Marvell's CEO: Scale-Up Is a Greenfield Market Worth "Potentially Very Large" Revenues — and None of It Is in the Forecast Yet
Bank of America Global Technology Conference, June 3, 2026 — CEO Matt Murphy lays out the next five years and why the biggest opportunity isn't even in the numbers
Scale-Up: The Opportunity Wall Street Hasn't Priced In
The most important thing Marvell Technology CEO Matt Murphy said at the Bank of America Global Technology Conference on Tuesday afternoon was also the thing he said most casually: none of the scale-up opportunity is in any revenue forecast today. Not the switching. Not the optics beyond $300 million. Not the integrated XPU attach. Zero. For a company guiding to $16.5 billion in revenue next year and tracking toward an $8 to $10 billion custom silicon business by 2028, that framing is significant.
Murphy, who had just returned from a self-described "very calm trip" to Taiwan — a reference that drew laughs from moderator Vivek Arya of Bank of America — spent the better part of an hour walking through why scale-up switching is the cleanest incremental TAM opportunity in Marvell's history. "Scale-up switching is completely greenfield," said Ashish Saran, head of investor relations, who joined Murphy on stage. "We could be leading the market from day 1. That's a massive opportunity." Unlike scale-out switching, where Marvell entered a market that Broadcom already dominated and built from zero to a projected $1 billion-plus this year, scale-up has no incumbent.
Murphy was direct about the architecture agnosticism that underpins the strategy. "I'm investing in all of it. I have the R&D capacity to do this now. I don't have to take that bet." His argument is that Marvell can integrate the switch silicon, the optics — whether co-packaged (CPO) or near-package (NPO) — and the XPU or GPU attach into a single coherent platform. "That's not in any revenue forecast today," he said. And then, almost offhandedly: "just assume whatever the switch ASP is, that's the content opportunity on the optics side."
CPO Numbers Are Moving Fast
On co-packaged optics specifically, Murphy disclosed that Marvell's scale-up optics revenue target for next year has already doubled since the Celestial acquisition closed in December. The initial figure at the time of the deal was $150 million for 2027. It is now $300 million. "It's going from zero to $300 million with who knows where it could go," Murphy said. Celestial, which brought a lead customer and optical integration capability, accounts for roughly half. The rest comes from Marvell's own light engine and NPO work. The company has been in silicon photonics production for a decade, with 15 billion field hours and four generations of high-volume manufacturing behind it — a manufacturing depth that Murphy repeatedly framed as a competitive moat against startups and concept-stage competitors.
The NVIDIA Partnership Is Broader Than the Market Understands
Murphy addressed the NVIDIA relationship with more specificity than he has in prior settings. He described the partnership as rooted in years of pre-existing R&D collaboration — "I think it is even pre-COVID probably" — and said the formalization covers custom silicon enablement, NVLink Switch and NVLink Fusion adoption, joint silicon photonics development with the Celestial team, and work on 6G and AI RAN. "It's quite encompassing," he said.
On Jensen Huang's now-viral COMPUTEX appearance, where Huang called Marvell "the next trillion-dollar company" on stage without a script, Murphy said the comment was genuine. "You can watch the video. That was definitely not scripted." Murphy's own reaction was measured: "I said, we have a little work to do still." He noted that Marvell's enterprise value was $3.5 billion a decade ago and is roughly $270 billion today — approximately 80x in ten years — and declined to speculate on what comes next. "I don't know what's possible with Marvell."
Three New $1 Billion Businesses, All Landing Within the Next Year
Saran highlighted what he called an underappreciated element of the growth story: revenue diversification. Marvell has called out three distinct new $1 billion revenue lines all expected to materialize within the next twelve months — broadband analog and SiGe high-performance analog components, cloud switching, and DCI with early scale-across workloads. "This is before scale-up becomes really big," Saran noted. Murphy added context: the whole company was $4 billion in revenue just a few years ago. Each of these businesses carries multiple customers. By 2028, Murphy said, Marvell will have 15 to 18 custom silicon products in production, spread across XPU and XPU-attached sockets, with no single program representing a disproportionate share.
Custom Silicon: The Lesson Learned and the Reset That Followed
Murphy revisited the custom silicon narrative that created investor frustration in 2023 and 2024. His version of events: Marvell consistently guided custom silicon to $1 billion for calendar 2025, exceeded that by 50%, and never once told investors to model $3 or $4 billion. "People put stuff in their model and they're like, well, I thought it was going to be this. I said, well, I didn't tell you that." He acknowledged partial responsibility for the communication failure — "that's on us, too" — but described the underlying business as "an unbelievable success story" that got misread as a miss.
The September 2024 reset was deliberate. Murphy said he shifted to highly prescriptive guidance on custom silicon precisely to prevent expectations from running ahead of disclosed targets again. For this year, custom is growing more than 20%. For next year, it will more than double. "Don't go crazy, but more than double." The $8 to $10 billion custom silicon target for 2028, based on a 20% share of a $40 billion market, remains intact. "We're tracking to that," he said.
Supply Chain: Tight, But Planned For
On supply constraints, Murphy was candid that the environment is difficult across logic wafers, substrates, and back-end packaging, but argued that Marvell's long-range forecasting process insulates it from the worst allocation battles. "I forecasted $10 billion of growth to my suppliers back when I was $3 billion, $4 billion in revenue." He credited ASE CEO Tien Wu — who appeared on stage with Murphy and Huang at COMPUTEX, the first time Wu had done so — as a personal champion of the Marvell partnership. "He bet on me and Marvell like personally." The current supply position, Murphy said, is sufficient to support all disclosed guidance. The open question is whether incremental capacity can be secured if CapEx from hyperscalers runs hotter than the 30% growth assumption embedded in the $16.5 billion model for next year. "To the extent CapEx bias is higher than that, we would probably do better."
Cycle Durability: Murphy Sees Structural, Not Cyclical, Demand
Asked directly about cycle risk, Murphy argued the current AI infrastructure buildout is categorically different from prior semiconductor cycles. The constraining factor is supply, not demand — the inverse of the pandemic-era dynamic and of the consumer cycles that created previous boom-bust patterns. "We don't see it slowing down." He cited insufficient leading-edge logic wafer capacity, memory supply timelines, and global power grid and permitting constraints as natural governors on both supply and demand velocity. His long-term framing was unambiguous: "This is a true global infrastructure build on AI, on the scale of an industrial revolution type of event."