Marvell Q2 FY27 — Revenue Guide Up $1.5 Billion, Market Value Down $22.9 Billion. Read the Incremental Dollar.

The quarter, stated plainly

Net revenue was $2,739.3 million against $2,006.1 million a year earlier and $2,417.8 million in the prior quarter, up 37% and 13%, and $39.0 million above the mid-point of the guidance given on May 27, 2026. Data center was $2,171.5 million against $1,490.5 million, up 46% year over year and 18% sequentially, and 79% of total revenue against 74% a year ago. Communications and other was $567.8 million, up 10% year over year and down 3% sequentially. Non-GAAP gross margin was 58.9% on gross profit of $1,614.0 million. Non-GAAP operating expenses were $610.8 million, non-GAAP operating income $1,003.2 million, and non-GAAP operating margin 36.6% — up 180 basis points year over year and 160 basis points sequentially, which the CFO called “the significant operating leverage in our model.” Non-GAAP EPS was $0.94 against $0.67, up 40%, and against a $0.87 estimate.

Operating cash flow was $605.5 million against $461.6 million. Capital expenditure was $126.7 million against $47.5 million, leaving $478.8 million derived. Cash was $3,932.8 million against long-term debt of $4,962.9 million, net $1,030.1 million derived, which management framed as 1.32x gross and 0.27x net to EBITDA. The company bought back $200.0 million of stock and paid $53.9 million of dividends.

Then it raised, twice. Fiscal 2027 revenue to roughly $12 billion, approximately 45% growth, from approximately $11.5 billion a quarter ago — about $500 million. Fiscal 2028 to roughly $18 billion, approximately 50% growth, from approximately $16.5 billion — about $1.5 billion. Data center growth for fiscal 2027 went from approximately 50% to approximately 60%. Murphy's framing of the trajectory was plain enough: “we were an $8 billion company last year. We just took everybody to $12 billion this year.”

That is a very good quarter and a genuinely large raise. It is not the argument.

The incremental dollar

Every number in what follows is one Marvell printed. Third-quarter revenue is guided to $3,150 million. Third-quarter non-GAAP gross margin is guided to 57.5% to 58.5%, a mid-point of 58.0%, against the 58.9% just reported. Multiply the guide by its own mid-point and third-quarter non-GAAP gross profit is $1,827.0 million derived. Subtract the $1,614.0 million Marvell just earned and the increase is $213.0 million derived, on a revenue increase of $410.7 million derived.

That is an incremental gross margin of 51.9% derived on a business reporting 58.9%.

Marvell did not hide the reason and did not have to be pushed to it. Quinn Bolton asked about the 90-basis-point step-down directly, and Dan Durn answered: “Part of the acceleration story in Q3, as you point out, mix is the primary driver. We've got a strong ramp in custom. And so you can see that play out in the profiling of the margins.” Decode that. The fastest-growing piece of the fastest-growing segment ships at a lower gross margin than the company average, so the faster it grows, the more it pulls the corporate rate toward itself. This is not a one-quarter artifact and Durn said so about next year too: “My preliminary view is gross margins next year are going to be in a similar range, same range as we're exiting this year.” Exiting this year is 57.5% to 58.5%. The company is guiding, in its own words, to a fiscal 2028 that does not recover the 58.9%.

Put the full non-GAAP gross-margin path in one line: 59.4% a year ago, 58.9% last quarter, 58.9% this quarter, a guided 58.0% mid-point next quarter derived. Six-month margin 59.0% against 59.6%. Revenue accelerated through every one of those quarters. The margin went the other way through every one of them.

The raise, priced

Now decompose the raise itself, because a raise is not one number, it is two. Marvell added approximately $500 million to fiscal 2027 revenue. In the same breath it took fiscal 2027 non-GAAP operating expenses from approximately $2.45 billion to approximately $2.55 billion — approximately $100 million, and the company was explicit that the reason was “the significantly larger revenue opportunity we now see.” Decode the pairing. The opex raise is not a cost overrun and Marvell is not apologising for it; it is the stated price of the revenue raise, disclosed in the same paragraph, which means the two numbers are one decision and have to be read as one.

Run the two together at the gross margin the company guided. Five hundred million dollars of revenue at a 51.9% incremental gross margin is $259.5 million of gross profit derived. Less $100 million of new operating expense, that is $159.5 million of operating income derived — an incremental operating margin of 31.9% derived on the money Marvell just added to the year.

Marvell's reported non-GAAP operating margin is 36.6%. Its stated long-term target is 38% to 40%, which it says it enters in the fourth quarter. The revenue it added to the year drops through below both.

