AI News

Marvell's $115 Billion Acquisition of Lumentum: Four Identity Transformations Completed Through Acquisitions Over Thirty Years

HuxiuUpdated Nov 15 19:25

AI summary

Marvell announced a $115 billion all-stock acquisition of laser company Lumentum, with the deal expected to close in the first half of 2028, bringing the light source segment into its portfolio. The article reviews its more than ten acquisitions over three decades: in the 2000s it bought its way into becoming a "general store" with a bloated product lineup but no number-one product, and in 2015 it exited mobile phone chips; after 2016 it divested WiFi and automotive Ethernet, recouping $4.3 billion, then acquired Cavium, Inphi and others, with revenue rising from $2.65 billion to $8.195 billion and the data center share climbing to 79%. Lumentum's 200G EML laser chips hold more than 90% market share, with capacity booked through early 2029, and Nvidia already invested and locked in capacity in March 2026.

Why read this

AI score 66

Using more than ten acquisitions over three decades as a thread, this article reviews the successes and failures of Marvell's four identity transformations and the place of its latest $115 billion deal.

Topics

Tags

Body · AI translation

[[**Over thirty years and more than a dozen acquisitions, Marvell completed four identity transformations. In 2026 it announced a $115 billion acquisition of Lumentum, bringing laser light sources into its portfolio and transforming into an optoelectronics-integrated connectivity company.**

**Key Points**

**1. All four identities were rewritten by acquisitions**

From hard drive chips to a consumer electronics "general store," then to data centers, and now to optical connectivity—its products, customers, and revenue structure have all changed; the core method is to use acquisitions to buy time.

**2. The first two reinventions: one failure, one success**

After 2000, acquisitions formed a "general store" with no product line ranked first, and in 2015 it exited the mobile phone chip business; after 2016, it sold WiFi and automotive Ethernet to bring back $4.3 billion, then bought Cavium, Inphi, and others, with revenue rising from $2.65 billion to $8.195 billion and the data center share rising to 79%.

**3. The third turn toward optics still lacked lasers**

From Inphi to Celestial AI, XConn, and Polariton, Marvell pushed optical connectivity from between racks into the package, but lasers—the scarcest link and the hardest to expand production for—remained controlled by others.

**4. The $115 billion acquisition of Lumentum fills in the light source**

Lumentum holds more than 90% of the 200G EML market, with capacity booked through early 2029; after Nvidia invested and locked up capacity in March 2026, Marvell acquired it in an all-stock deal, with the transaction expected to close in the first half of 2028. 2026-11-05 10:17 Marvell: It Bought Itself Into Another Company Big G Tech Watch

From a single hard drive chip in 1995 to buying a laser company in 2026, thirty years, four identities—every transformation was bought. This article is from the WeChat public account: Big G Tech Watch, author: Big G Tech Watch, cover image from: Visual China

On November 2, 2026, Marvell announced the $115 billion acquisition of laser company Lumentum, taking the source of light into its own hands.

This is its largest acquisition in thirty years and its fourth identity transformation. Thirty years ago it made hard drive chips; afterward it did consumer electronics, then data centers, and now it wants to become an optoelectronics-integrated connectivity company.

Today's Marvell and the original Marvell are no longer the same company. Its product catalog, customer list, revenue structure, gross margin, and competitors have all changed. It did not complete this through a single transformation, but through more than a dozen acquisitions over thirty years, one by one, buying itself into another company.

Thirty years, four identities. The story begins in 1995.

The first Marvell: from a single read-channel chip to a "general store"

In 1995, Sehat Sutardja, Weili Dai, and Pantas Sutardja founded Marvell in Sunnyvale, Silicon Valley. The initial funding was about $350,000, from family savings. All three were Berkeley graduates—a husband and wife plus a brother, a standard family startup team.

They chose a very narrow niche: the read-channel chip for hard drives. This market was then monopolized by Texas Instruments, using bipolar processes. Sehat Sutardja believed that CMOS, which the industry considered "too slow," could be faster, more power-efficient, and cheaper as long as it was designed well enough.

The technology was developed, but no one dared to use it. Samsung refused, and Western Digital refused—not because the chip was bad, but because the company was too small and no one was willing to replace an existing system for a supplier that had not reached mass production. The turning point came from a cold call by Weili Dai. She reached Kenneth Boon, a scientist at Seagate, who was worrying that Texas Instruments' new-generation chip could not meet speed targets. Seagate gave an almost impossible target, and Marvell delivered it in three months. Seagate not only placed an order but replaced Texas Instruments entirely.

