Marvell Technology has repositioned itself as a high-beta play in AI infrastructure, with approximately 73% of its revenue now derived from data center workloads rather than legacy mixed-signal products. The company’s custom network interface controllers, memory controllers and optical interconnects are tightly integrated into hyperscale AI clusters, tying its performance directly to cloud operator capex and next-generation model buildouts. Industry forecasts suggest that custom accelerators (XPUs) will surpass GPUs in unit shipments by 2028, and Marvell’s focus on bandwidth, latency and power-efficient interconnects positions it to capture value in the data path surrounding compute dies.

In the quarter ended November 2025, Marvell reported revenue of $2.07 billion, up 36.8% year-on-year on a reported basis, and low-40% underlying growth after adjusting for the mid-2025 divestiture of its Automotive Ethernet unit. Data-center revenue climbed about 38%, while gross margin was 59.7%, down 80 basis points due to product mix and cost pressures. Disciplined operating expenses—up just 1.2% year-on-year with SG&A flat—drove a 660-basis-point expansion in operating margin. Net income was $1.90 billion, boosted by non-recurring items, and adjusted EPS rose roughly 76–77% to $0.76. Cash from operations was $582 million, free cash flow reached $1.42 billion (up 127%) and net cash increased by $1.49 billion, underscoring strong cash conversion in the AI cycle.

Marvell is bolstering its connectivity and memory strategy through two key acquisitions. The $3.25 billion purchase of Celestial AI adds a photonic-fabric specialist whose first-generation chiplet delivers 16 Tbps per device and is forecast to generate $500 million of annual recurring revenue by FY2028. The $540 million acquisition of XConn enhances PCIe and CXL switching capabilities, expected to contribute $100 million of revenue by FY2028. Together these deals extend Marvell’s portfolio from photonic interconnects through CXL-pooled memory and custom NICs, reinforcing its position in disaggregated AI data-center architectures.

Key risks include a broad AI sector multiple de-rating, potential slowdown in hyperscaler capex as data-center growth moderates from recent highs, and geographical concentration with roughly 40% of shipments destined for China. Execution risks around the ramp of Structura memory controllers—which target $2 billion of revenue over three years—and Celestial AI integration could also temper medium-term upside. Consensus models project normalized EPS of $5.00 by FY2028 and imply a fair-value target near $112.50, representing approximately 35% upside, based on a forward multiple around 22.5x and mid-20s revenue growth in AI infrastructure.