Be precise about what that does and does not mean, because there is an arithmetic trap here and it cuts my way only partly. The sequential step from this quarter to next is accretive: $1,827.0 million of gross profit less $655 million of operating expense is $1,172.0 million derived, a 37.2% margin derived, up 60 basis points, and the incremental operating margin on that specific step is 41.1% derived. Operating margin genuinely expands. Both things are true at once, and the distinction is the whole point: the quarter steps up because opex is growing slower than a very large revenue base, while the raise — the marginal decision to guide the year higher — carries economics beneath the model it is being credited for. Growth at 31.9% drop-through into a 38% to 40% target does not get you to 38% to 40%. It gets you there more slowly, and only because the base is doing the work.

For what it is worth, the guide is internally clean. Take the $1,172.0 million, subtract the $36 million of other expense management named, tax it at the 11% they gave, divide by 921 million diluted shares, and you get $1.0977 derived against a $1.10 mid-point. Nothing is being smuggled. The numbers say exactly what they say.

The fourth quarter inside the full-year number

One more subtraction, and it is the one that shows how much of this year is still forward. Six-month revenue was $5,157.1 million. The third quarter is guided to $3,150 million. Fiscal 2027 is now “roughly $12 billion.” Twelve thousand million less $5,157.1 million less $3,150 million leaves approximately $3,692.9 million for the fourth quarter derived — a sequential step of about 17.2% derived, on top of the 15% the company is guiding for the third.

Management did say the fourth quarter accelerates, so this is not a contradiction. It is a magnitude. Marvell is asking to be paid today for a fourth quarter that has to grow faster, sequentially, than the quarter it has actually guided — at a gross margin it has told you will not improve. The single-largest input to this fiscal year is a quarter that has not begun.

240 tranches of $500 million

This is the part I have not seen written down properly anywhere, and it is arithmetic, not opinion.

On July 29, 2026, Marvell entered a commercial agreement with Google LLC for custom semiconductor products. On August 18, disclosed by 8-K the following day, it issued Google a warrant for up to 58,970,907 shares at $206.58, roughly $12.18 billion at the strike, running to August 18, 2033. Of those, 1,360,867 vest in equal quarterly installments over the first year. The other 57,610,040 vest against Google's discretionary purchases from the third quarter of fiscal 2027 through fiscal 2033, in 240 equal tranches — one tranche for every $500 million of Custom Products revenue.

Two hundred and forty tranches at $500 million is $120 billion. That is where the headline number the analysts kept quoting on the call comes from. Harlan Sur put it as “a $120 billion in cumulative revs over 6 years,” Ben Reitzes divided it by 6.5 and got “about $18 billion a year,” and Murphy's answer was “your math is not wrong.” Decode what the $120 billion actually is. It is not a commitment, an order book, or a contract minimum. It is the denominator of a vesting ladder — the amount Google would have to buy for Marvell to have given away all 58,970,907 shares. It is the size of the prize stated as the size of the price.

Now price a rung. Fifty-seven million six hundred ten thousand and forty shares over 240 tranches is approximately 240,041.83 shares per $500 million of custom revenue derived. At Friday's close of $216.595, that rung is worth about $2,404,019 in intrinsic value derived — 0.48% of the revenue that earned it derived. At $300, the median of the eight sell-side targets published today, the same rung is worth about $22,424,708 derived, or 4.48% derived. At $400, about $46,428,892 derived, or 9.29% derived.

That is the structure, and it is worth saying slowly. The warrant is a volume rebate whose cost per dollar of revenue is set by Marvell's own share price. It is not fixed, not capped, and not on the income statement. It gets more expensive precisely when the thesis is working — when custom scales and the stock re-rates, existing owners pay Google more for each incremental dollar, not less. Call it the rebate that indexes to your own stock. Full exercise is approximately 6.4% dilution against 921.2 million diluted shares derived, taking the count to roughly 980.17 million derived — and 6.4% is what it costs at $120 billion of revenue, which is the number everyone is quoting as the bull case.

Two things I could not determine, and I am not going to pretend otherwise. Marvell has not disclosed the gross margin of the custom business separately from the corporate rate, so I cannot tell you what percentage of custom gross profit a rung represents rather than revenue. And the warrant appears nowhere in the Q2 FY27 earnings release — not the expense recognition, not the accounting treatment, not a sentence. Nobody on the call asked what it costs per dollar. Eight analysts got a question and every one of them spent it asking how big the opportunity is.

One correction to the easy version of this story, and it matters. The warrant was public on August 19 and the stock fell about 6% on August 21 on exactly this concern. The market had a full week. Today's 10.29% is not the warrant being discovered. It is the warrant's arithmetic being handed a revenue ramp to multiply against.

The strongest case against this read

Here is the bull argument at full strength, and it is not weak.

