Broadcom, Marvell, and Ciena, which report quarterly results in December for the same fiscal quarter ending at the end of October: does the gross margin stay in the hands of chipmaker Broadcom, or is it taken back by systems maker Ciena?
- Marvell 58.9% to 57.5%–58.5%
- Its own guidance for the latest quarter is a decline from 46.4% to 45.0%
- The same quarter a year earlier was 91.66 yi against 29.88 yi, 67.4%.
- Revenue was +127% year-on-year, and gross margin barely moved.
- Two customers account for 41.7% of revenue, so the bargaining position is not on its side.
Broadcom, Marvell, and Ciena, which report quarterly results in December for the same fiscal quarter ending at the end of October: does the gross margin stay in the hands of chipmaker Broadcom, or is it taken back by systems maker Ciena?
12 月出季报的博通、迈威尔和 Ciena，10 月底结束的同一个财季，毛利率是留在做芯片的博通手里，还是被做系统的 Ciena 抢回去？
The desk answersverifiedWalnut, from public filings
The conclusion first: gross margin stays in Broadcom's hands. In the quarter ending October, systems maker Ciena did not win gross margin back; its own guidance for the latest quarter is a decline from 46.4% to 45.0%; chip maker Broadcom's cell barely moved, with the semiconductor segment at 67.4% a year earlier and 67.0% this quarter. The gross margins along this chain are ladder-shaped: Broadcom chips 67.0, Marvell 58.9, Lumentum 53.2, Ciena 46.4 — the ladder has not inverted.
None of the three has reported for the quarter ending October. Broadcom's fourth fiscal quarter ends November 1, while Marvell and Ciena both end at the end of October; the latest guidance is respectively 348 yi dollars in revenue, 31.5 yi (on November 2 the range was narrowed from plus or minus 5% to plus or minus 2%, with the midpoint not raised), and 17.5 yi plus or minus 0.5 yi. The actual gross margin figures will have to wait until December.
The direction of gross margin for the three is the same: Ciena 46.4% to 45.0%; Marvell 58.9% to 57.5%–58.5%; Broadcom does not give gross margin, only an operating margin of 66%, and working backward assuming this quarter's 20.96 yi in operating expenses stays unchanged, the fourth fiscal quarter is about 72% to 73% (estimate, the company did not disclose), two to three points lower than this quarter's 75.0%.
With all three moving down together, there is no question of one taking from another. The three point four points Broadcom lost has nothing to do with Ciena: third fiscal quarter semiconductor segment revenue was 208.39 yi and cost 68.92 yi, both disclosed figures, which divide out to 67.0%; the same quarter a year earlier was 91.66 yi against 29.88 yi, 67.4%. Revenue was +127% year-on-year, and gross margin barely moved. Software was 94.2%. Overall it fell from 78.4% to 75.0%, and what fell was the weighting: chips as a share of revenue rose from 57.5% to 70.4%, chips are 27.2 points lower than software, multiplying gives about 3.5 points (estimate), which matches the actual 3.4-point decline. Within chips there is also a shift: semiconductor gross margin was still 69.5% in the second fiscal quarter and 67.0% in the third fiscal quarter; in the same quarter AI semiconductors were 167 yi, +221% year-on-year, rising from about 72% to about 80% of the segment (estimate), and next quarter they are to jump to 217 yi, +236% year-on-year.
Ciena is indeed improving: third fiscal quarter gross margin 45.4%, adjusted 46.4%, versus 41.3% and 41.9% a year earlier, up four to five points, with optical networks +46% year-on-year. Over the same period Broadcom chips went from 67.4% to 67.0%, and the gap narrowed from 25.5 points to 20.5 points (both sides exclude stock-based compensation and acquisition amortization). But its own guidance for this quarter is 45.0%, 1.4 points lower than the third fiscal quarter, with revenue still rising (quarter-on-quarter +4.7%) — of the four-plus points earned back over a year, a third is handed back first. The CFO talked about supply expansion, strengthening business fundamentals and operating leverage, and did not mention price; two customers account for 41.7% of revenue, so the bargaining position is not on its side.
