Skip to main content
Guillermo Barreto
← All posts

Somebody just tried to buy the bottleneck

5 min readBy Guillermo Barreto
Somebody just tried to buy the bottleneck

I read this one this morning before work started, and I had to read it twice. Not because it was confusing. Because of the size of it.

Bloomberg reported on Wednesday, October 28, that Marvell Technology is in advanced talks to buy Lumentum Holdings. A deal could be announced as soon as next week. No final agreement yet. The talks are still private, and they could still fall apart. Reuters reported the Bloomberg story here.

Lumentum makes optical components for communications networks and data centers. In plain terms, it makes the parts that turn data into light and shoot it between servers. Marvell makes custom AI chips and networking products. Put them together and Marvell would own a much bigger chunk of what connects an AI cluster, not just the chips inside it.

The market reacted fast, and in opposite directions. Lumentum jumped about 11% in extended trading, which put its value at roughly $91 billion. Marvell fell more than 8%. Before the pop, Lumentum was worth about $83 billion. Marvell is valued near $221 billion. Dow Jones covered the share moves here.

That split reaction is the part that stuck with me.

Investors were not cheering a brilliant deal. They were pricing in how expensive and how hard this would be. Reuters noted that Marvell had about $3.92 billion in cash as of its latest quarter. You do not pay for an $83 billion company out of a $3.92 billion bank balance. Bernstein analysts told Reuters the deal would likely be impossible to fund entirely with cash, and that the scale raises questions about financing and integration. Neither company responded to Reuters requests for comment.

So why do it anyway? Look at Lumentum's own numbers. Its revenue for fiscal 2026 was $3,014 million, up 83.2% from $1,645 million the year before.

I am a WGU software engineering student who works in IT support. I have never bought a company. I have never worked on optical hardware. I spend my nights shipping small code and doing CodePath interview prep, and my days on tickets. I am not going to pretend I know how to value a company at 28 times sales or whatever multiple the analysts will argue about this week.

But I do know what a bottleneck looks like from the support side. It is the thing everybody waits on while everything else sits ready. For most of this year, I have been writing about the AI buildout hitting one wall after another. Power. Grid queues. Cooling. Permits. Land. This week it is the light itself.

Think about it this way. You can have the fastest chips in the world sitting in racks, and they still have to talk to each other. In an AI cluster, thousands of processors work on one job together, and they pass enormous amounts of data back and forth constantly. If the connection between them is slow or scarce, the expensive chips wait. The cluster is only as fast as its slowest link. Lumentum sells the link.

That is why Marvell would rather buy the bottleneck than stand in line for it. Marvell already made a similar move late last year, buying photonics firm Celestial AI in a $3.25 billion deal, according to Reuters. It also raised its fiscal 2028 revenue forecast earlier this month to $20 billion, up from $15 billion, on demand for its custom AI chips and networking products. Its shares have nearly tripled this year. This is a company that has been winning, and it is still willing to pay an enormous price to own more of the connection layer.

That tells me something about where the work is going. The job I am studying for is not only about writing code. It is about the physical layer underneath the code. Fiber. Optics. Interconnects. The stuff you can actually touch, install, test, and troubleshoot.

I keep coming back to the same thought I have had all month. The glamorous part of AI is the model. The durable part is the building around it. Power, cooling, networking, and now the light running through the fiber between the racks. None of that is genius work. It is learnable work. It is the kind of work a student in IT support can actually grow into, one ticket and one certification at a time.

A few honest caveats before I stop. This is a reported deal, not a done deal. Advanced talks have collapsed before, and the price and terms have not been reported. The share moves I quoted are after-hours moves, which swing around a lot. And I do not know whether buying Lumentum will actually fix Marvell's supply problem or just hand it a very expensive integration project.

Still. When a company worth $221 billion looks at a supplier worth $83 billion and decides it would rather own it than wait for it, that is not a footnote. That is a signal about what is actually scarce right now. It is not the idea. It is not even the chip anymore.

It is the light between them.

Cover photo: a laser shining through a fiber optic cable, by Hustvedt, via Wikimedia Commons (CC BY-SA 3.0).