Overview
- Nvidia, which announced the deal Tuesday, is taking a $2 billion equity stake in Marvell and forming a partnership to fold Marvell’s custom AI chips and networking into Nvidia’s rack‑scale platform.
- The collaboration centers on NVLink Fusion, a data‑center fabric that links many chips at very high bandwidth so large AI clusters act like one system, letting Marvell’s accelerators talk directly to Nvidia CPUs, GPUs, switches, and network cards.
- Both companies will co‑develop silicon photonics for optical links and work on AI‑RAN, which applies AI hardware and software to 5G and 6G radio networks to cut latency and power in telecom gear.
- Marvell shares jumped roughly 7% to 13% after the news while Nvidia rose, reflecting investor bets that solving data‑center interconnect bottlenecks will drive sales of chips, switches, and optical modules.
- Analysts note NVLink Fusion requires at least one Nvidia component and uses Nvidia‑controlled IP licenses, a semi‑closed approach that counters open efforts like UALink; Marvell’s recent Celestial AI deal adds photonic fabric tech that could speed adoption in hyperscaler racks.