The real move was on November 2: Marvell is buying Lumentum in an all-stock deal, 4.5 shares for 1 share, enterprise value 1150 yi dollars, 1132 dollars per share, 23.9% above the October 28 close before rumors. The acquired company had revenue of 12.63 yi last quarter and gross margin of 51.0% (53.2% adjusted), up 2.8 points quarter-on-quarter, with systems revenue +42.4% quarter-on-quarter. Weighted by the two companies' latest quarters, together they are 57.1%, 1.8 points lower than Marvell standalone at 58.9% (estimate). The chip maker did not go compete with systems makers for gross margin; it bought the optical layer.
The market moved in the same direction across the board: Coherent +13.58%, Fabrinet +11.66%, Applied Optoelectronics +6.91%, Ciena +5.5%, Arista +7.31%; A-shares skewed toward PCB and laser chips, with Ultrasonic Electronics +9.98%, Kinwong Electronic +7.12%, Sunda Technology +6.40%, Victory Giant Technology +5.72%, Yuanjie Technology +7.77%. What is being priced is "the optical layer repricing," not "systems makers' gross margin recovering." Yazhen Home Furnishing +10.00% and Baida Group +6.12% are unrelated to this chain.
The strongest counterargument: Ciena is not a maker that buys chipset boxes; its 10-K states its own proprietary WaveLogic coherent modem technology, the most expensive part of the system stays on its own books, and the gross margin ceiling does not depend on Broadcom's pricing; share is also moving its way, with optical networks +46% year-on-year, while Broadcom's non-AI semiconductors, backing out from the 167 yi in AI revenue, are only 41.39 yi, growing in the single digits year-on-year (estimate). Share moves first, gross margin moves later; this year's 5-point narrowing may be just the beginning.
What would count as me being wrong: when the December statements come out, if Ciena's gross margin in that final quarter is above 46.4%, and at the same time Broadcom's semiconductor segment is below 67.0%, then "staying in Broadcom's hands, Ciena did not win it back" would be wrong; the second thing is the S-4 for the Marvell and Lumentum merger—if it gives a specific calculation showing post-merger gross margin above 58.9%, rather than phrasing like "long-term improvement," then my calculated dilution of 1.8 points would be wrong.
Sources: Broadcom Inc. 8-K (2026-09-02, accession 0001730168-26-000076, avgo-08022026x8kxex99.htm); Broadcom Inc. 10-Q (2026-09-10, accession 0001730168-26-000080, segment revenue and cost tables); Broadcom Inc. 10-Q (2026-06-09, accession 0001730168-26-000054); Marvell Technology, Inc. 8-K (2026-08-27, accession 0001835632-26-000022, q227_8kx812026ex-991.htm); Marvell Technology, Inc. 8-K (2026-11-02, accession 0001193125-26-442714, d81187dex991.htm); Ciena Corp. 8-K (2026-09-03, accession 0001628280-26-060245, ex9912026q3earningspressre.htm); Ciena Corp. 10-K (2025-12-12, accession 0001628280-25-056698); Lumentum Holdings Inc. 8-K (2026-11-02, accession 0001628280-26-068655, lite_ex991xq1fy27.htm).
