Key Takeaways:
Matt Murphy spent a decade rebuilding Marvell Technology around data centers. On Wednesday he told investors the payoff is now $10 billion larger than he thought a year ago.
The chipmaker's two-year revenue outlook now stands at roughly $30 billion, with sales of about $12 billion expected this year and $18 billion next, Murphy said at Citi's Global TMT Conference. Twelve months earlier the same two-year window was pitched at about $20 billion.
"This has been a company we built brick by brick, year-by-year," the chairman and CEO said, citing acquisitions, divestitures and organic investment since Marvell began orienting itself toward data centers in 2016.
Data-center revenue reached about $2.2 billion in calendar 2023 and is targeted at $15 billion to $16 billion next year, against 9% of fiscal 2017 revenue when the segment was tied mainly to enterprise servers and storage. Murphy attributed the increase to stronger demand, higher customer capital spending, product execution, connectivity growth and supply-chain expansion.
The stakes sit in the margin math. Marvell expects to exit the current year inside its long-term operating-margin target of 38% to 40% and to finish next year at the top end of that range, CFO Dan Durn said, even as custom silicon grows as a share of the mix. Operating expenses are expected to grow at roughly half the rate of revenue.
Scale-up networking, which links accelerators inside a rack or cluster, is the largest single opportunity Murphy described. Marvell's Teralynx switching platform, acquired through Innovium, is on track to exceed $1 billion in revenue — against an initial projection of about $150 million. The company is pursuing Ethernet-based scale-up, UALink and Nvidia's NVLink in parallel, betting that its electrical SerDes (the interface that moves data between chips) and optical portfolio let it sell into whichever standard wins.
"There is a lot of shots on goal for us on scale up networking," Murphy said. NVLink dominates today, but connecting rising counts of GPUs and XPUs should open a separate market over the next several years.
Optics is the swing factor. Marvell has funded near-packaged optics, or NPO, for at least five years, including through the Inphi acquisition, and added co-packaged optics capability with the December 2025 purchase of Celestial AI. Revenue from broadband analog, NPO and CPO products starts from effectively zero this year; the company had guided to $0 to $300 million next year, and Murphy said the figure is now "much higher." Management expects NPO and CPO to sit alongside copper interconnect rather than replace it outright.
In data-center interconnect, Marvell was first to announce 2-nanometer digital signal processors for 1.6T optical modules, a step up from the 800G generation now shipping, and cited 15 billion device hours of performance and reliability data. Murphy said "scale-across" architectures could expand that addressable market by two to five times. The company sells DSPs both inside its own modules and to third-party module makers, and is funding component and manufacturing capacity for both.
Marvell's custom ASIC business, built from the Avera acquisition and its ASIC team formerly associated with IBM and GlobalFoundries, now counts more than 20 design wins across the four largest U.S. hyperscalers, spanning XPUs and "XPU-attach" parts. Murphy described the recently disclosed Google relationship as an XPU-attach engagement covering network interface controllers, CXL, storage controllers, near-memory compute and AI acceleration, without giving a revenue timetable and pointing to the company's upcoming Investor Day for detail.
Durn said rising custom-silicon mix can modestly reduce gross margin, but argued the business still supports attractive operating margins because customers fund development work. Connectivity remains the core franchise. "At the end of the day, it always comes down to the I/O," Murphy said.
The supply chain is where the ambition meets a constraint. Marvell has secured longer-term arrangements with TSMC, substrate suppliers and backend capacity providers, and Durn said capital is being directed at potential bottlenecks before they limit growth. That matters because the same foundry and advanced-packaging capacity is contested by Nvidia, Broadcom and AMD, all of which are scaling AI accelerator programs on comparable nodes.
Broadcom, which reports earnings this week, has framed its own AI revenue path toward $230 billion in addressable opportunity, and Nvidia's data-center results have reset expectations for the entire interconnect supply chain. Marvell's $18 billion target is small against those numbers, which is the point: it is a connectivity and attach story, not an accelerator story.
For investors, the question is whether the revenue lands before the spending does. Marvell shares rose 4.26% on the session, while Broadcom slipped 1.13% and Nvidia fell 0.91%. Marvell's capital allocation priorities remain growth investment, a flexible balance sheet, and dividends and buybacks with excess cash. The company plans to update its longer-term operating-margin target at its Analyst Day, the next scheduled checkpoint for the $18 billion number.
This article is for informational purposes only and does not constitute investment advice.
