Marvell拟收购Lumentum 押注AI光互联破局铜缆物理瓶颈

Marvell拟收购Lumentum 押注AI光互联破局铜缆物理瓶颈 Marvell虽面临股权稀释重压，但补齐激光制造短板可构筑全栈光互联壁垒，其股价回踩是逢低布局AI硬件2.0周期的核心买点。 视频深入剖析了Marvell计划收购Lumentum背后的全栈战略野心：面对下一代AI超算集群中铜连接“速率翻倍、传输距离减半”的物理瓶颈，Marvell试图将自身DSP芯片和定制ASIC能力与Lumentum的激光器制造相结合，实现共封装光学（CPO）的纵向一体化。尽管博通和英伟达公开强调压榨低成本铜缆的利润空间，但英伟达分别向两家注资20亿美元的举动揭示了其对光通信路线的隐秘对冲。主讲人指出，尽管并购短期内会带来股权稀释和财务负担，但合并后的全栈优势极其显著，投资者应视短期股价回调为逢低吸纳的契机，并在AI硬件2.0时代保持分散配置。 Marvell（MRVL）计划收购光通信制造巨头 Lumentum（LITE），交易预计下周宣布；Marvell 拟借此补全激光器与共封装光学（CPO）制造能力，构建光通信全栈垂直整合闭环。 (0:00) 面对“铜退光进”的技术路线之争，英伟达采取双向对冲策略，分别向 Marvell 和 Lumentum 各注资 20 亿美元（合计 40 亿美元），兼顾当前高利润率铜缆生态与未来光纤演进。 (5:00) 铜缆物理极限日益逼近：下一代 AI 集群传输速率每翻一倍，机架内铜连接有效距离即折半，超过 3 米的传输必须转向光互连。 (22:25) 由于双方现金储备有限，Marvell 预计将通过发股与举债方式推进换股并购，短期内股权稀释风险可能带来股价调整，但长期协同效应显著。 (15:00)

主讲人 Kevin Paffrath 指出，半导体公司 Marvell 拟收购光通信企业 Lumentum 标志着 AI 数据中心互联正加速迈向光学化；尽管交易带来短期财务稀释与资产负债压力，但其旨在打破铜连接物理瓶颈，构建全栈垂直整合的硬件壁垒。 技术路线之争：铜缆物理极限与光通信演进 (0:00) 数据中心内部互连正在经历技术迭代。目前数据中心之间已普遍采用光纤，机架间连接正逐步替代铜线，而芯片间互联则逐渐逼近物理极限。当前机架内 GPU 互联仍完全依赖铜缆，但在下一代计算集群要求的高速率下，铜缆的有效传输距离无法延伸超过 3 米，传输速率每翻一倍有效距离便折半。此外，Agentic AI 对 Key-Value Cache 的消耗促使数据中心将大量 NAND SSD 独立隔离在机架外存储机房，计算集群与存储机房之间的高吞吐调度进一步加大了对高速光纤的需求。 面对互连架构演进，行业巨头产生路径分化。博通（Broadcom）占据 ASIC 市场约 60% 份额，其首席执行官倾向尽可能压榨低成本铜缆的潜力以维持高利润率，避免追逐所谓“耀眼闪烁的事物”。相比之下，占据 15% 至 25% ASIC 份额的 Marvell 则选择全力押注共封装光学（CPO）与激光互连，力图突破“铜墙”实现弯道超车。英伟达黄仁勋秉持“能用铜就用铜，万不得已才用光”的务实原则，同时向 Marvell 与 Lumentum 各注资 20 亿美元（合计 40 亿美元）以对冲技术路线风险。 拟收购 Lumentum：构建光电全栈垂直整合 (5:00) 作为顶级定制芯片设计商，Marvell 的主要营收来源包括为亚马逊定制 Trainium 芯片（占其营收约 15.5%）以及为微软研发 Maia XPU（占营收约 8.4%），并参与了谷歌 Axion CPU 与 Meta 的芯片研发。英伟达推出的 NVLink Fusion 架构允许定制芯片作为协处理器直接接入其机架系统，这为 Marvell 提供了关键的生态接入点。 为了形成全栈闭环，Marvell 计划将自身在定制 ASIC、作为光模块处理中枢的数字信号处理器（DSP）芯片以及光子交换芯片方面的优势，与 Lumentum 的物理激光发射器件（Laser Diodes）及光模块制造能力相结合。除光纤与玻璃基板依赖康宁等供应商外，该笔并购若达成，将使 Marvell 补齐从计算芯片、DSP 调制到激光发射端制造的全套光通信拼图。 并购财务冲击与标的估值基本面剖析 (15:00) 潜在交易在短期内引发了资本市场的资金担忧。市值约 2210 亿美元的 Marvell 自由现金流收益率仅约 1.68%，手头现金约 60 亿美元；而作为收购标的的 Lumentum 市值约 800 亿至 830 亿美元，账面负债 38 亿美元，现金及应收账款 34 亿美元。因缺乏充裕现金，Marvell 极需通过增发新股及举债推进换股收购，从而对现有股东权益造成稀释。消息传出后 Marvell 股价下跌，而 Lumentum 股价走高。 尽管短期面临稀释阵痛，两家公司的运营数据仍显示出强劲动能。Marvell 净营收同比增长达 36%，营业成本增速低于营收增速，展现出定价权提升。Lumentum 扣除短期地缘因素后，预期净营收可增长 1.8 倍，毛利润增长达 2.7 倍，运营支出仅增长 14%，营业利润率达到 41%，此前账面亏损主要源于一次性债务提前清偿。此外，产业链上游的康宁虽握有高额订单且 PEG 约为 1.39，但其业务中约有一半属于缺乏增长动能的传统工业制造与显示玻璃，纯度不及专注光子学的 Lumentum。 主讲人认为，面对铜缆物理限制与 AI 硬件 2.0 周期，Marvell 激进的垂直整合具备长期战略协同价值；但投资者需留意巨额并购所带来的流动性压力与股权稀释风险，并建议在硬件板块内维持分散配置以防范单点波动。

[0:00] Broadcom CEO: On optics, oh, I try to avoid bright, shiny objects.

[0:00] Well, on the same day that we get crazy Google earnings, which we already broke down in a separate segment, we just got a mega deal between Mar-Vell and Lumentum, supposedly in the works that could happen as soon as next time. Well, the deal could be announced as soon as next week. Now, this is crazy because you're talking about Marvell, which is a $221 billion company, which Jensen himself has suggested could be the next $1 trillion company. And another company called Lumentum, ticker Lite, L-I-T-E, which is appropriate given that they work on lasers for fiber optics. Get it? Light? Anyway... They are an $83 billion company. This is an insane potential acquisition for Mar-Vell, which does not have the balance sheet to justify this. That's why the stock's going down. And that's why light is going up. So let's understand a little bit of the background here. So the basic gist of what's going on is the Marvell CEO is going all in on co-packaged optics, CPO, and Fiber optics, laser connections, Okay, that's different. from a copper connection. Who's in the copper camp? Jensen says you use copper as... much as you can, because it's way cheaper, right? So think about this as just connecting a few different things, either a data center to a data center. Iraqis. to a rack at the other side of the data center. or inside of the rack, Chip, To a chip. or even inside of a chip, a part of a chip to another part of a chip. Those are four layers that you can connect stuff in. Fiber optics pretty much already handled data center to data center and they're slowly getting to rack to rack. fiber optic cabling to have faster, lower latency connectivity between these data centers, which all becomes much more important when you're