What Marvell Actually ContributesThin moat

It is worth being precise about the division of labour, because it determines how much of the value Marvell captures. The hyperscaler supplies the architecture — the compute cores, the instruction set, the model of how the chip should work. Marvell supplies the pieces that turn that into silicon: high-speed serial interfaces, memory controllers, chip-to-chip interconnect, packaging expertise, and the physical design work of closing a billion-transistor die at a leading-edge node.

The IP is the durable part. World-class SerDes — the circuits that push data down a wire at hundreds of gigabits per second — takes many years and many silicon iterations to develop, and only a handful of companies have it. That IP is reusable across every program, which is why the marginal cost of the eighteenth design win is far below the first.

The fragile part is that architecture is where the value is increasingly seen to sit, and the customer owns it. Watch how much content Marvell captures per program over time. Supplying more of the chip is the difference between a design-services business and a semiconductor franchise, and the two deserve very different multiples. Marvell captures enough of it today for custom silicon to run near $1.5 billion a year¹.