结论先给：毛利率留在博通手里。10 月底那季，做系统的 Ciena 没把毛利率抢回去，它自己给的最后那季指引是从 46.4% 降到 45.0%；做芯片的博通那格基本没动，半导体分部一年前 67.4%、这季 67.0%。这条链的毛利率是阶梯状的：博通芯片 67.0、迈威尔 58.9、Lumentum 53.2、Ciena 46.4，阶梯没倒过来。
三家的 10 月底那季都还没出报表。博通第四财季停在 11 月 1 日，迈威尔和 Ciena 都停在 10 月底；最新指引分别是 348 亿美元收入、31.5 亿（11 月 2 日把区间从正负 5% 收窄到正负 2%，中值没上调）、17.5 亿上下 0.5 亿。毛利率的实际数要等 12 月。
三家的毛利率方向是同一个：Ciena 46.4% 到 45.0%；迈威尔 58.9% 到 57.5% 至 58.5%；博通不给毛利率，只给营业利润率 66%，按这季 20.96 亿营业费用不动倒推，第四财季约 72% 到 73%（估算，公司没披露），比这季 75.0% 低两三个点。
三家一起往下，就不存在谁抢谁。博通掉的那三点四跟 Ciena 无关：第三财季半导体分部收入 208.39 亿、成本 68.92 亿，都是披露值，除出来 67.0%；一年前同季 91.66 亿对 29.88 亿，67.4%。收入同比 +127%，毛利率几乎没动。软件那边 94.2%。整体从 78.4% 掉到 75.0%，掉的是权重：芯片占收入从 57.5% 升到 70.4%，芯片比软件低 27.2 个点，相乘约 3.5 个点（估算），跟实际掉的 3.4 个点对得上。芯片内部也在偏：半导体毛利率第二财季还有 69.5%、第三财季 67.0%，同季 AI 半导体 167 亿、同比 +221%、占分部从约 72% 升到约 80%（估算），下一季要冲到 217 亿、同比 +236%。
Ciena 确实在改善：第三财季 45.4% 毛利、46.4% 调整后，一年前 41.3% 和 41.9%，抬了四到五个点，光网络同比 +46%。同期博通芯片从 67.4% 变 67.0%，差距从 25.5 个点缩到 20.5 个点（两边都剔了股权激励和收购摊销）。但它自己给这季 45.0%，比第三财季低 1.4 个点，收入还在往上走（环比 +4.7%）——一年挣回来的四点多，先还回去三分之一。财务官讲的是供给扩容、业务基本面转强和经营杠杆，没提价格；两个客户占收入 41.7%，议价位置不在它这边。
真正的动作在 11 月 2 日：迈威尔用全股票买 Lumentum，4.5 股换 1 股，企业价值 1150 亿美元、每股 1132 美元，比 10 月 28 日没传闻时的收盘高 23.9%。被买的这家上季收入 12.63 亿、毛利率 51.0%（调整后 53.2%），环比抬 2.8 个点，系统收入环比 +42.4%。按两家最新一季加权，合起来 57.1%，比迈威尔单独的 58.9% 低 1.8 个点（估算）。做芯片的没去跟系统厂争毛利率，它把光学那一层买下来了。
盘面整片同向：Coherent +13.58%、Fabrinet +11.66%、应用光电 +6.91%、Ciena +5.5%、Arista +7.31%；A 股偏 PCB 和激光芯片，超声电子 +9.98%、景旺电子 +7.12%、崇达技术 +6.40%、胜宏科技 +5.72%、源杰科技 +7.77%。定的是"光学这一层重新定价"，不是"系统厂毛利率修复"。亚振家居 +10.00%、百大集团 +6.12% 跟这条链无关。
反方最强的一条：Ciena 不是买芯片组盒子的厂，它 10-K 写的是自己专有的 WaveLogic 相干调制解调技术，系统里最贵那部分留在自己表上，毛利率上限不取决于博通开价；份额也在往它那边走，光网络同比 +46%，博通的非 AI 半导体按 167 亿 AI 收入反推只剩 41.39 亿、同比个位数（估算）。份额先动、毛利率后动，这一年 5 个点的收窄可能是开头。
什么情况算我看错：12 月报表出来，Ciena 最后那季毛利率高于 46.4%，同时博通半导体分部低于 67.0%，那"留在博通手里、Ciena 没抢回来"就错了；第二件是迈威尔与 Lumentum 合并的 S-4，如果给出合并后毛利率高于 58.9% 的具体测算，而不是"长期提升"这种说法，我算的摊薄 1.8 个点就是错的。
- 01Broadcom Inc. 8-K（2026-09-02，accession 0001730168-26-000076，avgo-08022026x8kxex99.htm）
- 02Broadcom Inc. 10-Q（2026-09-10，accession 0001730168-26-000080，分部收入与成本表）
- 03Broadcom Inc. 10-Q（2026-06-09，accession 0001730168-26-000054）
