Nvidia At 27x Versus Marvell At 72x: The Same Earnings Beat, Two Different Valuations

Marvell beat earnings. Nvidia beat earnings. They reported within 24 hours of each other, both raised their guidance, and both told versions of the same AI-capex story.

The market's answer was not the same.

Marvell shares fell more than 10% after its Q2 report. Nvidia's dipped briefly, then recovered on the strength of Jensen Huang's call commentary. Both reactions make sense if you look past the headline revenue beats and ask what each company is actually being paid for.

The question this comparison raises is not "which company is better." It is "what kind of investment is each one at this price?" — and the answer, as it turns out, is not what most investors assume.

What Nvidia delivered

Nvidia reported $96.2 billion in revenue for the quarter — up 18% from the prior quarter and 106% from a year ago. Data center alone generated $89 billion. Gross margins sat at 75%. Non-GAAP earnings per share came in at $2.22.

On a trailing basis, Nvidia generated roughly $127 billion in free cash flow over the last four quarters. It carries $22.4 billion in cash against $91 billion in debt, with a debt-to-equity ratio of 14.6%. Return on invested capital: 87%.

The company guided Q3 revenue to $108 billion, plus or minus 2%. Vera Rubin — the next-generation platform beyond Blackwell — has already entered full production with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle, and Nebius. Nvidia's Groq 3 inference accelerator is in full production. CEO Jensen Huang announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion of third-party capital for AI infrastructure.

None of that is a roadmap. It is a set of operating facts from a company that delivered the quarter it guided to.

Now look at the valuation. Nvidia's market cap sits at $5.25 trillion. But the trailing P/E — price divided by the actual earnings the company produced over the last four quarters — is 27x. By almost any definition in technology history, a company growing revenue at 106% with 75% gross margins at 27x trailing earnings is not expensive. It is, in fact, one of the cheapest ways to own high-quality, proven AI growth on the market today.

What Marvell delivered

Marvell reported $2.739 billion in Q2 revenue, up 37% year-over-year. Data center now accounts for 79% of the total — up from 73% a year ago. The data center segment grew 46% year-over-year. Non-GAAP EPS came in at $0.94.

The company guided Q3 revenue to $3.15 billion, which implies over 50% growth. It raised its full-year fiscal 2027 revenue outlook to approximately $12 billion, and its fiscal 2028 outlook to approximately $18 billion — up from $16.5 billion. Management expects the custom silicon business to more than double in fiscal 2028 and aims for non-GAAP operating margins to reach 38–40% by fiscal 2028, up from roughly 36.6%.

Then there is the Google deal. In mid-August, Marvell announced an expanded commercial agreement with Google for custom AI inference accelerators, storage controllers, NICs, and memory interface controllers. Google received a warrant to purchase up to $12.2 billion in Marvell shares — roughly 7% of the company — contingent on revenue milestones through fiscal 2033.

These are real numbers from a real business that is accelerating. The growth is genuine.

And then there is the price. Marvell trades at $190 billion in market capitalization. Its trailing P/E is roughly 72x — more than two-and-a-half times Nvidia's. On a forward basis, the multiple stretches to over 250x. The stock is up 155% year-to-date and 134% over the last four months.

Here is the tension most investors miss: the growth that justifies this price has not yet happened. The $18 billion fiscal 2028 revenue target is a projection, not a result. The custom silicon business — which management expects to drive the next wave of growth — is already showing margin pressure. The Google deal's revenue through fiscal 2028 is already baked into the current guidance. The bigger payoff, if it materializes, is fiscal 2029 and beyond.

Marvell is a company with $2.7 billion in quarterly revenue, a 36% operating margin, and an ROIC of 5.5% — that is to say, a business generating modest returns on its invested capital — being valued as if it were already delivering $18 billion in revenue at 40% margins. That is not to say the thesis is wrong. It is to say the thesis is what you are buying.

The architecture of the comparison

These two companies sit in the same industry but occupy fundamentally different positions in the AI value chain.

Nvidia sells the training and inference engines — the GPU systems where AI models are built and run. Its CUDA software ecosystem is the moat that locks in adoption. When hyperscalers buy AI infrastructure, Nvidia's chips are the default starting point. The company captures value at the top of the stack.

Marvell sells the infrastructure that connects those chips together — optical interconnects, switching, custom silicon, and the "XPU attach" chips that sit alongside accelerators. It does not make GPUs. It makes the parts that let thousands of GPUs work as one system, and it designs custom chips for hyperscalers who want alternatives to Nvidia. Marvell captures value in the supporting architecture.

