While the rest of the chip sector was getting hammered on Thursday, Marvell Technology quietly dropped an earnings report that made Wall Street sit up straight. Record revenue, raised guidance, and a data center business growing so fast it now accounts for nearly three-quarters of the company’s total sales.
The numbers are hard to ignore. Marvell posted Q4 revenue of $2.22 billion — up 22% year over year — and full-year revenue of $8.2 billion, a 42% jump. Non-GAAP earnings came in at $0.80 per share, beating guidance by $0.10. But the real headline was the forward outlook: management raised its fiscal 2028 revenue target to approximately $15 billion, adding nearly $2 billion to the prior forecast and projecting non-GAAP EPS north of $5.
The engine behind all of this? Data center demand. Marvell’s data center revenue crossed $6 billion for the year, growing 46% and making up 74% of Q4 sales. CEO Matt Murphy said bookings in the segment are growing “at a record pace,” and the company now expects data center revenue to climb another 40% in fiscal 2027. The interconnect business alone is projected to grow more than 50% next year.
What’s especially interesting is the breadth of the growth. This isn’t a one-customer story. Marvell disclosed more than 20 design wins or product sockets entering production by fiscal 2028-2029 across the major hyperscalers. The custom chip business hit $1.5 billion annually and is expected to grow 20%+. Their AEC and retimer revenue is set to more than double. And the new Celestial AI acquisition is targeting $500 million in annualized revenue by 2028, scaling to $1 billion by 2029.
Supply chain? Locked down. COO Chris Koopmans said they’ve “secured the supply needed for all the growth” through next year and beyond. That’s not nothing in a semiconductor industry still dealing with capacity constraints.
Marvell shares jumped over 11% in after-hours trading. While everyone’s been focused on Nvidia and Broadcom, Marvell has been quietly building what looks like the connective tissue of the AI infrastructure buildout — the switches, the optical interconnects, the custom silicon that hyperscalers need to actually make their trillion-dollar AI investments work. The stock isn’t cheap, but neither is the growth runway they’re describing.
