Expand menu

On August 27, Marvell reported the best quarter in its history. Net revenue of $2.739 billion, up 37% year over year. The data center segment grew 46% to $2.171 billion, roughly four of every five dollars the company takes in. Guidance for the current quarter is $3.15 billion, plus or minus 5%. On the analyst call, management raised its fiscal 2028 outlook from about $16.5 billion to roughly $18 billion. The figures are in the 8-K filed with the SEC.

The next day the stock fell. CNBC reported a 6% decline during the session; The Motley Fool put it at 10%, and at about 8% in after-hours trading right after the report. Whichever number you take, the direction is the same: record revenue, immediate punishment.

What moved the price is not on the revenue line. It is in gross margin. Non-GAAP gross margin for the quarter was 58.9%, against 59.4% a year earlier. Guidance for the current quarter takes it down to a range of 57.5% to 58.5%. CFO Dan Durn explained it without decoration: revenue levels and product mix, with the acceleration of custom designs for cloud customers as the driver.

Half a percentage point. That was enough.

First, a correction about the $120 billion

Two things got blurred together this week. On August 19, Marvell announced an agreement with Google that headlines everywhere framed as "up to $120 billion." That number is not contracted revenue. It is the level of cumulative purchases Google would have to reach for full vesting of a warrant covering up to 58.97 million shares at $206.58, expiring in August 2033.

Put differently: the big number describes how much Google would have to buy in order to dilute Marvell's shareholders, not how much Marvell will bill next quarter. Repeating that $120 billion as if it were signed revenue is exactly the kind of mistake that forces you to delete the post afterward.

Why it matters

With that clear, the market's reaction makes sense. Marvell proved it can win the largest custom engagement in its industry. And in proving it, it also showed what that kind of work costs.

A chip designed for a single customer is sold once, to a buyer who knows precisely what it is worth and has the leverage to negotiate it. A standard product is sold many times and spreads its engineering cost across every buyer. The difference between those two economics is, precisely, gross margin. Marvell is growing toward the side that pays less per unit of effort, and it is doing so deliberately.

What this means if you run a technology company

This is not a semiconductor problem. It is the arithmetic of any company that sells custom work, and I live it every month at Indrox.

The big project always looks good in the meeting. It lifts revenue, fills the pipeline, buys peace of mind for the year. What that meeting does not show is that a large client arrives with large-client conditions: scope that moves, integrations nobody mapped, senior people tied up for months, and a price negotiated from a position of strength. Billings go up and margin erodes. Same as Marvell, three zeros down.

The difference between the two companies is that Marvell has a public market that prices that erosion within twenty-four hours. A services firm does not. It can grow for two straight years with margin sliding and only find out when cash flow stops covering the month.

Three criteria come out of that, and I would ask them of anyone running a technology firm:

Measure margin per project, not just total billings. The company average hides the project that is eating the team.

Decide explicitly which part of what you build to order becomes product. Marvell has both legs: the custom business that grows fast and compresses, and the standard portfolio that holds the margin up. If all of your revenue is bespoke, you do not have that second leg, and your margin floor is set by your largest client.

Charge for concentration. When a single client weighs too much, the volume discount they ask for is, in practice, a risk premium you are handing them for free.

My read

Friday does not strike me as bad news for Marvell. This is a company that chose to trade margin for scale in a market where scale is growing 46% a year, and the market billed it the same day it understood the trade. That is a defensible choice, and they defended it by name on the call.

What is not defensible is making the same trade without having decided to. Most services firms I know did not choose to lower their margin. It slid, one large project at a time, while the sales report looked better and better.

Marvell, at its scale, had the discipline to say out loud where its margin is going and why. That is the standard, and it does not depend on the size of the company. If you cannot name today which of your projects is compressing you, you are already paying for it and you just do not know yet.