working artificial intelligence and you want rapid responses for either customer service or voice AI or agents or whatever, all this stuff is going to get a whole lot more extreme and the investments in all this or just beginning But the point is, Mar-Vell takes the argument that we are going to start seeing... rack to rack basically go all fiber. we are going to see chip to chip GoFiber, And then we're going to see inside of the chip Fiber to fiber because the speeds are going to get so insane that you're going to hit what they call a copper wall. Now, Jensen says use copper as long as you can. It's cheaper, better margins for NVIDIA that way. Broadcom CEO, he says, Look. We'll do lasers... but we like copper. Okay, so Broadcom CEO guy, I actually like him. I think he's a very humble guy. Broadcom, I've got some notes here. Broadcom does 60% of the ASIC market. I'll just show them to you. This is actually in our course member only kind of like research tab. but whatever. Broadcom does 60% of the ASIC market. Meta and Anthropic joined Broadcom in co-development as well. Real competition they see is, you know, NVIDIA. But GPUs are like, your general purpose workhorse. ASIC is very specific. It's in the name. Specific. Application Specific Integrated Circuit. Marvell does only about 15 to 25% of the ASIC marketplace. So Broadcom is the leader over here. And so Marvell is taking a different approach to Broadcom. Broadcom says, look, we'll do copper as long as we can. In fact, we've got some new copper products that can go really fast. We're good for a while. Marvell is like, no, man, we're going to go optics because we think optics are going to be the future proof of the future. Well, that's redundant, but you know what I mean. And that copper is going to get left behind. You know what the Broadcom CEO literally says? He says... So he's asked about fiber optic, right? And this is literally his line. I wrote this down. This is in the course member research tab. Remember, you could be a member, you get lifetime access to this kind of stuff at reinvestone.com. This is just a sample of the stuff that we'll be talking about in the research terminal later this year. But anyway. He writes: on optics, Oh. I try to avoid bright, shiny objects. Obviously, you know, he's being facetious here because hyperoptics are bright and shiny. Anyway, I like the guy. I actually think he's a smart guy. I think the answer is you buy both, right? Like Broadcom and Marvell. Not personalized advice. I have exposure to both of them. full transparency. But Broadcom is used by Google to manufacture the TPUs. Marvell would get some TPU revenue as well. But these are two very big differences between Broadcom and NVIDIA's point of view. [5:00] and Marvell and Lumentum. Now you might think that makes Nvidia bearish on fiber. No. Nvidia invested $2 billion into... Mar-Vell, and $2 billion into... Lumentum. into both of them, NVIDIA has a collective $4 billion investment. Now that's really interesting. I think what Jensen is doing, he's hedging his bet. hey, let's use copper today. It's good for the margins. But, you know, if fiber works out great, We called it! We invested, and we call Marvell the future $1 trillion company. That would be like a 5X from here. Okay, well that's cool, and they want to obviously ride that wave up, but what do they see? Well, what they see is they're going to set up, NVIDIA is going to set up Marvell to do its best to integrate their products. into NVIDIA ecosystems. So that way NVIDIA and the whole CUDA moat can still be strong, and Marvell doesn't... you know chip away at Nvidia's lead. They're basically saying look, we want everybody to buy our GPUs. This is our ecosystem. Okay, we got the, we got the Ethernet, we got NVLink, we got the Vera CPUs, we got the Rubin GPUs, the Blackwells, whatever we got the whole stack baby, we got switches, you name it, we got Spectrum X, which is a copper based Ethernet, we have it all. You don't need anyone else, says NVIDIA. And then you got Mar-Vell, which is this little squeaky animal, going, yeah, but we're designing all these Asics! You know, all these people are doing custom chips, and, you know, they'd