Current Price: ~$93 | Rating: BUY | Entry Range: $85–$98 | 12-Month Price Target: $125

The Last Time Someone Built the Internet's Plumbing

In 1999, Cisco Systems was the most valuable company on earth. Not Microsoft. Not GE. Cisco. The internet needed routers and switches to function, and Cisco made the best ones. Every time a packet of data crossed the network, somewhere in that journey was a Cisco box. The company did not need to predict which websites would win or which software would dominate. It just had to be the infrastructure that carried all of it. For three remarkable years, that was enough to make Cisco worth $500 billion.

Of course, everyone knows what happened next.

The boom ended, and Cisco spent the following two decades trading at a fraction of its peak. The lesson most investors took from that episode was to be skeptical of infrastructure companies in the middle of technological manias. But there is a second lesson, less often discussed.

In the years before 2000, Cisco's revenue and earnings were real. The valuation was the problem, not the business.

That distinction matters enormously when you look at Marvell Technology today. The company is building the plumbing for the AI era. This includes the optical interconnects, custom silicon, and high-speed networking chips that allow tens of thousands of GPUs to function as a single coherent system. The question is not whether those products are in demand. They clearly are.

The question is whether Marvell is priced like a Cisco in 1999, or like a Cisco in 1996.

I think it's closer to 1996. Here is why.

What Marvell Does and Why It's Hard to Replace

Marvell designs semiconductors for data infrastructure. It does not own fabrication plants. It is fabless, relying on TSMC and Samsung to manufacture its chips, but the intellectual property that goes into those chips is Marvell's. The company operates in four main product categories: optical interconnects, custom silicon (also called XPUs or ASICs), data center switching, and storage controllers.

Optical interconnects are the cables and chips that move data at high speed between servers inside a data center. As AI clusters have grown from hundreds of GPUs to hundreds of thousands, the speed at which data moves between chips has become a binding constraint.

Marvell's PAM-4 DSP technology sits at the leading edge of this transition. The company was the first to ship a 1.6 terabit-per-second solution, entering volume production in the second half of fiscal 2026.

Custom silicon is the business that has surprised even Marvell's own management team.

Three years ago, it barely existed. In fiscal 2026, it generated $1.5 billion in revenue, doubling in a single year. The business involves designing application-specific integrated circuits (ASICs) for large cloud providers who want chips optimized for their specific AI workloads rather than the general-purpose GPUs Nvidia sells. AWS and Microsoft are confirmed customers.

In fiscal 2026, Marvell reported total revenue of $8.195 billion, up 42% year-over-year.

The data center segment accounted for 74% of that, with full-year data center revenue surpassing $6 billion. Non-GAAP EPS reached $2.84 for the year, up 81%. Q4 alone delivered $2.219 billion in revenue against consensus estimates of $2.21 billion, with non-GAAP EPS of $0.80 beating the $0.71 estimate by over 12%.

Three Reasons to Own This Stock Right Now

The first is the 1.6T interconnect cycle.

Data centers are transitioning from 800G optical connections to 1.6T, and that transition is happening faster than most analysts expected. Marvell entered production first, with multiple Tier 1 hyperscalers as customers.

On the Q4 earnings call, CEO Matt Murphy said he expects interconnect revenue to grow more than 50% year-over-year in fiscal 2027, well above the 30% growth the company had guided for just three months earlier. Each generation of interconnect technology takes 18-24 months before competitors can catch up. Marvell is currently in that window.

The second is custom silicon compounding.

The custom XPU business scaled from zero to $1.5 billion in a single fiscal year. Management guided for 20% growth in fiscal 2027 and at least a doubling in fiscal 2028, which would put that segment alone at $3 billion or more.

The addressable market for custom XPUs is forecast to grow at a 47% compound annual growth rate to reach $40.8 billion by 2028. Marvell's relationship with AWS and Microsoft, built over years of iterative design work, is genuinely hard to replicate quickly. Custom silicon programs take two to three years from design win to volume production.

The third is the valuation disconnect.

After reporting the best quarter in its history and raising two-year guidance by nearly $1 billion above prior estimates, Marvell trades at roughly 16 times calendar 2027 earnings. Morningstar raised its fair value estimate to $130, implying the stock is still meaningfully undervalued at current prices.

That gap does not exist because the business is underperforming. It exists because Marvell's stock fell sharply from its 2025 peak of $127, spent months in the $47-$78 range, and left institutional ownership at a level that does not yet reflect what the business has become.

