Marvell's newest guidance reset says the AI buildout is still broadening, not cooling. The harder question is whether that strength can make investors treat it more like a core AI infrastructure winner.
Marvell Technology has raised its revenue outlook again, and this time the signal is even harder to dismiss. The company is no longer just benefiting from a general data-center upgrade cycle. It is telling investors that demand for custom AI silicon, optical links, and Ethernet networking gear is strong enough to lift expectations for the next two fiscal years.
According to Marvell's May 27 results, first-quarter fiscal 2027 revenue reached a record $2.418 billion, up 28% from a year earlier and slightly above the midpoint of its prior guidance. The company also guided second-quarter revenue to $2.7 billion at the midpoint, which would represent 35% year-over-year growth. Management said AI-related bookings were strong enough to raise its outlook for both fiscal 2027 and fiscal 2028 from the targets it gave only one quarter ago.
That is the part investors should pay attention to. In March, Reuters reported that Marvell had lifted its fiscal 2027 revenue forecast to more than 30% growth, nearing $11 billion, while pointing to fiscal 2028 revenue of roughly $15 billion. The latest call moved those numbers higher, with management now discussing nearly $11.5 billion in fiscal 2027 revenue and about $16.5 billion in fiscal 2028 revenue. For a company that generated $8.2 billion in fiscal 2026 revenue, that is a meaningful reset.
The market has been trying to answer a simple question for months: is this still mainly Nvidia's story, or is it a broader infrastructure trade? Marvell's latest update points to the second answer. Its chips are not the consumer-facing products that dominate the headlines, and they are not the flagship GPUs that have become shorthand for the AI boom. They are the pieces that let larger AI systems work: custom application-specific integrated circuits, optical connectivity, Ethernet switching, and data movement technology.
That distinction matters because the AI buildout is becoming less about single chips and more about entire systems. Hyperscalers can buy more accelerators, but those accelerators lose value if the surrounding network cannot move data quickly enough. Training and inference clusters need memory access, low-latency links, switches, interconnects, and increasingly specialized silicon built around each customer's architecture. Marvell is trying to own more of that layer.
The company has been pushing that message hard. On its own AI materials, Marvell frames the current cycle as one of the largest infrastructure buildouts in history and positions itself as a supplier of custom silicon alongside interconnect and network-switch products. That framing matters because it separates Marvell from vendors selling more standardized parts. Custom silicon can take longer to develop, but once a hyperscaler commits to a design, the relationship can become more strategic and harder to replace.
Recent deal activity makes that strategy easier to understand. Marvell completed the acquisitions of Celestial AI and XConn in February, adding more technology around photonic fabrics and high-speed interconnects. It also became part of Nvidia's broader AI infrastructure ecosystem through an investment and NVLink Fusion partnership reported in April. That does not turn Marvell into Nvidia, but it does put the company closer to the center of the AI factory conversation than it was a year ago.
Competition remains the obvious constraint. Broadcom is still the better-known name in custom AI silicon, with deeper scale and a clearer place in several hyperscaler roadmaps. Nvidia remains the market's default AI winner because it controls the GPU platform, software stack, and much of the pricing power around accelerated computing. Marvell has to prove that its design wins can move from promise to revenue without sacrificing the margin profile investors expect from a premium chip supplier.
The valuation debate
That is where the stock story gets more complicated. Reuters reported on April 20 that Marvell shares rose after reports that Google was in talks with the company to develop two new AI chips, including one tied to Google's TPU ecosystem. Reuters also noted at the time that Marvell traded at 33.35 times forward earnings, compared with 27.84 times for Broadcom, while the median analyst price target stood at $125. Those numbers showed how quickly sentiment had shifted from discount story to premium AI exposure.
A richer multiple makes the next phase less forgiving. When a stock is priced as an AI infrastructure winner, meeting guidance is not always enough. Investors want evidence that the backlog is converting, that customer concentration is manageable, and that the company's newer optical and custom XPU opportunities are large enough to support the 2028 target. Marvell's updated outlook gives that case more weight, but it also raises the bar for execution.
For readers watching the AI trade from the outside, the message is straightforward. The spending wave is still broad-based, and the market is slowly admitting that the chip winners do not stop at one name. Marvell is one of the clearest reminders that the real AI infrastructure story lives deeper in the stack, where custom silicon and networking can matter as much as the flagship GPU. The next test is whether those design wins keep turning into visible revenue as cloud capex moves from buying raw compute to building more efficient systems.
Also read: Newcleo's Nasdaq plan shows nuclear is chasing AI money now • Google engineer arrested over Polymarket bets on search data • Powerlaw lists on Nasdaq as VC vehicles push private assets into public markets