love to be able to integrate them in your NVIDIA system, you know, your NVIDIA ecosystem. And so you might wonder, like, okay, well, what kind of chips are they integrating with? And the answer is a lot of them. Marvell... Does the Tranium XPU, which represents about 15.5% of Mar-Vell's revenue, they do 8.4% of their revenue comes from the, I don't know how to say it, Maya. custom XPU for Microsoft, and then the Axion CPU from Google some of their TPUs as well. And then a custom chip designed with Meta. So Marvell's doing all these deals with Amazon, Microsoft, Google Meta. Those are companies that happen to be blowing a lot of money on AI. And so NVIDIA is kind of like... You got to deal with all those guys? Alright, alright, alright. How about this? we'll create something called NVLink Fusion And you can be one of the few people to use it. We'll open our arms to you, Mar-Val. You can bring your crap into our racks. You want to ride my rack? Just plug into NVLink Fusion and Mar-Vell will be one of the partners. And then Mar-Vell's like, great. We have infiltrated the NVIDIA Rack. This is wonderful. We're deep, boys. We're deep. How can we go deeper? Let's buy the whole damn stack. That's basically what they're saying. By the whole damn stack. Because see, Marvell is a fabulous ASIC manufacturer. So they could design, but they don't actually do everything inside of the rack. So you kind of have to think like there's rack, Then you have cables. These cables can be produced by a company like Corning, which manufactures the glass that goes inside of the fiber optic cabling. And then you have something that injects the light, a device that's shooting the light, right? Lumentum manufactures those. then Mar-Vell does the chips that collector translate... that light data switch-to-switch data. But then Mar-Vell also has... these Custom ASICs. And so Marvell's like, look, we could do the GPU work for you. We could do the Marvell DSP. This is, you know, server to switch. We could do switch to switch, which is our photonics chip. We could then buy Lumentum, which injects the light. The only thing they haven't bought is Corning. which is in that vertical, which is essentially the glass and fiber manufacturer. Now, Corning is a $113 billion company. This is like a 150-year-old company I actually am really excited about Corning. I don't own Corning. I'm also really excited about... and I'm kind of pissed. Because... We've been talking about Lumentum in the course member live streams and I've been talking about how much I like them and how interested I am in them. You know, I'm not convinced I want to buy yet. And so I was a little pansy weenie baby and I didn't buy them yet. And, you know, now they're up 10% because of this acquisition. Of course, you know, whatever. But we've been talking about Lumentum because this whole fiber optic thing, this is just getting started. I think it's still early on this. I know some people are like, "Oh, it's late!" The DSPs are the digital signal processor, by the way. These are like the brain inside of the optical modules for fiber optics in Marvell So, you know, I think that's a good thing. [10:00] manufacturer. So by buying Lumenta, Marvell would also be buying the manufacturing capabilities that they do not have right now, which is pretty remarkable. So Jensen's quote on all this, by the way, is use copper. When you can't, use optics when you must. Hmm. Kind of an interesting little take there from Jensen. He's always got the little hot takes, right? But I personally think this is very exciting because Marvell... I was also reviewing another Marvell interview. This was a legacy one. But this legacy interview, just to kind of get some insight into where the CEO's head is, he thinks the next... This video is... This sponsor is brought to you by meetkevin.com/silo. Wait a second, points for stocks? You may have heard about our sponsor Silo today. It's basically what Bilt did for rent. But this time... It's for investing. So if you're not familiar with this, here it is. Silo is the first and only investing platform that gives you credit card style points, but on your brokerage account. Every month, they give you points based on your account value. You get an annual multiplier of 0.5x points on your total account value. So if you got $500,000 in an investing account, you'd earn $500,000. 