Upside comes from rising AI cluster bandwidth demands lifting connectivity silicon and custom programs. The key risk is customer concentration and timing of optical scale-up monetization.

Marvell can compound into 2030 by becoming the connective tissue of AI clusters (custom silicon + high-speed interconnect + emerging optical scale-up), with upside from turning lumpy hyperscaler co-design into multi-year platforms and adding higher-quality attach in security/telemetry.

AI pushes spend toward bandwidth, latency, power-per-bit, and verification; Marvell sells those bottleneck enablers, but lacks vertical control of fabs/energy and faces hyperscaler bargaining power.

The non-linear setup is that AI training clusters keep scaling faster than traditional enterprise IT, and the limiting reagent becomes interconnect (power, reach, signal integrity, and reliability). Marvell is positioned across multiple “must-buy” layers (switching, optical/electrical links, and custom chips) and can expand wallet share per deployment as architectures move from copper-heavy racks toward optics-heavy rack-scale systems. By 2030, a larger revenue base plus higher recurring attach (security lifecycle, fabric optimization, and ecosystem compliance) can support a durable premium versus average semis, though not a hyperscaler-style multiple given customer concentration and program cyclicality.

The two swing risks are (1) hyperscaler program concentration/dual-sourcing (custom AI and interconnect sockets can move quickly) and (2) the timeline and monetization of optical scale-up. Secondary risks: acquisition/earnout dilution optics, export-control shocks, and the possibility that security/services attach remains a niche rather than a material recurring stream.

$111.25