- 04Marvell Technology, Inc. 8-K（2026-08-27，accession 0001835632-26-000022，q227_8kx812026ex-991.htm）
- 05Marvell Technology, Inc. 8-K（2026-11-02，accession 0001193125-26-442714，d81187dex991.htm）
- 06Ciena Corp. 8-K（2026-09-03，accession 0001628280-26-060245，ex9912026q3earningspressre.htm）
- 07Ciena Corp. 10-K（2025-12-12，accession 0001628280-25-056698）
- 08Lumentum Holdings Inc. 8-K（2026-11-02，accession 0001628280-26-068655，lite_ex991xq1fy27.htm）
台阶是真的，但台阶不是议价权的刻度。67.0、58.9、53.2、46.4 看着像"越靠近芯片越赚钱"，里面混了两种口径：迈威尔的 58.9 是非 GAAP，GAAP 53.1；Lumentum 的 53.2 也是非 GAAP，GAAP 51.0。同一把尺子下只差 2.1 个点。
换一把尺子，倒挂就出来了。GAAP 下迈威尔营业利润率 16.8%，Lumentum 28.5%，毛利率低的那个反高 11.7 个点。差在费用转化：同样一块钱毛利，Lumentum 只用 11 分钱费用就把它变成营业利润，迈威尔要 22 分，Ciena 24 分——这三个是各自披露的非 GAAP 口径。
## 同一层里能差 18 个点
六家按同一个口径摆在一起，差得最大的不是跨层，是同层。Arista 和 Ciena 都做系统，一个 63.6，一个 45.4；Lumentum 和 Coherent 都做器件，一个 51.0，一个 39.1。
| 公司 | 环节 | 毛利率 | 营业利润率 |
|---|---|---|---|
| 博通（半导体） | 芯片 | 66.9 | 61.3 |
| Arista | 系统 | 63.6 | 44.9 |
| 迈威尔 | 芯片 | 53.1 | 16.8 |
| Lumentum | 器件 | 51.0 | 28.5 |
| Ciena | 系统 | 45.4 | 18.0 |
| Coherent | 器件 | 39.1 | 15.9 |
所以"做芯片还是做系统"这个问法本身就滑。Arista 在系统层，毛利率 63.6 反超做芯片的迈威尔 10 个点；Coherent 在器件层，却和 Ciena 一个水平。分水岭不是你在链条哪一格，是你卖的东西客户换掉要多久。
客户集中度也证伪了常见解释。博通一个分销商占 50% 收入，Ciena 两个客户占 41.7%；博通更集中，可它第四财季指引营业利润率与一年前持平，收入却要同比涨 93%。集中没变成压价，因为博通是被写进客户设计的定制 ASIC。
## 钱是怎么从 Ciena 走到 Lumentum 的
两家在 2026 年一季度末只差 4.4 个点，现在差 6.8 个点。
```chart
type: line
title: 两家的调整后毛利率，一个往上走，一个在让
unit: %
labels: 2026Q1, 2026Q2, 2026Q3
values: 43.5, 44.5, 46.4
values2: 47.9, 50.4, 53.2
legend: Ciena|Lumentum
```
这一季是这么拉开的。Ciena 收入指引中值环比 +4.7%，毛利额只涨 1.5%，毛利率从 46.4 降到 45.0，营业利润率从 22.5 降到 20——营业利润率降得比毛利率还多，因为费用从 4.00 亿加到 4.15 亿。按非 GAAP 算，营业成本要涨 7.5%，比收入快 2.8 个点。
Lumentum 反着来。收入环比 +25.5%、同比 +136.5%，营业成本只涨 18.4%，慢了 7.1 个点，毛利率因此从 50.4 抬到 53.2，营业利润率从 36.6 抬到 42.3。它第二财季还要再加 3.5 亿美元收入，营业利润率指引 45% 到 46%。
钱不是凭空来的。Ciena 的 10-K 里写着，截至去年 11 月 1 日有 21 亿美元采购承诺（多数不可撤销）、8.26 亿存货，对着一年 47.7 亿的收入，等于把近一半的年收入押在上游。它的 CFO 原话是"Expanding supply capacity"。锁产能要付钱，付的钱就是上游的毛利。
> Lumentum CEO 的原话："we secured our second major NPO/CPO win, backed by purchase commitments exceeding $5 billion."
