Nvidia invested $2 billion in Marvell to deepen their NVLink Fusion partnership. The deal aims to integrate Marvell's custom silicon into Nvidia's AI interconnect fabric for next-generation data centers.
Key facts
- Nvidia invested $2 billion in Marvell.
- Deal targets NVLink Fusion partnership.
- Marvell's custom ASIC business grew 30% YoY.
- Nvidia acquired Mellanox for $6.9B in 2020.
Nvidia invested $2 billion in Marvell to deepen their NVLink Fusion partnership, according to [MSN]. The investment aims to integrate Marvell's custom silicon into Nvidia's AI interconnect fabric, which pools GPU memory across clusters for large-scale model training and inference.
NVLink Fusion is Nvidia's protocol for connecting GPUs in a unified memory space, critical for training models exceeding 1 trillion parameters. Marvell brings expertise in custom ASICs and high-speed data infrastructure, having previously supplied chips for networking and storage in hyperscale data centers.
The deal signals Nvidia's strategy to lock in chip design partners as AI infrastructure scales. Competitors like AMD and Intel are pushing their own interconnect standards (Infinity Fabric, UPI), making ecosystem lock-in a priority for Nvidia. The $2 billion figure represents a significant bet on Marvell's ability to deliver custom silicon that differentiates Nvidia's platform from alternatives.
The companies did not disclose whether the investment includes an equity stake or specific product milestones. Marvell's stock rose 4.2% on the news, while Nvidia shares remained flat.
Why This Matters
This is the first major investment by Nvidia in a chip design partner since its $6.9 billion acquisition of Mellanox in 2020. Where Mellanox gave Nvidia control over networking (InfiniBand), Marvell brings custom silicon design — a capability Nvidia has historically built in-house. The deal suggests Nvidia is hedging against internal design bottlenecks as demand for custom AI chips surges.
Marvell's custom ASIC business has grown 30% year-over-year, per its Q1 2026 earnings, driven by AI accelerator contracts for cloud providers. By integrating Marvell's designs into NVLink Fusion, Nvidia can offer tailored interconnect solutions for hyperscalers like Google Cloud and Amazon Web Services, which increasingly demand custom silicon.
What to watch
Watch for Marvell's Q2 2026 earnings call on August 20 for updates on the partnership's revenue contribution. Also monitor Nvidia's GTC 2027 keynote in March for any NVLink Fusion product announcements involving Marvell silicon.