This was Marvell's first time punching above its weight. In 1998, Seagate became its first major customer. In the following years, the company nearly doubled every year. On June 27, 2000, Marvell listed on Nasdaq at an issue price of $15, closing near $56 on the first day, up nearly threefold, raising $90 million. Hundreds of employees became millionaires overnight, and the two founders made Fortune's list of the richest people under 40 that year.

After going public, Marvell began to "buy."

In October 2000, it acquired Israeli network chip company Galileo for about $2.7 billion in stock, entering Ethernet switching; in 2002 it bought SysKonnect, and in 2003 it bought Radlan for $49.7 million to add network software; in 2005 it bought QLogic's hard drive controller division for $180 million in cash plus $45 million in stock, and UTStarcom's SOC business for $24 million; in 2006 it bought Avago's printer ASIC business for $240 million.

Then came the most important deal: in July 2006, for $600 million in cash, it bought Intel's XScale mobile and handheld device chip business.

The imagination behind this deal seemed enormous at the time. XScale's PXA series application processors were used in star models such as BlackBerry and Palm Treo; and Marvell already had WiFi, Bluetooth, and storage controllers. It could supply almost all the peripheral chips a phone needed. In 2007, when the first-generation iPhone was released, its WiFi chip was Marvell's.

That was the closest Marvell ever came to the consumer electronics throne.

It was also the hardest it ever fell. XScale had an application processor but no baseband. Qualcomm integrated the baseband and application processor into a single chip, while MediaTek turned the entire solution into a "turnkey" package—phone makers only needed to slap on their brand. Marvell was stuck in between: in the high end, it couldn't beat Qualcomm; in the low end, it couldn't beat MediaTek. In 2015, Marvell exited the mobile chip business and laid off about 1,200 people. From a $600 million purchase to a full exit, it took less than a decade.

Looking back at these twenty years, Marvell's acquisition logic was really just one thing: buying revenue with stock.

Wall Street wanted growth, and in-house development was too slow, so it bought. By 2010, Marvell had climbed into the global top twenty semiconductor companies, ranking eighteenth. It had more than ten thousand patents and thousands of product models—hard drives, Ethernet, WiFi, phones, printers, set-top boxes, it had everything.

And it was also after that year that growth stopped. Revenue ground along between roughly $3 billion and $3.7 billion for four years, then fell back to $2.65 billion in a single year. The stock slid from the teens to under $10.

This was the endpoint of the first Marvell: a company that could do everything but was first in nothing. In every one of its product lines stood a more focused, larger rival. In hard drive controllers there was LSI, in Ethernet switching there was Broadcom, in phones there was Qualcomm, in WiFi there were Broadcom and Atheros.

It was like a big supermarket at an intersection: the shelves were full, but every street had a more specialized shop.

The board's audit committee launched an internal investigation, the SEC stepped in, and Nasdaq issued a delisting warning because the fiscal 2016 10-K was filed late; that same year, the company reached a $750 million patent settlement with Carnegie Mellon University. The audit committee ultimately did not find fraud, but concluded that management had exerted "significant pressure" on the sales team to meet revenue targets, and that revenue recognition on some transactions did not comply with company policy.

In April, founders Sehat Sutardja and Weili Dai resigned as CEO and president. Activist hedge fund Starboard Value by then held about 7% of the shares and pushed through a textbook management reshuffle. In July, former Maxim Integrated executive Matt Murphy became president and CEO.

When he took over, Marvell was worth less than $5 billion, with the stock around $10. Customers were worried about whether the company could survive.

Murphy later said at his first analyst day that this was a "company without a rudder." R&D spending was sprinkled like pepper across more than twenty product lines, and not one was in the industry's top three.

The first thing he did was not buy, but sell.

In 2019, the WiFi and Bluetooth business was sold to NXP for $1.76 billion; in 2025, the automotive Ethernet business was sold to Infineon for $2.5 billion. The two divestitures brought back $4.3 billion and completely pulled the company out of consumer electronics.

Many companies' transformations start with "buying"; Marvell's transformation started with "selling." To become another company, it first had to sell off its old body.

Then came the buying. In 2018, it acquired Cavium for $6 billion in cash and stock, gaining ARM server CPUs, DPUs, and network security chips; in 2019, it acquired Aquantia for $452 million to add multi-gigabit Ethernet; that same year it acquired Avera Semi, the GlobalFoundries subsidiary, for $650 million to add custom ASICs; in 2021, it acquired Inphi for about $10 billion to add optical DSPs; that same year it acquired Innovium for $1.1 billion to add data center switch chips.

