Marvell (MRVL) Earnings Tonight: Can AI ASICs + Optics Prove It’s the Next Broadcom… or Does $79 Snap and Send It to the Low $70s?

Marvell Technology (MRVL) reports fiscal Q4 results after the close today with the stock sitting at a technical fork in the road: shares have drifted down into a test of the 200-day moving average (~$79) and have been using that area as a “please don’t make me think” support level heading into the print. The setup matters because the narrative around Marvell has become binary. Bulls see a company in the middle of a credible transformation into a core AI infrastructure platform—custom ASIC compute plus optical interconnect plus switching and scale-up networking—while bears argue the growth story is still too dependent on a small handful of hyperscaler programs that can change direction, insource, or reshuffle suppliers with little warning. Adding to the pressure, Broadcom’s (AVGO) strong AI-driven report last night raised the bar for anything “AI infrastructure adjacent,” meaning MRVL likely needs to deliver both solid numbers and clean, confidence-building commentary to avoid a post-earnings air pocket toward the low $70s.

Consensus expectations are constructive but not euphoric, which is part of why some analysts like the risk/reward into the event. Wall Street is looking for roughly $0.79 in adjusted EPS on about $2.2B in revenue for the January quarter, implying ~20%+ revenue growth year over year, and for the April quarter the Street is looking for roughly $2.3B in revenue and ~$0.74 in adjusted EPS. Options pricing suggests the market is bracing for an ~11% move in either direction post-print, larger than MRVL’s recent average swing, which tells you investors think tonight’s call can actually settle (or reignite) the “dominant platform vs. customer concentration” debate. The rough math is simple: a beat-and-raise with strong data center momentum probably keeps the stock defended above the 200-day; an in-line print with muddy guidance or program commentary risks a technical breakdown and fast money heading for the exits.

Analyst tone into the report leans positive, but the subtext is “prove it.” Evercore reiterates Outperform and says it would be a buyer into earnings, arguing MRVL offers multiple ways to win, including a potential beat and raise, improving data center growth assumptions as hyperscaler capex expectations climb, and what it frames as an unusually cheap valuation for an AI semi name (it cites ~22x NTM P/E). A key point in Evercore’s setup is positioning: buy-side bogeys for AI revenue in 2026/2027 are described as below sell-side estimates, and MRVL is seen as one of the names most vulnerable to downward estimate revisions—so if MRVL executes, the “consensus is too high” fear can unwind quickly in the other direction. The message is basically: sentiment is already cautious, so you don’t need perfection—just clarity and follow-through.

Other desks are focused on the plumbing that makes the AI buildout work. J.P. Morgan’s Harlan Sur has called out continued momentum in Marvell’s lead custom program with Amazon (Trainium) and strong demand for optical DSPs that enable high-bandwidth, low-latency transmission in AI data centers. That framing is important because it emphasizes that MRVL isn’t only “another compute story”—it’s trying to be a picks-and-shovels supplier across the connectivity stack, which tends to be more durable across chip cycles if the product roadmap is real and attach rates rise. On the bear case side, Susquehanna has flagged longer-term sustainability of the custom XPU business as “debatable,” with particular attention on whether share can shift within hyperscaler programs and whether customer-owned tooling (COT) increases bargaining power over time. Translation: the risk isn’t that AI goes away, it’s that the customer decides you’re optional.

So what matters tonight? Start with data center revenue and mix. The Street expects data center to be the majority of MRVL revenue (often framed as close to two-thirds recently), and the company has previously talked about data center growth trajectories (with >25% growth for FY27 referenced in the notes). Investors will want to see that the data center line is not only growing, but growing for the “right” reasons—optics and interconnect demand broadening, custom ramp visibility improving, and networking content increasing as clusters scale. If management commentary indicates optics is outgrowing custom in 2026, that can be a positive because optics demand is more “systemic” to scaling AI clusters, but the market will still demand proof that custom isn’t lumpy or losing relevance.

Second, custom ASICs: the call will likely revolve around customer concentration and program health. The most obvious questions are AWS continuity (Trainium 2 ramp, Trainium 3 timing/volume, and any language around commitments into mid-2026), whether there are “air pockets” in 2026, and how the Microsoft Maia ramp shapes 2H26/2027. Several analysts suggest no air pockets are expected and that custom accelerates in 2027 led by Maia, with further acceleration as projects won convert into revenue later. Investors will listen for specifics: design win count, expected ramps by half-year, what’s baked into guidance, and any commentary on how much of growth is already contracted versus still “pipeline.”

Third, optics and AI interconnect: watch the optical DSP narrative and the upgrade cycle to higher bandwidth (800G/1.6T) because that’s where MRVL can look less like a “single-program supplier” and more like a platform embedded across multiple hyperscalers and architectures. If the company can credibly talk about broad demand drivers (scale-out networking, optical DSP attach, storage interconnect) and not just one customer’s compute roadmap, that’s how it earns the “dominant AI infrastructure” label.

Fourth, the inorganic story: Celestial AI and XConn are supposed to deepen the scale-up/scale-out networking portfolio and expand the TAM, but investors will want near-term realism. Key items: integration progress, what is and isn’t included in near-term guidance, when meaningful revenue contribution begins, and whether these deals change customer conversations today (even if revenue is later). You don’t need a revenue miracle this quarter, but you do need a coherent bridge between “we bought cool stuff” and “customers are pulling it into roadmaps.”

Finally, valuation and what it implies for reaction function. Evercore is explicitly arguing MRVL is among the cheapest AI semis on a near-term multiple, and other commentary frames the stock as discounted versus its own history. That helps explain why a clean report could catalyze a re-rate, especially if MRVL can show it participates in the same multi-year AI spending visibility that AVGO just leaned into. But the flip side is also true: if results are merely “fine” and the story stays concentrated, the market will keep the multiple capped and the chart can do the talking.

Putting it together: for MRVL to hold ~$79 and avoid a trip to the low $70s, it likely needs (1) in-line to better prints, (2) guidance that supports the “no air pockets” framing, and (3) conference call commentary that broadens the narrative from “a couple hyperscaler programs” to “multiple durable AI infrastructure legs” (custom + optics + networking) with improving visibility. If management can do that, the 200-day looks like a staging area rather than a trap door; if not, gravity will do what gravity does.