Marvell Investor Day: Connectivity Lifts Outlook to $20 Billion

I attended Marvell Investor Day in New York on October 6, 2026. Marvell Technology (MRVL), a designer of data infrastructure chips, set a long-term target of $70 billion to $90 billion of revenue for fiscal 2031, which roughly matches calendar 2030. It also raised its fiscal 2028 outlook to about $20 billion and sharply increased its estimate of the market it can serve, to about $400 billion by 2030. Marvell is a constituent of our ROBO Global Artificial Intelligence Index, in its Semiconductor subsector.Key Takeaways Marvell Technology (MRVL), a THNQ constituent, set a target of $70 billion to $90 billion of revenue for fiscal 2031 at its October 6, 2026, Investor Day, and raised its fiscal 2028 outlook to about $20 billion, from about $18 billion in August. Both are management forecasts for years ending in early 2031 and early 2028. Marvell sells custom processors, optical connections, and memory products that ease bottlenecks in AI systems. It says all five U.S. hyperscalers — the largest cloud operators — buy its modules linking separate data centers, and it sizes its potential market at about $400 billion by calendar 2030. Marvell’s acquisitions of Celestial AI and XConn both closed in February 2026, and Polariton — announced in April 2026 — add optical links between AI chips, memory-sharing switches, and faster optical components. Management expects optics that link processors inside AI systems to begin growing strongly in fiscal 2028 and puts plasmonics products, the technology Polariton brings, three to five years out. See more: ROBO Global Artificial Intelligence IndexWhat changed at Marvell Investor Day?Marvell’s new long-term target calls for $70 billion to $90 billion of revenue in fiscal 2031. The $80 billion midpoint is about 10 times fiscal 2026 revenue of $8.2 billion. Nearer term, Marvell’s fiscal 2028 revenue outlook rose to about $20 billion, from about $18 billion in August. Management also raised its fiscal 2029 custom chip target to more than $12 billion, from more than $10 billion, mainly on demand for chips that surround AI processors. All are management forecasts, dependent on customer spending and product execution. Marvell now sizes the chip market its products can compete for at about $400 billion by calendar 2030. Chairman and CEO Matt Murphy said that is about four times the last estimate Marvell gave investors. He credited the higher fiscal 2028 outlook to optical links and switch chips, saying, “It’s connectivity leading the race.”What does Marvell supply to AI data centers?Custom processors are chips built for one cloud operator’s workloads. XPU is the industry’s umbrella term for such AI accelerators. Marvell designs them with customers, then sells the finished chips along with custom chips that surround them. Switch chips route data to its destination. Marvell’s optical chips run the modules that carry data as light over fiber. Its Ara chips, which the company says are shipping in volume, clean up signals inside modules that move 1.6 trillion bits per second. The Ara line I saw on stage spans several versions, each tuned for a different job, such as lower power, link reliability, or built-in security. Marvell’s memory controllers and switches set how much data processors can reach, and how fast. Idle processors waiting on data are costly, especially during inference, when a trained AI model answers questions or takes actions.What do scale-up & scale-out mean?Scale-up connects processors so they work as one giant machine. Today that usually happens inside a rack, the cabinet holding computing equipment, and increasingly across several racks. Scale-out networks many of those machines into a cluster, like crews across a factory floor. Scale-across connects separate data centers, like factories at different sites. Co-packaged optics (CPO) places optical components inside the same package as a computing or switching chip, cutting the power needed to move data. Marvell said customers are deploying its CPO today. It also offers near-package optics, which sit just outside the chip package and, Marvell says, can be swapped out if they fail.Marvell customers & the NVIDIA partnershipMarvell says it supplies data-center interconnect modules, its scale-across products, to all five U.S. hyperscalers. Russ Esmacher, who runs the broader data-center interconnect business, expects its revenue to grow from roughly $500 million in fiscal 2026 to more than $1 billion in fiscal 2028. NVIDIA (NVDA), a THNQ constituent that designs AI processors and systems, said on March 31, 2026, that it had invested $2 billion in Marvell as part of a strategic partnership. Marvell will supply custom processors and networking that work with NVIDIA’s NVLink Fusion platform, so customers can combine their own chips with NVIDIA equipment.What Celestial AI, XConn & Polariton add to MarvellCelestial AI, whose acquisition closed in February 2026, brings Photonic Fabric — optical technology that links AI chips across racks — beyond the reach of copper wiring. Marvell has merged Celestial’s team with its own optics engineers. Dave Lazovsky, Celestial’s former CEO and now a Marvell executive, said scale-up optics could amount to thousands of dollars of spending per AI processor. He added that Marvell has design wins with leading hyperscalers. That means customers have selected the product, though production orders come later. Murphy said Marvell has virtually no scale-up optics revenue today and expects strong growth to begin in fiscal 2028. Photonic Fabric also targets memory. High-bandwidth memory sits about 1.5 millimeters from the AI processor, inside the same package, so adding memory usually means adding processors. Lazovsky said this turns AI processors into “the world’s most expensive memory controller.” Marvell’s Photonic Fabric memory module is designed to give many processors optical access to a shared memory pool. Marvell cites up to two to three times more AI output from the same footprint and power, and customer samples planned for 2027 will test that claim. Marvell’s Structera S memory switch, announced in March 2026, uses switch technology from XConn Technologies, acquired a month earlier. It is designed to pool memory across a rack, letting cloud operators get more from installed memory while supply is tight and prices are high. XConn’s engineers also work on Marvell’s UALink switches, built on an open standard for linking AI processors. In April 2026, Marvell announced the acquisition of Polariton Technologies, a Swiss developer of plasmonics. The technology builds tiny, fast, low-power modulators, the parts that encode data onto light. Marvell said it is investing in plasmonics for scale-across links at 3.2 trillion bits per second and beyond, with products three to five years out. For Marvell, the milestones to watch are customer qualification, production shipments, and recognized revenue from custom processors, optical connections, and memory products. Qualification establishes whether a product meets customer requirements. Shipments and revenue show commercial delivery and help assess progress toward Marvell’s fiscal 2028 outlook and fiscal 2031 ambition. THNQ is the underlying index for the ROBO Global Artificial Intelligence ETF (THNQ) and the L&G Artificial Intelligence UCITS ETF (AIAI.LN). Looking for regular updates? Subscribe here for weekly insights on robotics, AI, and healthcare technology, delivered straight to your inbox. For more news, information, and analysis, visit the Artificial Intelligence Content Hub. VettaFi LLC (“VettaFi”) is the index provider for THNQ and AIAI.LN, for which it receives an index licensing fee. However, THNQ and AIAI.LN are not issued, sponsored, endorsed, or sold by VettaFi, and VettaFi has no obligation or liability in connection with the issuance, administration, marketing, or trading of THNQ or AIAI.LN.