PAM4 and the Physics of Going FasterWide moat

The reason a signal processor is needed at all is that light travelling through fibre, and electricity travelling through copper, arrive distorted. PAM4 encoding squeezes two bits into each signal level to double throughput, at the cost of much tighter tolerances — and it falls to the DSP to reconstruct a clean signal from a degraded one.

This is a moat with an unusual property: it deepens with each generation. Moving from 400 gigabit to 800 to 1.6 terabit does not make the problem 2x or 4x harder; noise, power and manufacturing tolerance all worsen disproportionately, so the number of companies capable of shipping a working part at the newest speed shrinks rather than grows. Marvell's advantage is that it has been through this transition several times and has silicon-proven IP each time.

The vulnerability is that a hard problem is not an impossible one, and the rewards for solving it keep rising. Watch time-to-market at each new speed grade: being first with a working DSP at the newest rate is the entire advantage, because module makers design around whatever is available when they need it and rarely change mid-generation. Being early repeatedly is how the optics franchise helped carry data-center revenue past $6 billion in fiscal 2026¹.