Marvell's 10% Drop Priced the Calendar, Not the Business

Marvell Technology reported its best quarter ever Thursday night, and the stock fell about 10% on Friday to roughly $217. Revenue was up 37% from a year earlier. The company raised its revenue outlook for this fiscal year and the next. And the shares still dropped — on the day after the biggest AI deal Marvell has ever signed. It reads as a contradiction: record numbers, falling price. It is not a contradiction. The number and the price are both honest; they are just measuring different things.

The first thing on the tape, the number: the AI engine is real and it is getting faster. Data-center chips — the custom accelerators, networking silicon, and optics that hyperscalers buy by the rack for AI — brought in $2.17 billion, 79% of the company, up 46% year over year. The quarter itself was a record at $2.74 billion, and the acceleration shows in the sequence: Marvell grew 28% in the spring quarter and 37% in the one just reported, and it is guiding the fall quarter to about $3.15 billion, up 15% from there. Margins are climbing with the mix, to a 36.6% adjusted operating margin, which is how management found room to raise the full-year numbers: about $12 billion for the fiscal year that ends this January and about $18 billion for fiscal 2028.

The second thing on the tape, the price: the drop was not about demand. It was about the calendar. In mid-August Marvell announced a custom-silicon agreement tied to Google's TPU accelerator line, and Google took a warrant for up to about 59 million Marvell shares as part of it — a structure whose theoretical revenue ceiling through fiscal 2033 runs to $120 billion. That headline is a big part of why the stock had nearly tripled in 2026 before this week. On the call Thursday night, the CEO walked the market back from it: some Google revenue is already inside the forecast through fiscal 2028, but the more meaningful contribution begins in fiscal 2029 — the year that starts around February 2028. Wall Street had wanted the Google dollars to load up the next two years. They will not. Friday was the market re-pricing the timetable, and the fine print is worth two words: discretionary. That $120 billion is a ceiling tied to purchases Google chooses to make, not a committed order book, and Google still works with Broadcom in the same stack.

So here is the part a fall-10%-on-good-news headline hides: a big drop on a record quarter is not automatically a gift, because the price had already eaten most of the good news. Marvell's market value is still near $190 billion. That is roughly 50 times the fiscal 2027 earnings today's run-rate points to, and even if fiscal 2028's $18 billion lands with margins intact, still about 30 times fiscal-2028 earnings. The harshest number is the cash, which is where this story has to prove itself: the record quarter produced $605.5 million of operating cash flow, so even annualized, free cash flow is only about $2 billion against that $190 billion market value — on the order of 100 times cash. And there is a gap inside the quarter worth naming: adjusted net income was about $866 million while operating cash came in below that. Some of it is working capital building for a 50%-growing ramp — the cost of growing this fast — but it is a reminder that the adjusted numbers run ahead of the cash, and the case only becomes financially real when the cash catches up.

What would make it real is exactly the bridge in those guides. The balance sheet is finally clean enough to let the economics flow through: no short-term debt, about $3.9 billion of cash against roughly $5 billion of long-term debt, net debt near $1 billion for a company of this size. Carry fiscal 2028's $18 billion at today's margins and convert it at anything close to a normal rate, and you get something like $4-5 billion of free cash flow — a roughly 40-times-forward-cash story instead of a 100-times-cash one. Still expensive. Still a growth premium. But a premium with a path, rather than a valuation with a hope.

State the risk inside that path plainly, because it is the whole bear case in one line: the custom-silicon ramp, and its timing. A handful of hyperscalers drives nearly all of it, the Google piece of it is discretionary, and one program's schedule can move the entire consensus. If the ramp slips a year again, you own a stock trading near 50 times earnings whose relief keeps moving further out — exactly the mistake the market just made on Friday.

I can be wrong again. But the reset is real: the market paid up for Google now and got Google in fiscal 2029, while underneath, the guides went up, the margin went up, the cash went up, and the debt stayed low. The market is still pricing the old fear — that the acceleration stalls once the near-term Google dollars disappoint — while the numbers already show the new story. The proof point is not the share price. It is fiscal 2028: the $18 billion, the margin it lands at, and the cash it converts into, with the company's investor day on October 6 as the first checkpoint. If the ramp delivers, about 30 times fiscal-2028 earnings is the going price of a compounding AI franchise. If it slips, the calendar — not this business — was the story all along, and there will be more air pockets before it lands.