Marvell FY27: A $5 Billion Guide Raise Mattered More Than Jensen Huang
On June 2, Jensen Huang turned to Matt Murphy on a Computex stage in Taipei and called Marvell the next trillion-dollar company. The stock rose 32.52% that day, its largest single-session gain on record, adding roughly fifty billion dollars in market value before the close. Every desk on the Street ran the clip. Almost none of them ran the number underneath it.
The number was disclosed six days earlier, on the May 27 fiscal first-quarter call, and it carried no theater at all. Marvell raised its forward guide by roughly five billion dollars and lifted interconnect growth from 50% to over 70% year over year. The trillion-dollar line moved the tape. The guide raise moved the thesis. Those are not the same event, and conflating them is how investors end up paying for sentiment while telling themselves they bought fundamentals.
The Pop Was Theater. The Raise Was the Tell.
A CEO praising a partner from a stage is a sentiment event. It can be unwound by the next bad print, the next macro shock, the next rotation out of the AI cohort. A guide raise is a structural event. It is management putting a number in front of the Street and accepting the cost of missing it. One is reversible by mood. The other is only reversible by failure to execute.
The market treated June 2 as the catalyst because June 2 had a face and a quote. The actual catalyst had a spreadsheet. Marvell now guides to roughly $11.5 billion in revenue in FY27, a 40% advance, and roughly $16.5 billion in FY28, a 45% advance on top of it. Revenue does not compound at that slope by accident, and management does not raise into that slope unless the design wins behind it are already booked. The pop was the echo. The raise was the signal.
The Dual Engine
Most names levered to the AI build-out own one tailwind. They sell compute, or they sell connectivity, or they sell power, and the bull case rests on a single curve bending the right way. Marvell sits on two.
The first is custom silicon. Marvell designs application-specific accelerators for the hyperscalers that no longer want to rent Nvidia’s economics wholesale: Trainium for Amazon, Maia for Microsoft, a data-processing unit for Meta, Axion-class work for Google. In FY26 the data center segment reached a record $6.1 billion, with custom silicon alone scaling to a $1.5 billion annual run rate across eighteen cloud-provider design wins. These are not pipeline. These are awarded programs with multi-year tails.
The second is optical interconnect. When a computing problem is disaggregated across an entire data center, the constraint stops being the chip and becomes the link between chips. Marvell sells the digital signal processors, coherent optics, and silicon photonics that carry that traffic at 800G and 1.6T, and it sells them into a customer base where switching costs are punishing. Microsoft’s North American data centers source their optical chips from Marvell at the exclusion of everyone else. That is not a vendor relationship. That is infrastructure dependency.
One engine would justify a growth multiple. Two engines, compounding against the same secular demand curve, is the entire reason the FY28 guide is shaped the way it is.
The Guide Math
The headline figures are clean: $11.5 billion in FY27, $16.5 billion in FY28. The composition is where the conviction lives.
Interconnect is the segment management moved first, raising data center interconnect growth from a prior 50% to roughly 70% this year. Switching, a smaller line today, is guided to $600 million in FY27 and $1 billion in FY28, a near-doubling that reflects the scale-up networking attach as racks get larger. And the load-bearing assumption sits in custom silicon, where revenue growth is guided to accelerate from 20% in FY27 to more than 100% in FY28.
That last figure is the one to underline. The FY27 numbers are largely de-risked; the design wins are awarded and the optical attach is contracted. The FY28 guide depends on custom silicon roughly doubling, which depends on specific hyperscaler programs ramping to volume on schedule. If those programs slip, the $16.5 billion does not print, and the multiple the stock currently carries does not survive the miss. Everything bullish about FY28 routes through that single acceleration.
The Conflict the Tape Ignored
There is a detail in the June 2 story that most of the coverage walked past. On March 31, Nvidia took a $2 billion equity stake in Marvell as part of an NVLink Fusion partnership. So when Huang stood on stage and anointed Marvell a future trillion-dollar company, he was, in the plainest terms, an executive talking up the value of his own firm’s investment.
That conflict does not make him wrong. Nvidia has better visibility into hyperscaler interconnect demand than any analyst alive, and a $2 billion check is a more credible signal than a price target. But every investor who bought the pop inherited the conflict whether they priced it or not. The blessing and the position are the same fact viewed from two angles.
It matters more because of where the valuation already sits. Marvell trades near 72 times forward earnings and north of 55 times EV/EBITDA. At those multiples the stock is not pricing the next four quarters. It is pricing the FY28 guide as if it has already cleared, with the Nvidia endorsement underwriting the certainty. Remove the endorsement and the same numbers look a great deal more demanding.
Stock Trajectory
The Street repriced fast and hard. KeyBanc set the high at $385 on June 18, lifting from $260; Bank of America moved to $365 from $240; Stifel went to $350; B. Riley’s Craig Ellis took his target to $345 from $240; HSBC upgraded to $300 in late May. The bull case is uniform: Marvell re-rates toward Broadcom-tier status as the dual engine compounds, and the FY28 guide validates a multiple the market currently treats as aspirational.
The base case is quieter and more revealing. Consensus average price targets cluster in the low $240s, which sits below a tape trading near $271. That inversion is the tell. The average analyst is not ahead of the stock chasing it higher; the average analyst is behind it, modeling a name that has already run roughly 220% year to date and is pricing a multi-year ramp that has not yet been demonstrated. When the consensus target trails the price, the marginal buyer is paying for a thesis the median estimate does not yet endorse.
The bear case is not a Marvell-specific story. It is a cohort story. Every AI infrastructure name from the merchant silicon vendors to the optical players carries a derating multiple that exists only because demand has not yet disappointed. The moment the build-out cohort compresses — a hyperscaler capex pause, a macro shock, a single quarter where the orders soften — a stock at 55 times EBITDA falls faster than its fundamentals deteriorate, because the multiple was never about the fundamentals. Goldman Sachs still carries a $90 floor from December. That number looks absurd against the current tape, and it is exactly the kind of number that stops looking absurd the week the cohort breaks.
The decision-relevant catalyst is dated. Marvell reports fiscal second quarter on August 20. The interconnect raise will either be tracking to 70% or it will not, and the early read on the custom silicon ramp will either support the FY28 doubling or call it into question. Own the name into that print only if the custom silicon acceleration is the bet you actually want to make, because that single line is the one carrying the entire valuation. Jensen Huang gave Marvell a sentence. August will tell you whether the company earned it.