To The Moon

For the past three years, investors have been obsessed with one question: who will build the fastest AI chip?

I think the more important question is this: who gets paid every time those chips need to talk to one another?

That distinction sits at the heart of Marvell Technology's transformation. Following Jensen Huang’s public declaration that Marvell could become the next trillion-dollar company, investors suddenly began reassessing a business they had long treated as peripheral to the AI story.

The reaction has been violent. But I suspect many are still using the wrong mental model.

They see a semiconductor supplier.

I increasingly see an infrastructure business whose value rises with complexity itself.

The network may become more valuable than the nodes

Complexity Is Becoming the Product

The first phase of the AI boom rewarded raw compute. More GPUs meant more capability, more headlines, and more capital inflows.

The next phase may reward something less visible but far more structural.

As AI clusters scale into hundreds of thousands—and eventually millions—of interconnected processors, the constraint shifts from compute generation to compute coordination. Every additional accelerator introduces new communication pathways, new latency risks, and new inefficiencies if the system is not perfectly orchestrated.

In other words, scale does not just increase performance; it increases fragility.

Marvell’s portfolio sits directly inside that tension. Its optical networking, PAM4 DSP technology, and custom silicon programmes all converge on a single reality: AI systems are becoming more dependent on communication than computation alone.

That is not a cyclical tailwind. It is structural.

The Financial Picture: Expensive for a Reason

The market is already paying for this narrative.

Marvell has rallied more than 240% year-to-date and over 370% over the past twelve months, pushing its market capitalisation beyond $250 billion despite trailing revenue of $8.72 billion.

Traditional valuation metrics look stretched. The stock trades at roughly 22 times sales, 54 times forward earnings, and about 42 times EV/EBITDA.

On paper, that demands perfection.

But the underlying business is evolving quickly. Revenue growth sits at 27.6% year-over-year, profit margins are close to 29%, and levered free cash flow exceeds $2.2 billion annually. Cash of $3.84 billion against $5.28 billion of debt keeps the balance sheet manageable, but not immaterial.

The real question is not whether Marvell is expensive.

It is whether it is being valued as a supplier while behaving like a platform layer.

Conviction clusters where narratives become capital allocations

The Neutral Architect Thesis

The most important strategic shift is not Broadcom competition. It is architectural positioning.

Broadcom remains the industry's profitability benchmark, generating operating margins near 45% and free cash flow in excess of $25 billion. It is structurally dominant in execution and monetisation.

But I think investors are framing the rivalry incorrectly.

Marvell's opportunity may not depend on beating Broadcom.

It may depend on becoming something Broadcom is not.

Hyperscalers increasingly face a structural contradiction. They want access to NVIDIA’s ecosystem because it has become the default operating environment for AI infrastructure. At the same time, they want independence through custom silicon programmes designed in-house.

Historically, those goals have collided.

This is where NVLink Fusion becomes strategically important. It allows custom silicon to plug into NVIDIA-centric systems without fully surrendering architectural control. It does not resolve the tension—but it reduces the cost of managing it.

And that matters more than it sounds.

The hidden attraction is that hyperscalers increasingly want optionality. Designing a custom chip is difficult; redesigning an entire AI infrastructure stack every few years is even harder. A neutral integration partner reduces that friction.

In that world, $Marvell Technology(MRVL)$ is not competing to replace $NVIDIA(NVDA)$ or $Broadcom(AVGO)$.

It is competing to sit between them without being squeezed out.

Why Optics May Matter More Than Faster GPUs

The second underappreciated driver is optical connectivity.

Most investors still anchor AI progress to compute performance. That made sense when systems were small enough that communication was trivial.

Scale breaks that assumption.

As clusters expand, the cost of moving data begins to rival the cost of processing it. When accelerators spend time waiting for data rather than computing it, theoretical performance gains collapse in practice.

A useful way to think about it is congestion economics. Adding more compute without improving interconnect is like adding more cars to a city without widening the roads. At some point, throughput stops improving and begins deteriorating.

That is why optical interconnect and high-speed networking are moving from infrastructure detail to system-level constraint.

The irony is comical.

Investors spend fortunes analysing the engines while largely ignoring the motorways.

The moment infrastructure stops being invisible

Margin Pressure Today, Operating Leverage Tomorrow

Marvell’s profitability profile still reflects its transitional phase.

Operating margins of roughly 14.5% lag Broadcom significantly, and custom silicon programmes typically carry lower initial margins due to customer bargaining power and upfront engineering intensity.

That is the bear case in a sentence: Marvell is spending heavily today for uncertain returns tomorrow.

But the lifecycle dynamics matter more than the snapshot.

As programmes currently in development move toward production scale over the next two to three years, operating leverage could begin to materialise. Once design wins transition into volume deployment, incremental revenue should carry far higher marginal profitability than the current investment phase suggests.

If that transition occurs cleanly, margins could re-rate meaningfully. If it does not, today’s investment cycle will be judged far more harshly in hindsight.

Both outcomes remain live.

Verdict: The Bridge Is the Business

I do not find it difficult to see why sceptics are uncomfortable.

The valuation assumes sustained execution, continued hyperscaler investment, and successful conversion of design wins into production scale. Any disruption to that trajectory could compress multiples quickly.

There is also a deeper structural risk that deserves more weight.

The neutral-architect thesis only holds if hyperscalers continue outsourcing critical integration layers. Yet Amazon, Google, and Microsoft are not just seeking independence from NVIDIA. They are steadily moving toward independence from everyone. In-house silicon and internal networking stacks remain a long-term destination, not a temporary experiment.

That tension does not invalidate the thesis, but it does limit its certainty.

Even so, the direction of travel is clear. AI systems are becoming more complex, not less. Interoperability challenges are increasing, not fading.

That environment favours companies positioned at the seams of the ecosystem.

The bridge may ultimately outgrow the castles

Broadcom remains the profitability king.

NVIDIA remains the compute king.

In technology, the business that controls the bridges can sometimes become just as important as the businesses that own the castles.

@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire