Marvell’s most plausible non-linear path is not “one blockbuster chip,” but expanding its share of each AI deployment as architectures add more interconnect, more optics, and more fabric complexity. If it sustains
hyperscaler custom silicon exposure while adding PCIe/
CXL switching (XConn) and optical
scale-up roadmaps (Celestial AI), investors can underwrite a steadier AI-infrastructure revenue stream even with modest
multiple compression versus today’s AI-premium pricing.