Marvell Raises the Bar Again: CEO Says Supply Chain Has Flexed 50% in a Year as Networking and XPU Attach Reshape the Growth Story
Citi Global TMT Conference, September 9, 2026
Marvell Technology used its lunch keynote at Citi's Global TMT Conference to lay out just how dramatically its near-term outlook has shifted, while positioning networking silicon and scale-up optics as the next major growth vector ahead of an October 6 Investor Day. CEO Matthew Murphy and CFO Daniel Durn fielded questions from Citi analyst Atif Malik, who opened by noting Marvell is the top-performing AI semis stock year-to-date and reminding the room that Jensen Huang has called it a future trillion-dollar company.
Supply Has Flexed 50% in a Single Year
The most concrete new data point from the session was Murphy's account of how quickly Marvell's supply chain has scaled. A year ago, at the same conference, he told investors Marvell could do $9.5 billion in fiscal 2027 and $11 billion-plus in fiscal 2028, a two-year total of roughly $20 billion. The company now guides to $12 billion this year and $18 billion next year, or $30 billion over the same window. "So our supply chain has been able to flex up 50% over that time on a graduated basis," Murphy said, attributing the raise to a stronger demand backdrop, higher capex from hyperscalers, and near-flawless execution, including first-pass silicon success on advanced nodes more than 90% of the time.
Murphy was candid about how Marvell secured that capacity: long-term agreements and prepayments struck five years ago with TSMC, substrate suppliers, and back-end assembly partners, following the Avera, Inphi, and Innovium acquisitions that pushed Marvell into reticle-sized, harder-to-source packages. "Suppliers are a lot like you guys," he told the audience. "They want to figure out who's the management, where is this company actually going... if they don't believe your story, you're not going to get what you need." He said he had just returned from a week visiting suppliers across Asia ahead of the Investor Day and came away with what he called a "resounding yes" on capacity commitments. He was equally blunt that constraints remain acute in some areas and that late entrants will struggle: "You can't get some of the substrate capacity until the end of the decade if you haven't started."
Scale-Up Networking Becomes a New, Largely Untapped TAM
Malik pressed Murphy on the scale-up networking opportunity, where standards including Ethernet-based ESUN, UALink, and NVIDIA's NVLink are competing for position. Murphy described Marvell's approach as a full-stack bet spanning switching silicon, copper SerDes, and both near-packaged optics (NPO) and co-packaged optics (CPO), built on the Teralynx switching platform acquired via Innovium five years ago. That business, he said, was projected at $150 million in revenue at acquisition and is now tracking to over $1 billion, with faster growth ahead. Marvell is also marketing NVLink as an NVIDIA technology partner following its investment and partnership announcement, while separately investing in UALink and its own Ethernet-based ESUN roadmap. "It's brand new," Murphy said of the scale-up market. "Today, it's really dominated just by NVLink, but in the next few years... it's going to be a new market and one where Marvell can really make a meaningful impact out the chute."
On the optics side, Murphy pushed back on the market's tendency to treat NPO and CPO as competing, mutually exclusive technologies. "I encourage all the investors here to get away from treating this as a binary thing," he said. "You're going to see all of these coexist in the market for several years." He noted Marvell's silicon photonics platform, developed through the Inphi acquisition and now combined with December 2025's Celestial AI deal, already generates roughly $1 billion in analog TIA and driver revenue supplying third-party NPO module makers, a base he expects to grow sharply as Marvell's own NPO and CPO solutions ramp starting next year and accelerating into calendar 2028 and 2029.
Google Warrant Deal Recast as Validation of XPU Attach, Not Concentration Risk
Murphy used the Google relationship, structured around the previously disclosed $120 billion warrant-linked revenue arrangement, to argue that Marvell's once-criticized customer concentration has evolved into a diversified, IP-rich franchise. He traced the custom silicon business back to the Avera acquisition, a roughly $300 million run-rate spinout of GlobalFoundries' IBM-derived ASIC team, which grew into a business overly reliant on a single large hyperscaler, "the curse of the mega socket," as he put it. The June 2025 Custom AI Day was Marvell's attempt to reframe the narrative around more than 20 design wins across all four U.S. hyperscalers, split between XPU and what the company terms XPU attach, chips such as NICs, CXL controllers, storage controllers, and near-memory compute silicon built to customer specification using Marvell IP.
The Google 8-K filing, Murphy said, "is really an XPU attach relationship" spanning the same product categories detailed a year ago, and he framed it as proof the category is real rather than a distraction from the core XPU story. "There's no one trick pony in there," he said, adding that Marvell is engaged across the full hyperscaler spectrum on similar attach opportunities that lack the public disclosure Google's warrant structure required. He reiterated the company remains on track for $10 billion to $11 billion in custom ASIC revenue by fiscal 2029, with custom sales set to double next year, while cautioning investors not to over-rotate back into single-socket anxiety: "I don't want anyone here to also get hung up that there's one big socket we got to go obsess over."
Margin Mix Shifts as Custom Scales, But Operating Leverage Holds
Durn addressed the structural gross margin dilution that comes with custom silicon growth, explaining that XPU attach programs carry a different margin profile than merchant products because Marvell customizes its own IP for customer environments, while full XPU programs are funded through customer non-recurring engineering payments that are accounted for as a contra-R&D offset. That accounting treatment allows operating margin to hold up even as gross margin compresses modestly. Durn said Marvell will exit this year at the low end of its 38% to 40% long-term operating margin target, touch the top of that range next year, and reset the long-term target itself at the Investor Day. Murphy added that despite custom revenue doubling next year, company-wide gross margin is expected to stay roughly flat versus the current exit rate, since the bulk of the recent revenue raise, from $13.5 billion to $18 billion for next year, has come from higher-margin interconnect, switching, and storage growth rather than custom alone.
Capital Allocation and the Next Growth Engine
Durn outlined a three-tier capital allocation framework: organic and inorganic growth investment first, balance sheet flexibility second, and shareholder returns via dividends and buybacks third, while noting that in the current environment additional capital is being directed toward supply chain prepayments to preempt bottlenecks. Asked what comes after the current XPU and XPU attach wave matures, Murphy pointed to the combined scale-up switching and optics attach opportunity as the next multibillion-dollar category, alongside continued growth in data center interconnect as it transitions toward "scale-across" architectures at 1.6T and beyond. "Don't count out the existing businesses growing at 50% a year plus off a pretty big base," he said, while flagging switching-plus-optics as the area he expects to become one of Marvell's largest growth engines over a three-to-five-year horizon.