On a stage in New York on Tuesday, the chief executive of Marvell Technology made a promise that, by the arithmetic of his own company, would require it to grow roughly tenfold in five years. Matt Murphy told investors the chipmaker now expects revenue of $70 billion to $90 billion in its fiscal 2031, with earnings above $30 a share. Marvell had never before put a number on that year.
The target landed far beyond what Wall Street had written down. Analysts had modeled about $47 billion in fiscal 2031 revenue, according to CNBC. Marvell’s trailing twelve-month revenue sits near $9.4 billion. The midpoint of the new range, $80 billion, implies a company between seven and ten times its present size by the end of the decade.
Shares jumped nearly 10% on the day. The rally was not a response to anything Marvell sold last quarter. It was a bet on the shape of the data-center business the company believes it is becoming.
Murphy also lifted the nearer target. The company raised its fiscal 2028 revenue outlook to about $20 billion from $18 billion, the fifth upward revision to that figure in roughly a year. It sketched gross margins of 56% to 59% and sized its 2030 addressable market at about $400 billion. Behind all of it sits a single assumption: that the money pouring into AI data centers keeps pouring for years.
At the $80 billion midpoint, Marvell said roughly $37.5 billion would come from interconnect, the networking gear that moves data within and between data centers. About $30 billion would come from custom chips, the processors that hyperscalers order to run their own AI workloads. Switching and storage would contribute about $10 billion, and communications and other business about $2.4 billion.
The interconnect projection is as telling as the custom-chip one. As AI clusters grow, moving data between thousands of chips becomes a bottleneck as binding as the chips themselves, and Marvell’s networking gear sits at that junction. The company is betting that the more silicon the hyperscalers buy, the more of its own products they need to link it together.
The custom-chip line is where Marvell’s story has changed most. A decade ago the company was best known for storage controllers and networking silicon, the quiet parts of a data center that nobody saw. Today it is one of a small group of firms, alongside Broadcom, that design processors to a customer’s specification rather than selling a standard product off the shelf.
Hyperscalers turned to custom silicon because standard graphics chips are scarce and expensive, and because a chip tuned to one company’s software can run faster and cost less per query. Amazon, Google, Microsoft and Meta have all moved some of their AI workload onto chips they helped design, and Marvell has been among the biggest beneficiaries. Custom work has pulled the company from the periphery of the AI boom to a place near its center.
The company’s own history makes the forecast harder to dismiss. Marvell was founded in 1995 by three siblings who named it for the word “marvelous,” and it spent its early decades in the unglamorous trade of storage and networking controllers. Its move into custom silicon, built through years of quiet design wins, is what put it in a position to stand on a stage and describe a tenfold future. Competitors and customers now treat it as a serious rival to Broadcom in a business that barely existed a decade ago.
That shift has reordered the chip industry’s pecking order, and it is the reason Marvell can forecast a number that once would have sounded absurd. Management told investors it expects data-center capital spending to reach $3 trillion by 2030. Marvell’s forecast assumes it captures a growing share of that spending. It does not assume the spending is guaranteed, but it does assume the industry’s direction holds.
Wall Street, for the most part, went along. Analysts at TD Cowen upgraded the stock, and several firms raised price targets after the presentation. The earnings-per-share goal of more than $30 sits well above the consensus near $19. The gap between the company’s framework and the Street’s estimates is the whole story in miniature: Marvell is asking investors to believe the AI build-out is earlier in its arc than the skeptics think.
The risk is that a target is not an order book. A five-year revenue framework is a statement of intent built on assumptions about customer budgets, chip prices and competition, and every one of those can move. Marvell’s largest customers are a handful of hyperscalers that can switch suppliers, delay purchases or bring designs in-house, and the custom-chip business, for all its growth, depends on a small number of very large relationships.
Marvell did not say it holds contracts that guarantee the number. It said the market will be that big and that it intends to win a share of it. The distance between those two statements is where a stock can get ahead of the facts, and it is the question investors will be asking every quarter from now until 2031.
What Marvell is really selling is a view of the next five years: that AI will keep demanding more chips, that more of those chips will be custom, and that a company once known for storage controllers will be one of the few capable of designing them at scale. The forecast was easy to write down. Hitting it is the hard part, and Tuesday’s share-price jump shows how much of the market has already decided the company will.