Competition and the Macro Backdrop

Broadcom is the most important competitor, and it is a serious one. Broadcom's custom silicon business, serving Google's TPUs, is larger and more mature than Marvell's. Broadcom's market cap is roughly 5x Marvell's, which means it can out-invest Marvell in R&D. The risk is that Broadcom wins more of the next generation of design programs. Marvell's recent acquisitions of Celestial AI and XConn, focused on photonic interconnects and PCIe/CXL switching, are partly a response to this pressure.

On the macro side, capital expenditure from the five major US hyperscalers continues to accelerate.

Microsoft's fiscal 2026 CapEx guidance is in the $80 billion range. Amazon's is similar. That spending feeds directly into Marvell's order book. The risk is a sudden pullback in hyperscaler spending, which could happen if AI monetization disappoints or if economic conditions deteriorate materially.

Tariff exposure is also real. Marvell relies on TSMC in Taiwan for fabrication, and any disruption to that supply chain would affect the entire industry, not just Marvell.

Cheap, Fair, or Expensive?

At roughly $93, Marvell trades at approximately 16 times fiscal 2027 consensus earnings and 4 times fiscal 2027 revenue. The sector median for large-cap semiconductors sits around 29 times forward earnings and 9 times sales.

Marvell's PEG ratio of 0.62 is one of the lowest in the semiconductor space for a company growing this fast. Goldman Sachs puts the PEG at 0.1 on its own estimates.

The counterargument is that Marvell earned the discount.

Over the past year, it traded down more than 25% while peers held up better, driven by concerns about near-term growth and AI spending sustainability. Those concerns have now been answered twice, in the December 2025 earnings and again in March.

The lingering discount reflects investor skepticism about whether fiscal 2028's $15 billion revenue target is achievable, not doubt about the near-term business.

I think the $15 billion number is credible. Management's guidance has been conservative by historical standards. They guided $9.5 billion in September, $10 billion in December, and $11 billion in March. The trajectory of upward revisions is itself a signal about demand visibility. Morningstar's fair value estimate of $130 implies 22 times fiscal 2028 earnings, which is actually modest for a company growing revenue 30-40% annually with 60% gross margins.

What Could Go Wrong

Customer concentration is the most serious structural risk. A small number of hyperscalers represent a growing share of Marvell's revenue.

If AWS or Microsoft decides to bring chip design in-house, or shifts spending toward a Broadcom-designed solution, Marvell's custom silicon trajectory changes materially. The company acknowledged this risk in its own earnings release, noting that dependence on a few customers for a significant portion of revenue is a disclosed uncertainty.

The second risk is execution on the technology roadmap. Marvell is simultaneously ramping 1.6T interconnects, scaling custom silicon to 2nm and 3nm process nodes, and integrating two acquisitions. Any one of those could slip. The photonic interconnect technology from Celestial AI is cutting-edge and unproven at volume. Integration risk is real.

The third risk is the macro environment. Marvell's beta is 1.98, meaning it moves roughly twice the market on both sides.

A broader technology selloff or a genuine pullback in data center spending would hit MRVL harder than most. This is not a stock to hold if you expect a recession. It is a stock for investors who believe AI infrastructure spending continues for several more years - which I do, but that conviction needs to be yours before buying.

My Take

Marvell is one of the most compelling risk/reward setups in semiconductors right now.

The business has inflected. Revenue is growing at 42% annually with 60% gross margins, the custom silicon program is doubling year-over-year, and management has established a pattern of conservative guidance followed by meaningful upside. The stock is trading at a 40-45% discount to peers on forward earnings, and it just delivered the best quarter in company history.

I am initiating a position in MRVL. My entry range is $85–$98.

A 12-month price target of $125 represents approximately 15 times fiscal 2028 earnings - a discount to where I think it deserves to trade, but a number I find defensible on fundamentals alone. What would change the thesis is any meaningful reduction in hyperscaler CapEx guidance, evidence of design-win losses to Broadcom in the next generation of custom programs, or a breakdown in the interconnect ramp. I will revisit the position if any of those materialize.

The Earnout Investor provides analysis and research but DOES NOT provide individual financial advice. Jamie Dejter may have a position in some of the stocks mentioned. All content is for informational purposes only. The Earnout Investor is not a registered investment, legal, or tax advisor, or a broker/dealer. Trading any asset involves risk and could result in significant capital losses. Please, do your own research before acquiring stocks.