READING BETWEEN THE LINES

(AMD), (AMZN), (AVGO), (GOOG), (MRVL), (MSFT), (NVDA), (UBER)

I was rummaging through an antique shop in San Francisco last month when I stumbled upon an old telegraph key from the 1880s. 

The proprietor, a chatty fellow with more stories than customers, told me something remarkable: telegraph operators back then could identify each other purely by their typing rhythm on Morse code keys. 

They called it a “fist,” and it was as unique as a fingerprint. I bought the damn thing for fifty bucks, and it’s been sitting on my desk ever since, reminding me that even in the 1800s, the real challenge wasn’t the technology itself – it was moving information from point A to point B without losing fidelity.

That’s the business Marvell Technology (MRVL) is in, and it’s why I’ve been circling this stock like a chess player contemplating a risky opening gambit.

The market has been scratching its head over Marvell for most of 2025. The stock soared to euphoric heights in January, then spent the following months in purgatory while Wall Street tried to figure out if the AI infrastructure story was real or just another round of Silicon Valley magical thinking. 

What’s fascinating is this peculiar disconnect: analysts keep raising their earnings estimates, yet the stock trades well below those January peaks. In my experience, that’s either a screaming buy signal or a bear trap with your name on it.

When looking at Marvell, most people only see the headline-grabbing announcements from Nvidia (NVDA) and OpenAI, or AMD’s (AMD) hundred-billion-dollar partnership deals, and wonder where Marvell fits in the pecking order. They’re asking the wrong question entirely. 

The real story isn’t about who builds the fanciest AI processor, but about who keeps thousands of those processors talking to each other without the whole system collapsing into expensive rubble.

When you’re constructing AI clusters that consume enough electricity to power a small city and coordinate computational workloads across thousands of chips simultaneously, the networking fabric becomes just as mission-critical as the silicon doing the thinking. 

Marvell owns a commanding position in high-speed networking, optical interconnects, and switching infrastructure. It’s the modern equivalent of those telegraph lines connecting operators across continents, except now we’re moving petabytes instead of dots and dashes. 

The company has locked down a five-year roadmap with Amazon (AMZN) Web Services, which still commands the largest slice of cloud infrastructure globally. Some nervous investors fret about customer concentration, but I see it differently. 

Having AWS as your anchor tenant is like owning beachfront property with Warren Buffett as your next-door neighbor. It provides stability and credibility while you build out the rest of the neighborhood. 

And Marvell is definitely building. The back-channel chatter suggests meaningful conversations with Microsoft (MSFT), which is pursuing silicon self-sufficiency with the intensity of a startup founder who just got their first term sheet.

Now let’s talk about the uncomfortable truth that most bullish investors gloss over. Marvell’s custom compute business, while growing nicely, represents a fraction of what Broadcom (AVGO) commands in that arena. 

Management has been refreshingly transparent about targeting roughly 20% market share in custom AI chips. That’s not exactly world domination, and there’s a reason for the modesty. 

Custom silicon requires longer development cycles, deeper customer collaboration, and frankly, more ways for things to go sideways. Unlike merchant chips from Nvidia that plug-and-play into existing ecosystems, custom designs demand bespoke engineering for each client. Every design win matters enormously, and the execution risk is real.

What keeps me leaning forward in my chair, though, is the networking angle. As AI clusters scale from megawatts to gigawatts, the networking complexity explodes exponentially. 

Think about my telegraph key sitting on my desk. Those operators moved information at maybe 40 words per minute on a good day. Now we’re talking about moving data at speeds measured in terabits per second across increasingly complex topologies. 

Marvell’s expertise in moving data at lightning speed without cooking the servers positions them at the center of an irreversible trend. Roughly 75% of their revenue still flows from networking and optics, which provides a stable foundation while the custom compute story develops.

The valuation riddle also captures my attention. Marvell trades at about 22.5 times forward EBITDA, barely above its historical average, despite operating in what everyone from your Uber (UBER) driver to Jamie Dimon agrees is a multi-trillion-dollar secular growth opportunity. 

Either the market is pricing in catastrophic execution risk, or it’s sitting on its hands waiting for concrete proof points. My instinct says the latter, especially considering management just authorized five billion dollars in share buybacks. 

But here’s where I pump the brakes. The recent rally has pushed the stock into technically overbought territory, and I’ve paid enough tuition in this market to know that momentum often peaks right before reality reasserts itself. 

What I’m monitoring closely is whether Marvell can land a marquee win outside the AWS ecosystem, particularly with Microsoft or Google (GOOG). That would validate the diversification narrative and likely trigger the valuation re-rating that bulls have been anticipating for months.

I’m also watching the competitive chessboard. The flurry of partnership announcements between OpenAI, AMD, and others tells me the custom chip race is accelerating faster than most people realize. 

If Marvell can’t demonstrate tangible progress in expanding its compute footprint by mid-2026, the market’s current skepticism might prove prescient rather than pessimistic.

That telegraph key on my desk reminds me daily that technology changes, but the fundamental problems remain constant. Marvell’s developing its own distinctive “fist” in this market. 

But just like those telegraph operators who could spot an imposter by an inconsistent rhythm, I’m waiting to see a few more consistent beats before I commit capital. 

The signal looks promising, but I want to be certain I’m reading it correctly before I start tapping out buy orders.