50 bips. That's 250,000 points every single year. Dang, I'm missing out on some points. Points are redeemable for cash. You know, a lot of us are probably thinking that. We're like, man. Wait. What? I want points. What can I do with them? And here you go. Points are redeemable for cash at one cent Per point. or you could use their travel portal to book at 1.3 cents per point depending on your membership tier so here's the big thing they are also now launching their very first transfer partner It may even be live by the time you're seeing this integration. Now, What is really worth noting? is that while I can't yet reveal the airline, as I can say, It is a Star Alliance partner at a one to one transfer rate, which is pretty neat because Star Alliance partners allow you to book flights with 26 different airlines, including United, Turkish, Swift Air, Lufthansa, Singapore Airlines. And so if you're clever with how you redeem and use these points, you could fly around with some, you know, pretty nice seats, business class or even first class, simply because you had your portfolio at silo and they were paying you. basis points on your portfolio. And so with the launch of their very first transfer partner, Silo is officially exiting beta and switching to their long-term points earning rate of 0.25%. in November. But... Anybody who signs up this month locks in those 50 basis points for at least the first year. and you might even get grandfathered in but of course check their program details so if you want to lock in .5x points on your stocks, 50 bips, and earn up to 250,000 points. as a new member, make sure to use my link, meetkevin.com slash... Silo. It's easy. Meet Kevin Alcom/Silo. S-I-L-O, to get started by the end of this. He thinks the next phases of growth are not just moving data. Like, moving data is what they're doing with fiber optics. It's also storing data. And this gets really deep. And I'm sorry, this is like... This is like so niche and nerdy, And I apologize, but I have this belief that First of all, I... How should I put it? I think the next boom... isn't necessarily going to go directly to memory, but a place that got sold off with memory that might recover faster is storage. This sounds crazy, but... Agentic AI uses so much damn KV cash over time and robotics is just going to make this worse, what's going to happen is you're going to need more storage. And eventually you might use a lot of off-rack storage where you have like All the NAND SSD you know, whatever, is over in that room over there. And all the GPUs in the A6 are over here. Well, how are you gonna connect those two things together? How about fiber optics? Fast, baby. See how all these things start? linking together you know, plugging into each other. It's crazy. And I'm not here to say that I can tell you exactly. which company is going to perform the best out of this whole stack. I think in this environment, spoiler just between us, okay? Don't tell anyone. Don't tell anyone. Thank you. I think you lean in on hardware, not personalized financial advice, but I think you have to be pretty damn diversified inside. [15:00] it because if you go all in on one of them it could end up pulling a google on you right So you diversify amongst these hardware plays and you just bet that the whole damn ship of AI is going everywhere. bump Unless you're a bear, then you do the opposite. But if you're a bear, you're... possibly not even watching. Okay, so let's look at the actual fundamentals a little bit. So the last fundamental that we did on Marvell in the course member live stream was on October 5th. Now remember, all this stuff that we go through, some of it's going to be, well, we're going to have a lot in our research terminal in the future. That's part of what's coming out in the Reinvest One. You pay once, you get lifetime access. But Reinvest Research is coming out. But you also get, if you join our course member live streams, analysis, technical analysis. We'll look at these plays and we'll try