Marvell targets up to $90 billion in revenue as custom AI chips boom

What’s going on here?

Marvell Technology held its Investor Day in New York on Tuesday and laid out far more ambitious growth targets than analysts expected. The company said it expects revenue of $20 billion in fiscal 2028, compared with a FactSet consensus of about $18.2 billion, and $70 billion to $90 billion in fiscal 2031, against expectations of roughly $47 billion. Marvell's shares rose about 6% in morning trading to around $288, and Broadcom, its main rival in custom chips, gained about 4%.

What does this mean?

Marvell's pitch centers on custom silicon: chips designed for one customer's specific workloads rather than general-purpose processors sold to everyone. Chief executive Matt Murphy argued that the giant cloud companies increasingly prefer owning chips tailored to their own AI systems instead of renting standard computing power. Marvell supplies much of what turns those designs into working products, including chip design, high-speed connections between processors, memory interfaces, advanced packaging and access to cutting-edge factories. All four of the largest cloud operators work with the company.

The move in Broadcom suggests investors see demand for custom chips widening across the industry rather than shifting to one winner. Both companies compete with Nvidia, whose off-the-shelf AI accelerators remain the default choice, so every custom program has to beat Nvidia on cost or performance to win. Marvell has tied part of its relationship with Google to actual orders: in August it granted Google warrants that unlock as Google buys chips.

The main question now is timing. Marvell earns money only when design wins turn into large production orders, which can take years and depend heavily on a handful of customers. After a roughly 239% rise this year, the stock already reflects a lot of future success, which leaves little room for delays.

Custom chips appeal to cloud companies for simple economic reasons. A chip built only for the calculations a company actually runs can use less power and cost less per task than a general-purpose processor, and owning the design reduces dependence on a single supplier. The trade-off is time and upfront cost: designing a chip takes years, which is why companies like Marvell and Broadcom, which can deliver complete designs, have become so valuable to the largest buyers of AI hardware.

The bull case

If cloud giants keep shifting AI spending to custom chips, Marvell could grow several times larger by the end of the decade, and its targets imply a business far bigger than today's. Design wins with all four major cloud providers spread the opportunity across several customers.

The bear case

Long-range forecasts are uncertain, and revenue depends on a few very large customers whose plans can change. Nvidia's improving products, competition from Broadcom and any slowdown in AI spending could delay or shrink the programs Marvell is counting on, and the stock's big run leaves it vulnerable to disappointment.