点的余量。换成年报里分产品的口径，这条线在 46.7% 附近，因为今年上半年产品毛利率比整体高 34 个基点。
真正该盯的是 2027 年一季报：那一季的售价是 2026 年底谈的，成本是 2026 年锁的，两头的时间差会在那张表上第一次撞上。
## 什么算我看错
- 旭创 2026 年报"光通信收发模块"毛利率低于 46.7%（对应整体 46.39%），也就是 Q4 单季跌破 46.39%。那说明价先掉、且掉得比料快，我的时点判断就是错的。
- 2027 年一季报单季毛利率环比再跌 100 个基点以上，同时源杰单季毛利率从 83.42% 回落。价跌料也跌，那是需求整体转弱，不是我说的"议价权往上游移"，因果关系换了一条。
- 反过来，如果 2027 年一季报旭创毛利率继续环比上行、新易盛和太辰光同时止跌回升，那这一轮压价只打在低端产品上，我这套"链上被夹"的框架对旭创就不适用。
来源：
1. Fabrinet，8-K，2026-11-02，accession 0001408710-26-000034，附件 EX-99.1 文件 fn-20261102xex991q127.htm（2027 财年第一季业绩、第二季指引及"截至 2026 年 12 月 25 日"、第一季原指引 13.75 到 14.25 亿美元）。
2. Fabrinet，10-Q，2026-11-03 提交、报告期 2026-09-25，accession 0001408710-26-000041，正文 fn-20260925.htm（附注 3 收入表：数据中心 777,132 千美元、通信基础设施 443,917 千美元、合计 1,474,459 千美元；利润表营业成本 1,300,661 千美元）。
3. Fabrinet，10-K，2026-08-18 提交、财年截至 2026-06-26，accession 0001408710-26-000028（三个终端市场 2024 至 2026 年度收入表）。
4. Fabrinet，8-K，2026-08-17，accession 0001408710-26-000026，附件 EX-99.1 文件 fn-2026811xex991q426.htm（2026 财年第四季业绩）。
5. Fabrinet，2027 财年第一季业绩稿与 10-Q 全文检索，"800G"与"400G"均命中 0 处（2026-11-07）。
6. 中际旭创，2026 年第三季度报告（利润表、资产负债表：存货 238.78 亿元、预付款项 15.67 亿元、合同负债 0.11 亿元）、2026 年半年度报告、2025 年年度报告与 2025 年一季报、三季报（单季数由累计相减所得）；tushare 300308.SZ 与 marketdata，2026-11-07 取数。
7. 新易盛 300502.SZ、太辰光 300570.SZ、光迅科技 002281.SZ、天孚通信 300394.SZ、源杰科技 688498.SZ 2026 年三季报及源杰 2025 年三季报（利润表营业收入与营业成本，单季数由累计相减所得），tushare 与 marketdata 接口，2026-11-07 取数。