The broad AI trade is over. What has replaced it is a narrower bet on companies converting surging infrastructure spending into hard revenue, and Marvell Technology is the clearest example of how that plays out in a stock price. Shares of the Santa Clara-based semiconductor designer have risen 131 percent year-to-date, a gap so wide against the broader market that it demands an explanation beyond momentum.
Marvell designs chips used in data centers, cloud computing, AI networking, storage and telecommunications. Its customers are hyperscale cloud operators, and those operators are spending without restraint on AI infrastructure. The company's first-quarter data center revenue came in at $1.8 billion, up 27 percent year-over-year, driven by optical interconnect products, custom silicon programs and Ethernet switching solutions.
A few months ago, Marvell's management had guided for high-single-digit sequential revenue growth through fiscal 2027. That forecast is no longer operative. The company now expects second-quarter revenue of $2.7 billion, a 10 percent sequential increase and 35 percent growth year-over-year at the midpoint. By the third quarter, management expects revenue to hit $3 billion. Full-year revenue is now projected to grow 40 percent year-over-year to nearly $11.5 billion.
The revision is not the result of one strong quarter. Robust bookings across the data center portfolio pushed results above expectations, and the forward guidance reflects a structural change in demand rather than a one-time pull-forward. Data center revenue is expected to grow 50 percent year-over-year in fiscal 2027 and then accelerate to roughly 55 percent growth in fiscal 2028.
The explanation for that acceleration lies in the physics of AI computation. Advanced AI models require enormous volumes of data to move across thousands of processors at very low latency. As model complexity increases, the networking layer inside a data center—not just the compute chips—becomes a constraint. Marvell's answer to that constraint is its 800-gigabit connectivity products and 1.6-terabit solutions. Its interconnect business is now expected to grow more than 70 percent year-over-year in fiscal 2027.
Marvell has also deepened its relationship with Nvidia. The two companies are collaborating on silicon photonics, NVLink Fusion integration and AI radio access network—known as AI-RAN—applications. That partnership matters because Nvidia's architecture choices dictate which networking and connectivity suppliers win alongside its GPU shipments. Being designed into Nvidia's stack is a durable revenue source, not a one-quarter benefit.
Analysts now project earnings-per-share growth of 42.13 percent for fiscal 2027. The figure for the following fiscal year was not available in full at the time of publication.
Broadcom is the other name Wall Street is treating as a structural AI infrastructure winner. Like Marvell, Broadcom earns revenue from custom silicon and high-speed networking products that sit inside the same hyperscale data centers driving Marvell's growth. The two companies are not competing for identical design wins—Marvell is stronger in optical interconnects while Broadcom holds dominant positions in custom AI accelerator chips and Ethernet switching at the highest speeds—but both are benefiting from the same dynamic: AI capital expenditure is flowing into the networking and connectivity layer, not only into GPU clusters.
The selectivity now visible in AI equity positioning reflects a maturation in how institutional investors think about the theme. In 2023 and early 2024, most AI-adjacent stocks moved together on sentiment. Now the divergence inside the sector is wide enough that owning the index is not a substitute for picking the right names. Companies that can demonstrate actual revenue growth, expanding margins and visible forward bookings are drawing capital away from names that are still in the promise phase.
Marvell at 131 percent year-to-date is the proof case. Nvidia trades at $225.16, essentially flat on the day, reflecting a market that has already priced in substantial AI infrastructure dominance at that level. The opportunity being pursued by investors rotating into Marvell is the argument that networking and custom silicon suppliers—one layer removed from the GPU itself—carry more upside because they are less fully valued against their forward growth rates. Whether that argument holds depends on whether the fiscal 2027 and 2028 data center projections prove accurate, and the most recent guidance revision moved in one direction only: higher.
