Constructive on the business, cautious on the price and the timing. Nothing in this quarter broke the thesis — but the stock is now priced for an outcome management explicitly declined to quantify for another six weeks. Confidence: moderate.
The quarter
Revenue grew 37% year over year to a record $2.739 billion. Data center revenue grew 46% to $2.17 billion and now makes up 79% of the company, up from roughly 74% a year ago.
The number that matters most: both the sequential and year-over-year growth rates accelerated from the prior quarter, which ran +11% sequential and +27% annual. Going from +45% growth on an $8.2B base to +50% on a $12B base is genuinely hard, and it is the strongest argument in the bull case.
Operating leverage is working too. Non-GAAP operating margin reached 36.6%, up 180 basis points year over year and 160 sequentially. Management expects to enter its 38–40% long-term target range in Q4 and reach the upper end during FY2028.
| Metric | Q2 FY27 | Year over year | Sequential |
|---|---|---|---|
| Revenue | $2.739B | +37% | +13% |
| Data center | $2.17B | +46% | +18% |
| Comms & other | $567.8M | +10% | −3% |
| Non-GAAP gross margin | 58.9% | — | +10bps |
| Non-GAAP operating margin | 36.6% | +180bps | +160bps |
| Non-GAAP EPS | $0.94 | +40% | +18% |
| GAAP EPS | $0.33 | — | — |
| Cash from operations | $605.5M | — | slightly down |
The sell-off
Three things happened on the conference call. None of them appear in the headline numbers.
Q3 was guided to 57.5–58.5%, a 90 basis point sequential decline. Then the CFO said Q4 lands in the same range. Then he said FY2028 lands in "the same range as we're exiting this year."
Read that again. Revenue goes from roughly $2.7B a quarter to roughly $4.5B a quarter — a 50%+ increase — and gross margin never expands. Not once.
The stated reason was mix. Custom silicon is ramping hard, and custom carries structurally lower margins than Marvell's merchant connectivity products — it is closer to IP-plus-foundry economics than a proprietary chip franchise. The growth is real, but it dilutes the margin structure, and the entire operating-margin story now rests on opex discipline rather than pricing power. That is a thinner cushion: if revenue comes in light, there is no gross margin buffer to absorb it.
Analysts from Barclays, JPMorgan and Melius each pushed for the FY2029 custom silicon number in light of the Google agreement. The CEO confirmed direction — "a lot of upside bias," "definitely goes higher" — but declined to size it, pointing to the Investor Day.
For a stock at these multiples, that is an unsatisfying answer. The market had already paid for the Google deal in the +11% move on August 19. It wanted the number and did not get one.
Marvell went into this print up roughly 184% year to date, at about 84x trailing earnings, with a momentum score in the 98th percentile and a value score in the 1st. NVIDIA had guided FY2028 growth of about 70% the night before; Marvell's ~50% looks slower by direct comparison, fairly or not. Costs also crept: FY27 opex guidance rose to ~$2.55B from ~$2.45B, and the FY28 tax rate steps up from 11% to 13%.
Volatility
This is not a smooth uptrend. In the last ten weeks the stock made an all-time high of $329.88 on June 18, fell to $162.90 by July 29 — a 50.6% drawdown in six weeks — then rebounded 49%, and is now giving some of that back. Past earnings reactions have ranged from +18% to −20%. Weekly ranges of 15–25% are routine.
The Google warrant
Disclosed August 19 by 8-K with no press release, tied to a commercial agreement signed July 29: Marvell granted Google a warrant for 58,970,907 shares at $206.58 — about $12.18 billion at the strike, or roughly 7% of shares outstanding, exercisable through 2033.
A small time-based tranche vests quarterly over year one. The remaining ~57.6 million shares vest in 240 tranches of about 240,000 shares each, one per $500 million of custom product revenue recognised between Q3 FY2027 and FY2033.
That is where the widely quoted "$120 billion" comes from — 240 × $500M. It is worth being precise about what that means:
The warrant is a real positive and genuine validation of the XPU-attached category Marvell pioneered. It is not $120 billion of new revenue, and treating it that way is how people get hurt in this name.
Valuation
Building from the guidance rather than consensus: the implied Q4 FY27 quarter is ~$3.69B of revenue at ~58% gross margin and ~39% operating margin, which produces roughly $1.36 in EPS and about $4.20 for FY2027. Rolling FY2028 forward at $18B, 58% gross margin, opex growing at half the revenue rate and a 13% tax rate gives roughly $6.45.
| Price | FY2027 P/E | FY2028 P/E | |
|---|---|---|---|
| $241.45 | ~57× | ~37× | Last close |
| $222.19 | ~53× | ~34× | After hours |
| $206.58 | ~49× | ~32× | Warrant strike |
| $185.00 | ~44× | ~29× | Where I'd get interested |
For a business compounding revenue at 50% with expanding operating margins, 34× forward-forward is not outrageous — roughly a 0.7 PEG. Broadcom trades near 20× forward and NVIDIA near 26× trailing, so Marvell carries a real premium, though it is growing faster than both from a smaller base.
The problem is not the multiple. It is that the multiple is applied to a number eighteen months out that requires flawless execution across supply, three separate hyperscaler custom programs, and a scale-up optics business that barely has revenue yet.
The trailing P/E of ~84× is flattered by a one-time gain. Trailing GAAP earnings include roughly $1.9B from the automotive ethernet divestiture — that quarter shows a 91.7% net margin, which is obviously not operating income. Strip it and clean trailing GAAP earnings are around $735M, putting the honest trailing GAAP multiple closer to 290×.
Second, the GAAP-to-non-GAAP gap is wide: $996M of GAAP operating expense versus $611M non-GAAP, a $385M quarterly difference — about 14% of revenue — from stock compensation, intangible amortisation and acquisition costs. Non-GAAP is the right lens here, but GAAP profitability is a fraction of the headline.
The case against
Decision framework
Conclusion
Marvell's business improved this quarter. Revenue accelerated, data center accelerated, operating margin expanded 180 basis points, and management raised guidance for the second consecutive quarter — by $500M for this year and $1.5B for next. That is not what a deteriorating fundamental looks like.
What deteriorated was the quality of the projected growth. Gross margin is flat-to-down for six-plus quarters, opex is rising faster than previously communicated, and the incremental bull case rests on an FY2029 number management deliberately withheld until October.
Selling an 8% gap-down after a beat-and-raise is usually the wrong side of the trade. But at 34× FY2028 with 7% dilution and no gross margin expansion, this is not a valuation opportunity either — it is a momentum stock with a real business attached, and it should be sized accordingly. If you want to add, scale in and keep meaningful capital available for October 6, which is the next real catalyst in either direction.
| Date | Event |
|---|---|
| 2 Sep 2026 | Broadcom FQ3 earnings — direct read-through on custom silicon |
| 6 Oct 2026 | Investor Day, New York — long-term model reset, FY29 custom sizing. The catalyst. |
| 1 Dec 2026 | Marvell Q3 FY2027 earnings (tentative) |
Street consensus: 23 Buy / 4 Hold / 0 Sell, average target $295.42. Recent targets into the print: Wells Fargo $310, Rosenblatt $300, Citigroup $275, Susquehanna $265, Morgan Stanley $224 — the notable dissent, and the closest to where the stock is trading tonight.