Marvell’s Custom Silicon Bet Pays Off
Marvell Technology raised its fiscal 2028 revenue target to $20 billion, sending shares up 5.81% as investors digested a far more ambitious long-term roadmap. The chipmaker also lifted its fiscal 2029 custom silicon revenue outlook and projected fiscal 2031 revenue of $70 billion to $90 billion — a scale that would place it among the largest semiconductor suppliers tied to the artificial intelligence buildout.
The upgrade is anchored in demand for custom AI accelerators, the application-specific chips that hyperscalers commission to reduce reliance on merchant GPUs. Google’s in-house AI silicon program has emerged as a central driver, with Marvell positioned as a key design and manufacturing partner for the custom compute and networking components that surround these systems.
Why Custom ASICs Matter
Hyperscale operators are increasingly pursuing dual-track silicon strategies: buying Nvidia GPUs where they are available while designing proprietary accelerators optimized for their own workloads. Custom ASICs offer better performance-per-watt and lower cost at scale for specific inference and training tasks, but they require deep co-design between the cloud provider and a semiconductor partner.
- Margin profile: Custom silicon carries lower gross margins than merchant chips, but volumes are contracted and predictable.
- Moat: Multi-year design cycles create switching costs that lock in revenue once a program ramps.
- Risk: Concentration in a handful of hyperscaler customers exposes Marvell to program cancellations or in-sourcing.
The raised targets suggest management sees visibility extending well beyond the current cycle. Data center interconnect, optical DSPs, and custom compute now form the core of the growth story, with legacy storage and carrier businesses playing a diminishing role.
Read-Through for the Broader Market
The announcement reinforces a theme that has driven semiconductor equities for several quarters: the AI capex cycle is broadening from GPU vendors into networking, memory, and custom logic. Companies with hyperscaler design wins are being repriced on multi-year revenue visibility rather than quarterly beats.
For crypto markets, the implications are indirect but real. The same custom silicon supply chain underpins the data centers that host AI training clusters, and by extension the decentralized compute networks that compete for GPU-adjacent workloads. As hyperscalers absorb leading-edge capacity, decentralized GPU marketplaces gain a relative cost advantage for smaller inference jobs — a dynamic worth watching as tokenized compute narratives mature.
Forward Look
Investors will scrutinize whether the $20 billion fiscal 2028 target is underpinned by signed programs or pipeline optimism. The fiscal 2031 range of $70 billion to $90 billion implies a compound growth rate that assumes at least one additional hyperscaler ramps meaningfully. Any signal of a second major customer — or of Google expanding its silicon roadmap — would be the next catalyst. Conversely, evidence of in-sourcing by Google or a rival winning a flagship socket would pressure the thesis.
