Marvell Technology gave Wall Street the number it wanted on May 27, and the stock rewarded it. Revenue rose 28% from a year earlier to $2.42 billion, edging past the $2.40 billion analysts expected, with adjusted earnings of $0.80 a share just ahead of the $0.79 consensus. Shares climbed more than 6% in extended trading. The beat was respectable. The guidance and the strategy underneath it are the reason the stock moved.
For a company that spent years as the quiet plumbing of the data center, Marvell has turned into one of the cleaner ways to bet on the custom side of the AI buildout, the part Nvidia does not own outright. That distinction is doing a lot of work right now.
The Custom-Silicon Story Is the Whole Story
Strip out the optical and networking businesses and the headline is custom AI chips. Marvell’s custom silicon, the XPUs that hyperscalers design with Marvell to run their own AI workloads, reached roughly $1.5 billion in fiscal 2026 and the company is guiding for north of 20% growth this year. Custom has scaled from effectively zero to about a quarter of data-center revenue in a few years, which is the kind of mix shift that re-rates a stock. That re-rating has already shown up in the share price, which has roughly doubled in 2026 as investors warmed to the custom AI-chip thesis.
The logic is straightforward and it is why the category is hot. The biggest cloud operators do not want to pay Nvidia’s margins on every chip forever, so they design their own accelerators and hire a partner like Marvell or Broadcom to turn the design into working silicon. Every XPU a hyperscaler ships is a GPU it did not have to buy. That makes Marvell a hedge against the exact concentration risk that has investors nervous about the rest of the AI trade, a theme that runs through the broader push toward custom AI silicon and in-house ASICs now reshaping the chip market.
Nvidia Wrote a Check, and That Tells You Something
Here is the wrinkle that makes the Marvell story more interesting than a clean earnings beat. Nvidia has invested $2 billion in Marvell, and the two companies struck a partnership under which Marvell supplies custom XPUs and NVLink Fusion compatible scale-up networking. Read that twice. The company whose pricing power custom silicon is supposed to erode is now a shareholder in one of the firms enabling it.
This is the circular dynamic running through the entire AI hardware economy, where the dominant chipmaker takes equity stakes in suppliers, customers, and would-be rivals, then books revenue from the same ecosystem it is funding. For Nvidia, the Marvell stake is a way to stay close to the custom-silicon wave rather than being disrupted by it, and to keep NVLink as the connective standard even inside chips Nvidia did not design. For Marvell, a marquee investor and a networking-standard tie-in is validation and a moat. The question nobody on the call wants to dwell on is what happens to all these interlocking bets if AI capital spending finally cools.
What the Guidance Actually Signals
The forward number is what moved the stock. Marvell guided second-quarter revenue to about $2.70 billion, plus or minus 5%, comfortably above the roughly $2.60 billion analysts had modeled. Reuters reported that the company forecast quarterly revenue above estimates on AI chip demand, sending shares higher, and that is the cleanest possible read on the print: demand is still accelerating, not plateauing.
That matters beyond Marvell. Custom-silicon order strength is a real-time gauge of whether hyperscaler AI spending is holding up, and a confident guide here is a quiet positive for the whole infrastructure complex heading into a stretch of bellwether earnings. When the company building other people’s AI chips raises its outlook, it is hard to argue the buildout is slowing.
The Bottom Line
Marvell is executing the rarest thing in this market: a credible second act that does not depend on beating Nvidia head to head. It sells the picks and shovels to the companies trying to escape Nvidia, while counting Nvidia as an investor. That is either brilliant positioning or a sign of how tangled the AI supply chain has become, and the honest answer is probably both. For now, the numbers are pointing up and the guidance is backing them. The risk is the same one hanging over every name in this trade: the day the capital spending slows, the custom-silicon growth that looks structural today will be tested in a hurry.