And there was "binding." What he wanted was not orders, but "multi-year visible revenue." Marvell signed a multi-generation agreement with AWS spanning more than five years, covering custom AI chips, optical DSPs, AEC DSPs, PCIe retimers, DCI optical modules, and Ethernet switching—not a single transaction, but a whole system-level partnership. Some capacity reservation agreements disclosed in the 10-K ran from four to ten years.

The results showed up in the financials: FY2016 revenue was $2.65 billion, FY2026 revenue was $8.195 billion; the data center business rose to 74% of revenue, and further to 79% in the second quarter of FY2027.

Ten years ago, data center accounted for less than 10% of its revenue. Today, it accounts for nearly 80% of the company's revenue.

By this point, Marvell had become a second version of itself: a data center infrastructure company. Cavium gave it compute, Aquantia gave it Ethernet, Avera gave it custom ASICs, Innovium gave it switching.

But it was still just a company that "sells chips."

The third Marvell: starting with Inphi, moving toward light

In October 2020, Marvell announced the acquisition of Inphi for about $10 billion in cash and stock. When the news came out, Wall Street generally thought it was expensive—Marvell's annual revenue at the time was still under $3 billion, yet it was paying three times its own revenue for a company that made PAM4 DSPs.

But what Inphi gave Marvell was not a product line, but a position.

If Cavium changed Marvell's customers, Inphi changed its identity.

In an 800G optical module, the two most expensive types of components are the DSP and the laser. The DSP is responsible for "fixing" the signal at the electrical level, and is the first hurdle in each speed upgrade; the laser is responsible for converting electricity into light, and is the physical starting point of transmission. Inphi is the global leader in optical DSPs, and Marvell's purchase of it was equivalent to standing at the junction of "electricity" and "light."

More importantly, Inphi let Marvell understand light for the first time.

It began to realize: no matter how strong its own DSP is, it still has to wait for the laser. The laser's wavelength, linewidth, power, and yield determine whether the DSP can run at full capacity; the production capacity of optical components determines how many modules the module makers can deliver; and the delivery of module makers determines how quickly cloud providers' clusters can come online. In this chain, companies that make electrical chips look important, but are in fact "supporting players."

So starting in 2021, Marvell followed the line of "connectivity" and bought its way upstream, inch by inch.

In 2021, it bought Innovium for $1.1 billion to fill in data center switching; in December 2025, it bought Celestial AI for $3.25 billion, acquiring Photonic Fabric to bring light inside the package; in early 2026, it bought XConn for $540 million to fill in PCIe/CXL switching; in April 2026, it bought Polariton to fill in plasmonic modulators.

If you arrange these deals by "distance," a very clear route emerges: inside the package (millimeter scale) → chip-to-chip (centimeter scale) → board level (tens of centimeters) → rack-to-rack (meter scale) → data center-to-data center (kilometer scale).

Marvell took "connectivity" from electrical to optical, and from outside the rack to inside the package.

By this point, it was no longer a "chip company." What it sells is connectivity—from one chip to another, from one rack to another, from one data center to another. Chips are merely the way it delivers connectivity.

But it was still missing one thing: the light source itself.

It makes DSPs, makes switching, makes custom ASICs, makes retimers, makes photonic fabric, but it does not make lasers. And lasers are precisely the scarcest, most rigid, and hardest-to-expand link in the entire chain.

This is why Lumentum appeared.

The fourth Marvell: taking the source of light into its own hands

The one that filled in this link was Lumentum. Its predecessors can be traced back to Uniphase in 1979 and JDS Fitel in 1981, which merged into JDSU in 1999; after the 2015 spin-off, the communications optics business became independent as Lumentum. Through several rounds of acquisitions, it obtained indium phosphide laser chips and data center optical modules, and by 2026 its global market share in 200G EML laser chips exceeded 90%, with production capacity long since sold out and booked into early 2029.

When Nvidia took the lead in investing and locking up Lumentum's capacity in March 2026, Marvell had only two choices left: either keep waiting in line, or buy the entire company. It chose the latter, acquiring Lumentum in an all-stock deal, with Matt Murphy continuing to lead, and the transaction is expected to close in the first half of 2028.

From hard drives, networking, and mobile phones, to data centers, and now to optics, every acquisition has been a self-rewrite. Lumentum made it a company that truly owns a light source.