to understand, hey, where are the technicals? Where are the fundamentals? What's a buy? And we're just trying to understand more. We're trying to learn more. This is a learning exercise where every day we're trying to expand our knowledge with what the hell is actually going on with artificial intelligence. And there's a lot. So we think we do a really good job of this, but we're going to do an even better job when our research terminal comes out. And the terminal is literally going to have a module on building an agentic AI business or improving your business with agentic AI, being a better consumer with agentic AI. You get the course member live streams, the nine courses, the research terminal, the stock terminal, the homes terminal. Like you get it all in one bundle at reinvestone.com. We think it's a killer deal. And... You really want to join it in 2026. This is going to be like full self-driving where all of a sudden you can't get lifetime access to it anymore after 2026 ends. And then it's going to be like, damn, I wish I got lifetime access. Because I think it's in a year from now, people are going to go, holy crap. That's actually really good. We'll see. It's early. Okay, so this is the Mar-Vell... Statement of operation. So this is our income on Marvell. We have a 36% increase in net revenue. Cost of goods sold actually grew less than that, so their pricing power grew. So we've got strengthening pricing power at Marvell. Very bullish about Marvell's pricing power. I love them. They also divested an automotive business, and they had free cash flow, a free cash flow yield of only about 1.68%, which is $1.6%. On the lower side. But you've got some free cash flow. They did also borrow... And they issued a little bit of stock. So, you know, they're also in this race of, "Hey, we got some building to do here." You can see my note over here I wrote... Broadcom Has some optical ambitions, but Copper Connectivity, they're a Copper Connectivity heavyweight. Marvell leans into CPO. a CPO optics. Okay. Right. I wrote, I think of Mar-Vell as the discount... StepSys to Broadcom. She's a bit more pricey, which is kind of ironic because I actually think Marvell has a higher price than Marvell right now. But I guess what I mean by this is market cap wise. OK, market cap wise, you're talking a 200 billion dollar market cap versus about a trillion dollar market cap at Broadcom. It might even be more than might be like a trillion and a half now. I can't remember. But anyway, that's so when I say it's a discount, the Broadcom valuation is lower than Marvell's. But Mar-Vell is that smaller... up-and-coming step row that's making the bet on optics... And this absorption of light. Damn, it pisses me off. No, it's actually great. I'm actually really bullish Marvell on it. I would buy the Marvell dip on this. Because I want the combined company. Like these are, I'm very happy if this goes through. obviously the Marvell stock is going to go down, Light stock is going to go up. That happens every time there's an acquisition because Marvell doesn't have the money. They do not have $83 billion in cash sitting around. They got like $6 billion in free cash sitting around. So this is going to be debt and shares. It's going to be dilutive. Yeah, Light's going to have some cash on their balance sheet as well, but not much, dude. They got $3.8 billion in bills. And they got 3.4... in cash and receivables. So there's no cash in their $80 billion market cap. So you're going to have to take on debt and stock to do this at Marvell. So yes, Marvell stock goes down in the near term. But this is a behemoth that they are creating. This is delicious. Oh, thank you for the shout-out, per 82, 36. I love Reinvest1 Thanksgiving. Oh, yeah. Appreciate that. So So, Take a look at this. This is what I wrote. Okay. This is what I wrote on September 25th. Oh, 925. That's my crush's birthday. Well, this was my high school crush. I don't know why I still remember her birthday. It just makes Lauren mad. Don't tell her I said that. It's very transparently known. I got rejected by the girl in middle school, okay? I asked her out for pizza, and she said no. And I still haven't