Both positions benefit as AI infrastructure expands. But they benefit in different ways, at different margins, and with different customer dynamics.

Nvidia's customers buy because the workloads require Nvidia. Marvell's customers buy because they are building systems that need connectivity and customization — and those same customers, particularly Google, can decide tomorrow whether Marvell's custom chips or Broadcom's are the better fit. Marvell's data center revenue is 79% of the total, concentrated among a handful of hyperscalers. Nvidia's data center revenue is 93% of the total, but its customer base spans hyperscalers, startups, enterprises, and sovereign AI programs worldwide.

This matters because margin structure follows competitive position. Nvidia trades at 27x trailing earnings with 75% gross margins. Marvell trades at 72x trailing earnings with a non-GAAP gross margin of roughly 59% — and management signaled that the custom silicon ramp will put sequential pressure on those margins in Q3 as lower-margin custom programs scale faster than higher-margin optics and connectivity.

The growth rate difference tells another part of the story. Nvidia grew revenue 106% year-over-year. Marvell grew 37%. Both numbers are enormous in any other context. But Marvell's stock has run so far ahead of its current results that the market needs growth acceleration — not just growth — to justify the multiple. When Marvell guided to 50% growth for next quarter, the stock still fell. As one market observer put it, when a stock runs far ahead of results, even robust growth can disappoint if it does not exceed elevated expectations.

What the earnings reactions mean

The different stock reactions to two positive earnings reports are not a contradiction. They are the market telling you what it thinks each valuation should do next.

Nvidia's initial post-earnings dip reflected something familiar: the company has beaten estimates for eight straight quarters, and "beating" is now the price of admission. The stock recovered only after Jensen Huang spoke, characterizing AI as reaching an "inflection point". That recovery was not about the numbers — it was about the conviction that the numbers are not peaking. At 27x trailing earnings, there is room for the market to accept that conviction.

Marvell's decline was about the opposite problem. The company delivered a beat and a raise, but the stock had surged 48% in the month before earnings, up 155% year-to-date. The Google deal that drove much of that rally was announced on August 19, eight days before the earnings report — and the revenue from that deal through fiscal 2028 is already in the guidance. Investors who bought after the Google announcement and again before earnings paid twice for the same narrative. Marvell has now beaten consensus in six of its last seven quarters, and the average day-of price change over that stretch was minus 2%. The stock does not just need good news. It needs news that exceeds what has already been absorbed into a 72x P/E.

This is the practical lesson: a stock that grows 37% and trades at 72x is not being valued for what it is today. It is being valued for what it might become by 2028 or 2029. Every quarter between now and then, Marvell must grow fast enough and maintain margins well enough to justify a price that is already pricing in the future outcome.

So which one is the buy?

This is where the comparison becomes personal, not mechanical — because the answer depends on what kind of investor you are and what you believe the next 12 to 36 months will bring.

If you are looking for a position in AI growth where the earnings, cash flow, and margins are already real, Nvidia at 27x trailing earnings is the anchor. The company is producing $96 billion in quarterly revenue, $127 billion in trailing free cash flow, and guiding to $108 billion next quarter. The risk is not about whether the growth stops — it is about whether it slows enough to make the forward multiple uncomfortable, and whether competition in inference or from custom silicon chips gradually erodes the CUDA premium. But those risks are priced for a company that is currently printing record profits.

If you are looking for a bet on the next wave of AI infrastructure spending — specifically on the connectivity, custom silicon, and optical interconnects that scale as GPU clusters get bigger — Marvell has the positioning and the trajectory. The Google deal is a landmark contract. The custom silicon business could deliver $10 billion in revenue by fiscal 2029. The optical interconnect business is growing over 60%. But at the current price, you are buying the 2028-2029 outcome, and you need to be comfortable with the possibility that the execution does not match the guidance, that custom silicon margins stay lower than management targets, or that hyperscaler capex cycles turn down before the thesis lands.

The debate is not about whether either company stays important in the AI infrastructure build-out. Both are essential. The debate is about what you are being paid to wait, and whether the return profile of holding through the uncertainty matches what could be found elsewhere.

For most investors, the cleaner case is the one where the earnings justify the price today rather than in two years. That distinction matters far more than which company reports the bigger percentage growth in any single quarter.