Can Marvell really grow revenue tenfold and earn 30 a share?

What grabbed me when I first read through Marvell's early-October investor day was not the 80 billion headline but the near-term arithmetic. The company printed roughly 8.2 billion of revenue in fiscal 2026 and now guides to about 12 billion for 2027 and 20 billion for 2028. That means roughly 47 percent growth followed by roughly 67 percent, with the pace quickening even as the base gets much larger. You rarely see a listed chip stock accelerate like that two years in a row, and re-ratings in the stock market usually start from exactly this kind of inflection. To my eye, there is a serious execution story here that deserves attention before anyone dreams about 2031.

Two details make the near term more believable. First, the 2028 goal has been lifted five times inside a year; under twelve months ago the discussion was about 13 billion, today it is 20 billion, ahead of the 18.2 billion analyst consensus (Benzinga tracked that revision streak meeting by meeting). Second, the fiscal second quarter of 2027 already delivered record quarterly sales of 2.739 billion, up 37 percent year on year, with non-GAAP gross margin at 58.9 percent (the figures come from Marvell's own release). So the bars in the investor presentation rest on a balance sheet that is already breaking records, not on wishful thinking. No wonder the shares jumped something like 6 to 10 percent on the day; rallies in this market tend to follow consensus-beating revisions.

The engine is unambiguously the data center franchise. Data center sales stood at 2.2 billion in 2024 and are now running toward 18 billion in 2028; an old 2029 objective has effectively been pulled forward by a year. Hitting targets is one thing, pulling them forward is another, and the stock market pays a confidence premium for the second. The key concept here is XPU attach: the hardware around the main processor that hauls, stores and manages data. Marvell's operating chief told SquawkNews in an interview that this line has expanded much faster than expected. Custom silicon sales, seen above 12 billion in 2029, belong to the same story and imply a near-tripling from 2028 to 2029. My reading: Marvell no longer bets on a single die but on the whole silicon collar around the processor.

Near term: an accelerating growth cadence

The most serious feature of the long-range model is that the 80 billion is fully itemized. In the midpoint scenario, interconnect contributes about 37.5 billion, custom silicon 30 billion, switching and storage 10 billion, and communications plus other lines 2.5 billion, summing to roughly 80 billion (the split as summarized by Yahoo). Note what the largest slice is: not one customer contract, not the custom chip, but interconnect across optical and copper links. That breadth matters because single-design stories can collapse on one delay. Here, if one of four engines sputters, the others can keep carrying. For a stock market that punishes concentration, that spread is worth something. All of them still ride the same AI infrastructure cycle, of course; but a spread portfolio shrinks single-bet risk.

The most debated piece is the strategic pact with Google. The company describes up to 120 billion of opportunity over six and a half years across custom programs that plug into the TPU ecosystem, and Reuters documented the accompanying warrant mechanics. The other side of the coin came from NextPlatform: announcement language for giant pacts and shipment rhythm on the ground are different things, with each tranche vesting on performance. I find that skepticism healthy. The good news is that the whole 2031 construct does not rest on this pact; interconnect and switching are independent engines outside Google. So the investor backs several ramps, not a single signature, which is firmer footing on which to justify a valuation multiple.

The market-size assumption is equally bold. Management sees total addressable spend reaching 400 billion in calendar 2030: 235 billion custom silicon, 85 billion switching and storage, 65 billion interconnect, 15 billion communications and the rest. FuturumGroup notes the company quadrupled its market estimate and plans to take the scale-up optics arm from near zero to a ramp starting next year. The assumed paces are fierce: about 55 percent a year in custom silicon, 65 percent in interconnect, 45 percent across the total. Picture a highway widening while its toll booths widen too; Marvell argues the silicon content inside every server will keep climbing. That is a share-gain thesis, and it needs proving in the numbers quarter after quarter. The stock will ultimately be judged on whether that content growth shows up in reported mix.

Anatomy of the 2031 model: where 80 billion comes from

The slickest frame in the deck was a three-rung ladder: data center capital investment grows about 35 percent a year, Marvell's serviceable market grows 45 percent, and the company itself grows 55 to 60 percent. The pie expands, Marvell's slice of the pie expands faster than the pie, and the company expands faster than its slice; the gaps are share gains and product mix. In stock market terms, the ladder is a promise of sustained outperformance. Management says these goals derive from live customer ties and design programs rather than a top-down market guess, a bottom-up build that reads plausibly on paper. But keep the base assumption in view: worldwide data center investment must climb from 711 billion in 2025 to 3.1 trillion in 2030. If that first rung wobbles, the others shake too. For me the 2031 goal is a conditional option, not a promise.

The margin side raises the natural question as custom weight rises: do deals cut with giant customers holding heavy bargaining power compress profitability? For 2031 the company targets 56 to 59 percent non-GAAP gross margin, 44 to 46 percent operating margin and free cash flow above 36 percent. The defense runs like this: on custom work the customer funds part of the engineering, lightening Marvell's own research bill, and swelling volumes switch on operating leverage. Operating costs are in fact planned to rise at roughly half the pace of sales. Merchant connectivity lines carry fatter gross margins but the research burden sits fully in-house. My math: if that equation holds, scale offsets mix pressure; if it breaks, the 30 earnings dream dies. For the stock, this margin bridge is the hinge of the whole bull case. Balance-sheet discipline matters here as much as any multiple.

The 35-45-60 ladder, margins and risk

The most quoted figure is the 30 in 30 slogan: more than 30 of adjusted earnings per share in calendar 2030. With the shares near 275 the scenario math is simple: a 15 multiple means about 450 (roughly 64 percent upside in dollars), 20 means 600, 25 means 750, 30 means 900. The multiple awarded will hinge on the growth rate still visible at that point; markets pay up for businesses still sprinting and haggle with mature ones. Jensen Huang's next trillion-dollar company line can be tested the same way: at the current share count 1 trillion needs about 1,140 a share, or a 38 multiple on 30 of earnings, falling to 29-33 times if earnings reach 35 to 40. So the trillion is not inside the model, yet no longer pure fantasy either. In my experience the stock market starts discounting that kind of long-dated option well before it matures, as confidence builds.

The Street's reaction was telling. After the day, Susquehanna lifted its objective from 265 to 340 and stayed constructive; Investing reported the shares around 287 with a 238 percent gain since January. Other houses wrote 350 to 400. My split is this: 2027-2028 delivery and the 2031 option are different assets. The first is measured in orders, shipments and margin discipline; the second depends on architecture shifts, from scale-up optics to the copper wall to memory disaggregation, breaking Marvell's way. I would track the near term quarter by quarter and price the long term like a discounted call option. On that balance Marvell works as a carriable growth holding in my portfolio; but I refuse to count 2031 dollars as banked today. The next layer of diligence should be architectural, because wiring choices rather than slideware will decide the outcome.