forgotten. Anyway, this is a mass [20:00] Photonics play for artificial intelligence. Mega boom, like, memory has already happened. So there's been this mega boom in light. The stock's actually come down a lot since we did this analysis. And I thought they were a little expensive then. So the last, you know, four weeks, the stock has come down a little bit. So I said at 1.2 peg, there are cheaper options like Broadcom and Marvell are actually cheaper. you know, post the Iran situation. They see the net revenues growing 1.8x and gross profits are up 2.7x. So you're talking massive pricing power here. Their operating expenses are only up 14%, margins 41%. This is sexy, man. Yes, we did have a loss over here. Why did the loss happen? That was because of an extinguishment of debt. So you kind of have to ignore this nonsense over here on the right. It's more one time. That's Sydney Swee's birthday. Was her name Sarah? Nope. Name was Kristen with a Y. Uh, anyway. So, um... Okay, investing activities. and let's see a hundred billion okay okay okay accounts receivable 3.8 billion of bills oh yeah we already covered that this is the financial information assets Yeah, we've already covered that as well. Okay. Let's look at... Let's go into this a little bit more. Okay, so corning has massive supply agreements. Corning's an interesting one too. Let me do a quick valuation check on Corning. uh, Corning is GLW. They're at about a 1.39 peg. Great pricing power. Um, But they have a legacy business in it, right? So half of the business at Corning is eh. And the other half is great for fiber optics. It's just not as pure of a play. So you're really paying a fair price for an entire company when only half of the business is booming. That said, you know, Dell did really well when only a tiny portion of their business is booming. And honestly, Dell has gotten cheap. Dell has gotten a whole lot better. cheaper. So. This is very interesting. Now if I go look a little bit more here right now inside of racks gpu to gpu links are 100 copper But Copper can't scale beyond three meters... at the speeds desired for the next generation clusters. And the more the speed goes up, the more the copper distance that you could cover, like halves. I think they say for every doubling... in speed you want to accomplish, you have the distance that copper can cover. Thank you. And this is why Corning is seen as such a big opportunity in scaling out with fiber. But this is why Mar-Vell is going, F it. Even if we have to tank it in the short term, we want the full stack. We want the laser. We want the CPOs. We've already got the DSPs. We've got the ASICs. We've got the connectivity to get inside of the NVIDIA rack. We almost have everything to be fully vertically integrated. This CEO is going full on Elon on your butt. and I like the guy I think... I think he's... more ambitious than the Broadcom guy. I think the Broadcom guy is more conservative. That does increase execution risk. But I like the guy. I respect the guy. I respect his acquisition. I'm a big fan. Uh, uh, and, and even though the one that I own, Marvell is going down on this and light, the one I wanted to buy, the one that got away is going up on this. I have to say, overall, this is brilliant. Big fan. And this just reiterates the AI boom that is coming. It is just getting started. And you get an Iran deal, man. Come on, baby. Here's Kevin with that Iran deal. But everything flip-flops. And we go towards hardware 2.0 rally. And these valuations that are super cheap now, then you're going to get the euphoric rally. Then people are going to go, holy smokes. That's what a bubble actually looks like. That's my opinion. Maybe I'm wrong. So in the meantime, I'll just play this. Here we go, come on, push it. Find the dip, find the dip. - If you liked that video, check this one out. I think you're gonna love it. - Why not advertise these things that you told us here? I feel like nobody else knows about this. - We'll try a little advertising and see how it goes. - Congratulations, man, you have done so much. People love you, people look up to you. - Kevin Paffer, financial analyst. And YouTuber, Meet Kevin, always great to get your take.