Analysis
Marvell reported record fiscal second-quarter revenue of $2.739 billion, up 13% sequentially and 37% year over year, with data center revenue of $2.17 billion -- 79% of the business -- growing 46%. Non-GAAP EPS came in at $0.94, a penny ahead. The company raised its full-year outlook to roughly $12 billion and lifted its fiscal 2028 target to about $18 billion from $16.5 billion, implying roughly 50% growth. The stock fell 6%, CNBC reported.
I think this is the most useful print of the week, and not because of Marvell. Guiding to 50% growth and getting sold is the market telling you what is already in the price. Nvidia set the bar the day before with a quarter that sent it up nearly 9%; every custom-silicon and networking name now gets measured against that, and "beat and raise" has quietly stopped being sufficient. Marvell's custom ASIC business -- the XPUs it designs for hyperscalers alongside its optics and networking silicon -- is precisely the position investors said they wanted a year ago when the argument was that merchant GPUs would lose share to custom accelerators. The company delivered the thesis and the multiple compressed anyway.
What that says about the AI trade is more interesting than what it says about Marvell. Expectations have moved from growth to acceleration. Semis are being priced as though revenue must beat a rising forward curve every quarter, which is a structure that only breaks in one direction. And the margin question underneath is real: custom silicon carries lower gross margins than merchant parts, so a mix shift toward ASICs grows revenue while diluting the profitability the multiple was built on.
“Guiding to 50% growth and getting sold is the market telling you what is already in the price.”
For private-market founders and their investors, the read-through is direct. Public comparables set private marks with a lag. If a company growing 37% with 79% data center exposure and a raised long-term target trades down, the private AI infrastructure companies raising at 25-40x forward revenue are underwriting a public bid that just got choosier.
Room for disagreement: a 6% move after a large run-up is noise, not a verdict. Marvell had rallied hard into the print, and single-day reactions to guidance are among the least predictive signals in equities. Broadcom, the far larger custom-silicon incumbent, has repeatedly sold off on results and then made new highs within weeks. If the fiscal 2028 target holds through two more quarters, everyone arguing about Friday's tape will look like they were watching the wrong number.
The number I will be tracking is gross margin as the ASIC mix rises. Revenue growth is settled; margin is not.