CompetitorsNarrow moat

Marvell's competitive position cannot be summarised in a share number, because it holds two very different positions in two markets against largely the same opponent. In custom AI silicon it is the challenger, holding perhaps a fifth to a third of design share against Broadcom's clear lead. In optical DSPs the figures are close to reversed: Marvell is the leader, with roughly 60-70% of the market, and Broadcom is the one pushing in¹. Anyone who describes Marvell as 'the number two to Broadcom' has read only half the scoreboard.

The four pages here are deliberately four different kinds of relationship rather than four rivals. Broadcom is the mirror — the same company met twice, winning one fight and losing the other. Nvidia is the opponent that invested $2 billion in Marvell and opened its rack interconnect to it, while Marvell's largest customers buy Marvell silicon specifically to depend less on Nvidia². Alchip, GUC and MediaTek attack from underneath on cost with a different business model entirely, and they are the ones actually taking sockets. And the laser and module makers are the competitor Marvell has deliberately chosen not to become — a restraint that wins it business.

One rivalry is covered elsewhere and not repeated here: Credo, Astera Labs and the interconnect specialists nibbling at the low-power end of optics have their own page under the Electro-Optics moat aspect, as does Broadcom's grip on switching silicon.

The honest summary is that Marvell wins where being early matters and loses where scale matters. Optical transitions reward the company that has working silicon first; custom accelerator programs reward the company with the most programs already running. Watch which of those two dynamics is setting the growth rate in any given year — because the multiple assumes both are working at once.