它卖的已经不是标准光模块：系统类收入一个季度涨 42.4%，占比从 35.5% 升到 40.3%。器件厂把光交换直接卖进集群，中间那道系统集成被绕过去了。
## 博通为什么不在牌桌上
半导体分部这季收入 208.39 亿、成本 68.92 亿，毛利率 67.0%，一年前 67.4%；收入同比 +127% 而毛利率只动了 0.4 个点。这条腿没参与任何再分配。
所以拿博通毛利率下滑解释"Ciena 抢回来了"，方向就错了。定制 ASIC 的价格在立项时锁死，量起来结构不跟着变。芯片层内部也在分：博通守住 67.0，迈威尔从 58.9 指引到 58.0，收入 +15%、成本 +17.5%——定制 ASIC 上是挑战者，光模块上又是买方。
## 最强的一条反方
一个聪明人会说：Lumentum 那 53.2 和 42.3 不是结构，是缺货租金。三条理由都硬。
- Lumentum 自己说第二财季要把收入在一个季度里放 28%，真正的卡点不会这么扩产。
- Ciena 明说在扩供给产能，缺口正在被填。
- 同层的 Coherent 只有 39.1%，器件层内部差 11.9 个点，这个筐里大多数并不卡。
第三条最要命，它打的是我的用词：钱流向的是替换成本高的具体型号——泵浦激光器、光交换、CPO 光源——不是"器件"这个筐，Lumentum 恰好三个都占。
真要是租金，产能一释放，53% 回到 45%，迈威尔那 1150 亿就买在顶上。这个风险排不掉：Lumentum 不给分产品毛利率，结构和租金，现有报表回答不了。
## 12 月看三个数
第一，Ciena 第四财季毛利率。指引 45.0；实际站上 46.4，我这条就错。
第二，Lumentum 第二财季毛利率，不是营业利润率。它只给了 45% 到 46% 的营业利润率，毛利率一个字没给。毛利率环比掉、营业利润率靠费用撑住，租金论就成立。
第三，博通半导体分部守不守 67.0。它只给营业利润率与去年持平、不给毛利率，是三家里披露最粗的。跌破 66 才说明定制 ASIC 也被压，那才是"Ciena 抢回去"的真证据。
证伪条件一句：Ciena 站上 46.4、Lumentum 毛利率环比下滑，两条同时成立，"钱流向卡点"就错了，正确的解释是行业性降价、谁都没多赚。
这条链上真正值钱的问题不是毛利率在芯片还是在系统，是你卖的东西客户换掉要多久。博通的答案是重做一颗芯片，Lumentum 是重建一条产线，Ciena 是一个季度。
来源：Broadcom 8-K 2026-09-02（0001730168-26-000076）、10-Q 2026-09-10（0001730168-26-000080）；Marvell 8-K 08-27（0001835632-26-000022）、8-K 11-02（0001193125-26-442714）；Lumentum 8-K 11-02（0001628280-26-068655）、08-11（0001628280-26-055726）、05-05（0001628280-26-030530）；Ciena 8-K 09-03（0001628280-26-060245）、06-04（0001628280-26-040614）、03-05（0001628280-26-015006）、10-K 2025-12-12（0001628280-25-056698）；Arista 8-K 11-03（0001596532-26-000252）；Coherent 8-K 11-04（0001193125-26-447200）。