Marvell fell on a day the entire AI-infrastructure complex fell, two sessions after NVIDIA's own print. NVDA closed down 4.58%, ALAB down 4.81%, CRDO down 3.08%, AVGO down 0.77% derived. Nothing in a $500 million raise explains a sector. And the people who actually rebuild the models overnight did not blink: eight sell-side notes carried an August 28 timestamp, every single one maintained or reiterated, and there was not one downgrade. Three of them raised the target into the decline — Oppenheimer from $300 to $325, Morgan Stanley from $224 to $246, TD Cowen from $225 to $245. The median of the eight is $300 and the mean is $291.4 derived, against a $216.595 close. That is 38.5% of implied upside derived from a group that read the same disclosures I did.

The positioning read is at least as good. MRVL ran from $229.29 on August 24 to $245.11 on August 26, then opened the print day at $253.44 and made $254.60 — and closed that same session at $241.45, before a single number was released, on 41,014,669 shares. The stock had already given back 5% of the run-up before the print. On this telling, what sold was the run, not the thesis, and this is a record that sold off on expectations, not flaws — the same shape as AMAT in August. Buy the flush.

I take that seriously and I still think it is incomplete, for two reasons. MRVL fell 10.29% on a day its closest comparables fell 3% to 5%, closed $1.045 above the session low derived, and never had a bid after the first hour — open $225.20, high $228.88, low $215.55. Sector beta explains roughly half of that and leaves the other half unaccounted for. And the price-target raises are not the rebuttal they look like. Every one of them marks the numerator — the revenue line management just raised. Not one of them is a statement about the multiple. That is precisely the disagreement: the sell-side is marking revenue, and the tape is marking margin.

What this is, in the framework

This site has a standing prior for exactly this setup and it has now printed three times in three weeks. For the picks-and-shovels vendor, the funding question is a margin question. When the name is the vendor collecting hyperscaler capex rather than a levered operator spending it, the balance-sheet risk everyone looks for is not there — Marvell's is fine, $3,932.8 million of cash against $4,962.9 million of debt, 0.27x net. The risk reappears one line up, in gross margin, because the party paying for the growth is the vendor's own margin. Cisco's fourth quarter said it in August: orders and guidance up, non-GAAP gross margin down on AI-hardware mix, stock down on the print. Marvell's version of that sentence is 90 basis points and a 51.9% incremental.

The second prior is the discipline that keeps me honest here. Mix is the softest margin to capitalize, and it is direction-agnostic — a mix that deflates margin is exactly as provisional as one that inflates it. I am not entitled to floor Marvell's margin at 58.0% on one guided quarter any more than I would have been entitled to capitalize Super Micro's spike. Which is why the falsifier below is a margin line and a date, not a narrative.

What is genuinely new in this print is the third thing, and it is why I am coining it rather than filing it. In the Nvidia quarter the vendor started underwriting its customers with guarantees, terms, and equity stakes — the vendor becomes the credit. Marvell has done something adjacent and cleaner: it has paid its customer in its own equity, on a schedule that indexes to its own share price. Nvidia's version costs cash and shows up in receivables. Marvell's costs nothing until it works, and then it costs the most.

What settles it, and when

The date is Marvell's third-quarter fiscal 2027 report, which on the current cadence lands in early December 2026. The October 6 Investor Day is where the fiscal 2029 custom range gets sized, and it will move the stock, but it is a guidance event and guidance is not evidence.

Two things decide this, and both are lines Marvell will print.

• Wrong if Q3 FY27 non-GAAP gross margin prints at or above 58.5% — the top of the company's own range — on revenue at or above $3,150 million, and the fourth quarter is guided inside 57.5% to 58.5% with revenue at or above $3,692.9 million. That would mean the custom ramp does not dilute the way the guide implies, the incremental dollar is not 51.9%, and today's decline was sector beta plus a spent run-up. The multiple is the entry, not the problem.
• Right if Q3 FY27 non-GAAP gross margin prints at or below 57.5%, or the fourth quarter is guided below 57.5%. That is the custom ramp confirming as a revenue story that is not an earnings story, and every subsequent raise gets read the way this one was — as more volume at a worse price.

The bottom line is narrow and I will hold it. Marvell added $1.5 billion to next year and lost about $22.90 billion of market value derived in a session, roughly fifteen times the raise derived. That is not a market rejecting the demand. Demand is the one thing this print settled: data center up 46%, guided to roughly 75% next quarter, and the raise was broad-based across connectivity, switching and custom. It is a market marking down the quality of the dollar — 51.9% incremental gross margin, 31.9% drop-through on the raise, 11.9% of revenue in stock compensation against 7.7% a year ago, $326.2 million against $153.6 million, growing 112.4% against revenue's 37% derived, and a customer warrant that gets more expensive as the equity works. Marvell is winning the sockets. The question this print did not settle, and the one the Investor Day has to answer, is what the sockets are worth after everyone who helped win them is paid.