Why Hyperscalers Want a Second SourceNarrow moat

Most of Marvell's opportunity exists because its customers have decided it should. A hyperscaler committing billions to custom accelerators is creating a dependency, and the single most reliable way to keep a supplier honest on price and schedule is to have a credible alternative in the room. Broadcom's roughly 70% share is precisely the condition that makes buyers want to fund a second design house.

This is a genuine and underrated source of demand, and it explains behaviour that otherwise looks irrational — customers awarding Marvell programs it did not win on pure merit, or splitting work between two houses at some cost to efficiency. Google's decision to engage Marvell for TPU-adjacent silicon while Broadcom designs the TPU itself is close to a textbook example¹.

The uncomfortable corollary is that demand which exists to create competition can be withdrawn once competition exists — and that a company whose role is to be the alternative has limited pricing power by definition. Watch Marvell's custom-silicon gross margins. A second source that earns genuinely good margins has real technical differentiation; one that earns thin margins is being used as a negotiating instrument.