Marvell Q2 FY2027: The Incremental Gross Margin on the Q3 Guide Is 52%, Not 58%

Marvell reported fiscal second quarter revenue of $2.739 billion against roughly $2.71 billion of consensus, $39.0 million above the midpoint of its own May guidance, with non-GAAP earnings of $0.94 against $0.93. Data center revenue reached $2.17 billion, up 46% year over year and 18% sequentially, now 79% of the company. Non-GAAP gross margin was 58.9% and non-GAAP operating margin 36.6%. The third quarter guide is $3.150 billion plus or minus 5%, about 15% sequential and better than 50% year over year, with non-GAAP earnings of $1.05 to $1.15. Full-year fiscal 2027 revenue moved to approximately $12 billion from $11.5 billion and fiscal 2028 to approximately $18 billion from $16.5 billion, the second consecutive quarter in which both years were raised. The stock closed the session at $241.42, down 1.5%, and traded to $224.34 after hours, off 7.1% from the close and 8.5% from Wednesday's $245.11. It remains up roughly 184% year to date against a 52-week range of $61.44 to $329.88 and a market capitalization near $212 billion.

The headline number is the $18 billion. The number that decides whether the $18 billion is worth what the market was paying for it sits inside the third quarter guide, and nobody quoted it.

Q3 revenue is guided $411 million above the June quarter. Non-GAAP gross margin is guided 57.5% to 58.5% against the 58.9% just delivered. At the midpoint that produces $1.827 billion of gross profit versus $1.613 billion, so the incremental gross margin on the sequential dollar is 52.0%. At the low end of the band it is 48.2%. At the high end, 55.8%. Every point of that range is below the corporate average, which is the arithmetic definition of a mix that dilutes as it grows. Management named the cause directly: custom silicon acceleration in the second half of fiscal 2027 is a sequential margin headwind. What the guide adds is the magnitude. The marginal dollar of Marvell revenue in the current quarter is worth roughly nine tenths of what the average dollar was worth in the last one.

This matters because the growth being celebrated is precisely the growth that carries the lower margin. Custom is guided to more than double year over year in fiscal 2028. The company's standing target is more than $10 billion of custom revenue in fiscal 2029, which management now says carries upside bias after the Google agreement. Run that mix forward and gross margin does not dip and recover, it grinds. The offset is operating leverage: non-GAAP operating expenses are guided to approximately $655 million in Q3 against an implied $611 million in Q2, so opex grows 7% while revenue grows 15%, and operating margin expands to roughly 37.2% at the midpoint even as gross margin contracts. Management expects to enter the 38% to 40% long-term band in the fourth quarter and to sit near the top of it in fiscal 2028. That is the entire bet in one sentence. Marvell is trading a structurally declining gross margin for a structurally rising operating margin, and the trade works only while revenue growth outruns the fixed cost base by eight points a quarter. It is a volume argument dressed as a margin argument.

The second thing the market appears to have priced and the recaps missed is where the fiscal 2028 raise came from. The obvious reading is that the $1.5 billion increase reflects the Google warrant deal announced eight days earlier. Murphy said the opposite on the call: the raise is broad-based, some of it is custom, and the most meaningful component is scale-up optics and switching. Warrant-related revenue is already partly embedded in fiscal 2028, and the large contribution arrives in fiscal 2029 and beyond. Both halves of that answer cut against the position going into the print. If the raise is mostly optics and switching, then the largest single guidance event of the year owes little to the deal that added roughly forty points to the stock in three sessions, and the Google program has not yet produced a number anyone can model. If the raise had been mostly custom, the margin arithmetic above would be worse. There was no version of that answer that helped a stock at 39 times forward earnings.

The warrant itself remains the underpriced variable. Google holds the right to 58,970,907 shares at $206.58, close to 7% of the company, vesting in 240 tranches of about 240,042 shares, one per $500 million of custom product revenue recognized through fiscal 2033. The strike was in the money the day it was issued and it is in the money now. Every $500 million of the revenue that validates the thesis issues shares against it. Marvell has moved its customer acquisition cost off the income statement and onto the capitalization table, where it will never appear in a gross margin line and never be discussed as a margin item. Apply it to the fiscal 2029 custom target and the arithmetic is unforgiving: $10 billion of custom revenue, if a meaningful share of it is Google, vests a large block of stock at a price the market has already left behind. The $120 billion ceiling is a ceiling, not a forecast, but the dilution mechanism does not care about the ceiling. It runs on the actual dollars.

The moat argument has not changed and it survives all of this. Marvell's defensible position is SerDes and optical DSP, the 800G and 1.6T interconnect ramp, and the 51.2T switching franchise, all of which attach to accelerators regardless of who designs the accelerator. Interconnect is guided above 70% growth this year and scale-out switching is projected to exceed $600 million, roughly double fiscal 2026. Attach revenue survives a hyperscaler deciding to bring its compute die in house; the compute die does not. Nvidia's roughly 2.5% convertible preferred position is best read as buying exposure to Marvell wins that contain no Nvidia silicon. The complication introduced by the warrant is that the Google agreement covers mostly attach products, inference accelerators alongside storage controllers, NICs, memory interface controllers and near-memory compute. Marvell has just sold the most durable part of its franchise into a single account at a seven percent equity discount. That is a defensible trade for volume and a validation of the attach category management pioneered. It is also the first time the moat has been priced, and the price is now public.

Against a covering base of 38 analysts, no sells, an average target near $257 and a spread running from $126 to $400, the disagreement was never about this quarter. It is about the terminal number, and it resolves on October 6 in New York, not on August 27. Base case is $220 to $270 into the investor day, the range the stock has held through two guidance raises. Bull case is a retest of the $329.88 record if October 6 raises the fiscal 2029 custom figure above $10 billion and, more importantly, splits the socket count into compute versus attach so the durable revenue can be valued separately from the revenue that depends on a hyperscaler not changing its mind. Bear case is cohort derating with Broadcom and Astera Labs to $150 to $175, on the market deciding that a business whose incremental gross margin is 52% and falling, whose growth is concentrated in one account, and whose capital intensity now includes roughly $1 billion of fiscal 2027 supplier prepayments, belongs at a hardware multiple rather than a platform multiple. That is not a Marvell-specific outcome and it would not require the fundamentals to deteriorate at all.

The decision-relevant number is the incremental gross margin implied by the January quarter guide when it is issued in December. Custom is accelerating through the second half by management's own timeline, so the mix pressure should be worse in Q4 than in Q3. If the January guide implies incremental gross margin above 50%, the mix headwind is a transition cost and the operating leverage argument holds. If it implies something in the low forties, the $18 billion arrives with materially less profit attached than the market has been underwriting, and the October investor day will have been asked to defend the wrong number.