Worth noting is the timing: in recent months, optical module concept stocks have already corrected sharply and market sentiment has clearly turned cold, yet it chose this moment to buy the source of light.

What this $115 billion deal actually means and whether it is worth it—I wrote a separate piece dedicated to breaking it down. For now, remember one thing: once again, it decided not to be its former self.

Conclusion: what it becomes next depends on what it buys next

If you draw Marvell's thirty years of acquisitions into a single chart, you will see three career changes and four identities.

The first time, from 2000 to 2015, from hard drive chips to a "general store." From read channels to Ethernet, from WiFi to mobile phones, from printers to set-top boxes, what Marvell bought was "more markets." This career change failed. It turned Marvell into a company that had everything but was strong in nothing. The 2016 crisis was the bill for this failure.

The second time, from 2016 to 2021, from a "general store" to data centers. It sold WiFi and automotive Ethernet, and bought Cavium, Aquantia, Avera, Inphi, and Innovium. This career change succeeded. Revenue went from $2.65 billion to $8.2 billion, and market capitalization from less than $5 billion to $172.2 billion.

The third time, from 2021 to 2026, from data centers to optics. From Inphi to Celestial AI, from XConn to Polariton, and then to Lumentum. This career change is not over yet, and no one knows where its endpoint lies.

Three career changes, the same method: use acquisitions to buy time.

In the semiconductor industry, this is not a practice unique to Marvell. The rules of this industry are: fast technology iteration, high R&D investment, and short market windows. An InP production line takes 3 to 5 years to build, an optical DSP ecosystem takes 5 years to cultivate, and a relationship with a major customer takes 10 years to hone. But the AI window is only 2 to 3 years. Relying entirely on in-house development means that by the time you build it, the market has already changed direction.

So the giants in this industry are all built through acquisitions. Broadcom is, Intel is, AMD is, and Marvell is too.

But what makes Marvell special is this: what it buys is not just technology, but identity.

Every acquisition is a self-rewrite. Galileo made it a networking company, XScale almost made it a mobile phone company, Cavium made it a data center company, Inphi made it an optical connectivity company, and Lumentum made it—a company that owns a light source.

It has bought hard drives, bought networking, bought mobile phones, bought printers, bought DPUs, bought optical DSPs, bought photonic fabric, and now it has bought lasers.

Every purchase takes it a little further from "the original Marvell."

In September 2024, Sehat Sutardja passed away at the age of 63. The company he founded was long no longer the one it was when he left—its products, customers, and revenue structure had all changed. The only thing that did not change was that method of "using acquisitions to buy time."

Jensen Huang said that Marvell is the next trillion-dollar company. That statement is about its future. But Marvell's history over the past thirty years shows that every identity transformation of this company relied not on in-house development, but on acquisitions.

So, for the $115 billion acquisition of Lumentum, what truly deserves attention is not the price, but this: the company has once again decided not to be its former self.

What it becomes next depends on what it buys next.

Appendix: Marvell's thirty years of acquisitions (partial)

Sorted by transaction announcement date; amounts are publicly disclosed equity/enterprise value.

YearTargetAmountWhat it bought

2000Galileo TechnologyApprox. $2.7 billion in stockEthernet switching

2006Intel XScale$600 millionMobile application processors

2018CaviumApprox. $6 billionARM server CPUs, DPUs, security

2019Aquantia$452 millionMulti-gigabit Ethernet

2019Avera Semi$650 millionCustom ASICs

2021InphiApprox. $10 billionOptical DSP

2021Innovium$1.1 billionData center switching

2025Celestial AI$3.25 billionPhotonic fabric (in-package optical interconnect)

2026XConn$540 millionPCIe/CXL switching

2026Lumentum$115 billionLasers, optical components, OCS

This article comes from the WeChat public account: Da G Kan Keji, author: Da G Kan Keji (technology industry observer, with more than 20 years of experience in technology media and brand public relations for listed companies, formerly editor-in-chief at a well-known domestic technology media outlet, and brand director at Fortune 500 companies and domestic and international listed technology companies, with long-term attention to the latest developments in technology fields such as consumer electronics, semiconductors, and AI.) AI industry signal channel: Frontier Technology

4 articles already published on Huxiu. This content is authorized for publication by the author; the views represent only the author and do not represent Huxiu's position. If you have any objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing the world and wants to change the world is on the Huxiu APP[ ]

Source

Huxiu