Eighteen Wins and the PipelineNarrow moat
Marvell Technology (MRVL) — moat facet
Wins are announced and losses rarely are — so watch revenue per win, not the count.
Marvell describes its custom business in terms of design wins rather than revenue, because the revenue arrives years after the win: 18 cloud design wins, more than 50 active AI opportunities, across over 10 major customers1. That disclosure style is standard in the industry and it is also convenient, since a win is announced when it is secured and a loss is rarely announced at all.
Taken at face value the pipeline is genuinely broad. Ten customers is more than the handful of hyperscalers everyone names — it implies sovereign programs, second-tier clouds and large enterprises building their own inference silicon, which is exactly the market broadening that would make Marvell's position more durable than a two-customer business.
The caution is that wins are not equal. A full accelerator program is worth many multiples of a storage controller or a network interface chip attached to someone else's accelerator, and the disclosure does not distinguish. Watch custom-silicon revenue against the win count over time: if the number of wins keeps climbing while revenue per win falls, Marvell is winning the periphery of programs whose centre belongs to somebody else.
The count keeps rising and the customer base has broadened beyond the obvious hyperscalers to more than ten buyers. Breadth is the single best defence against the socket risk, so this genuinely widens — with the caveat that wins are disclosed and losses are not.
Design wins turning into production show up as a rising data-center share; a fall would mean the pipeline is converting slowly.
Source: Marvell Q2 fiscal 2027 results release (Exhibit 99.1, 27 August 2026) ↗- Third-party estimate18 cloud design wins, more than 50 active AI opportunities, across over 10 major customers.Marvell custom-silicon disclosures — roughly $1.5B annual run rate across 18 cloud-provider design wins, with more than 50 active AI opportunities across over 10 major customers; custom revenue guided to grow in FY2027 and potentially double in FY2028 on new hyperscaler programs including Microsoft's next-generation Maia — FY2026-FY2028 · publ. 2026 · source ↗