Marvell: The Second Source That Leads One MarketNarrow moat
Broadcom (AVGO) — moat facet
Broadcom beats Marvell roughly three to one in custom silicon and loses optical DSPs by about the same margin — proof its advantages are specific, not general.
Marvell is the only company that competes with Broadcom across most of what it does, and the scoreboard is more interesting than the usual summary allows.
In custom AI accelerators, Broadcom leads clearly: roughly 70% of design share against Marvell's estimated 20-25% by one count, nearer 60/35 by another1. In data-center Ethernet switch silicon, Broadcom's Tomahawk and Trident families hold a near-monopoly position with hyperscalers, and Marvell's Teralynx contests it from well behind2. Broadcom is comfortably ahead on both fronts.
Then look at optical DSPs — the silicon that drives the transceivers wiring racks together — and the ranking inverts. Marvell holds roughly 60-70% of the high-speed PAM4 DSP market against Broadcom's approximately 30%, with the two of them together above 90%3. Broadcom is the challenger there, attacking with its own 3nm parts into a position Marvell won by being first at each speed grade.
That single reversal is worth more attention than it gets. It shows that Broadcom's advantages are not general — they are specific to markets where design-in cycles are long and switching costs compound. Where the contest is time-to-market at a new standard, the same playbook does not work.
Watch whether Broadcom's DSP share rises through the 1.6-terabit transition. Taking optics would make Broadcom close to unavoidable in an AI data center; failing to would confirm that its moat has an edge, and where it lies.
The scoreboard has barely moved: Broadcom leads custom silicon and switching, Marvell leads optical DSPs, and both are spending to attack the other's ground at the next technology step. The 1.6-terabit optical transition is where this could actually change, and it has not arrived yet.
The one major market where Broadcom is the challenger rather than the incumbent, and the two together hold above 90%. It shows Broadcom's advantages are specific to long design-in cycles rather than general. Watch whether Broadcom's share rises through the 1.6-terabit transition.
Source: Third-party PAM4 DSP market-share estimates ↗- Third-party estimateBroadcom holds roughly 70% of custom-accelerator design share against Marvell's estimated 20-25%; another estimate puts it nearer 60/35.Custom-silicon design-share estimates — Broadcom approximately 70% of custom AI accelerator design share against Marvell's estimated 20-25%; Broadcom serves Google's TPU, Meta's accelerator program and OpenAI's custom silicon — 2026 · publ. 2026 · source ↗
- Third-party estimateBroadcom's Tomahawk and Trident families hold a near-monopoly position in hyperscaler Ethernet switch silicon.Third-party AI-networking market analysis — NVIDIA's Ethernet switch share rose from under 4% to approximately 21.5% in a single year, won by bundling networking with the GPUs customers already required; Broadcom retains a near-monopoly position in hyperscaler Ethernet switch silicon with its Tomahawk and Trident families — 2026 · publ. 2026 · source ↗
- Third-party estimateMarvell holds roughly 60-70% of high-speed PAM4 DSPs against Broadcom's ~30%, the two together above 90%.Third-party analysis of the high-speed optical DSP market — Marvell and Broadcom together hold more than 90% of the PAM4 DSP market; within it Marvell holds roughly 60-70% against Broadcom's approximately 30%; Marvell was first to 200G-per-lane 1.6T silicon in 5nm (Nova, 2023) followed by the 3nm Ara platform, and Broadcom's Sian3 targets the same 3nm 200G-per-lane ground — 2026 · publ. 2026 · source ↗